What we measured: the hypotheses everyone repeats
Google Ad Grants gives a nonprofit $10,000 a month for search ads. The money is free and renews every month. Even so, across the accounts we measured, the average account spends a few percent of it.
The bottom half of the table shows the scale best. 273 of the 706 accounts spend $935 a month between them — all of them together, not each. That is under $3.50 per account against the $10,000 granted. On paper the grant works for these organizations; in practice it brings them nothing.
Does that mean Google Ad Grants does not work? No. Below we measured exactly why these accounts stall.
For years we heard the same hypotheses from Google support and from Ad Grants specialists: thin content, slow pages, a Quality Score penalty, too small a daily budget. In this article we measured them one by one, to see which of them survive the numbers.
A single organization sees only its own account and has nothing to compare it with. We do: PING.NGO, our Ad Grants account audit and the only tool of its kind, audits more than 700 accounts in the program continuously. So the question can be asked across accounts instead of one at a time: what separates the ones that spend the grant from the ones that stall.
Leave your contact and we will send you the whole study
Everything below is what a single account cannot tell you: the method, the results and the one place where the grant actually leaks. Leave your contact and we will get back to you with a link to the full text.
- How we measured: what we counted inside accounts and what between them
- The three gates every keyword has to pass
- The hypotheses that failed — including the ones the industry repeats
- Diagnosis: how to tell where a given account stands
Not every keyword enters the auction
Before the numbers and the proof, the problem needs a name, because it is not what it looks like. Ad Grants accounts do not have too few keywords, nor badly chosen ones. They have keywords that never enter the auction — and a keyword that does not enter the auction produces no impressions and no clicks.
The difference between an account that spends the grant and one that stalls is not phrase selection. Both carry the same share of phrases with no demand and the same share rejected by Google. The difference is how many phrases start at all: on a stalled account, impressions come from 1.1% of its keywords; on an account that spends the grant, 24%.
So the question is not “which keywords to pick” but “what makes the keywords you picked start at all”. The rest of this article answers the second one.
What data we measured
We pulled the full spend history from every account — $41.8M, 26.4M clicks and 676M impressions.
The factor tests themselves run on the last twenty months, and that is deliberate. The Ad Grants rules changed along the way — Performance Max campaigns became available in 2024, among other things — so data from before those changes describes a different program than the one an organization works in today.
The Google Ad Grants rules
Ad Grants is not a normal Google Ads account with a free budget. It runs under its own rules, which narrow what you can do, and part of this study only makes sense once you know those limits. Four rules matter most.
| Restriction | What it means | What it does to account performance |
|---|---|---|
| $2 cost-per-click cap | With the maximize clicks bidding strategy a hard $2.00 cost-per-click ceiling applies. It cannot be raised. | An account with no conversions yet has to start on that strategy — so it starts under the ceiling. |
| Exemption from the cap | The ceiling does not apply on conversion-based strategies: maximize conversions, maximize conversion value, target CPA and target ROAS. | Switching to a conversion strategy lifts the ceiling — but conversion tracking has to work first. |
| No single-word keywords | Single-word keywords are not allowed, with exceptions: the organization’s own brand names, medical conditions and a few other categories on Google’s list. | The keyword list has to be built from multi-word phrases, which by itself pushes it towards narrower demand. |
| Performance Max | Available in the grant since 2024. | The newest and least understood change to the rules; we treat it separately in this study. |
The first two rules form a trap: to lift the bid ceiling you need conversions, and to get conversions you first have to win auctions, which the ceiling prevents. A large part of this study is about getting out of it.
Five numbers that change an account
The whole article comes down to this table. Each row is one measured factor, one number before and one after — the rest of the text is the evidence, the method and the caveats.
| What moves the result | Z | Na | What the number measures |
|---|---|---|---|
| Lifting the $2 cost-per-click cap a conversion strategy instead of maximize clicks |
5.1% | 21.9% | Share of keywords that enter the auction at all |
| The phrase written into the landing page title | 22% | 37% | Chance that a keyword ever gets an impression |
| Pages promoted: 1–2 versus more than 40 | $51 | $6,867 | Median monthly account spend. The bottleneck is the page, not the campaign |
| Phrases the site already ranks for organically | 4.0% | 31.3% | Share of phrases that get impressions. An eightfold difference — and 89% of those phrases are missing from the ad account |
| Turning on Performance Max | — | +$721 | Median change in monthly spend. PMax does not bid on specific phrases, so it reaches where a keyword list cannot |
Conclusions: what works and what does not
We started from a list of everything that — according to Google support, the industry, the documentation and our own experience — could affect account performance. Then we measured each item separately.
| Factor | Verdict | What the measurement showed | What to do about it |
|---|---|---|---|
| Bidding strategy (the $2 cap) | WORKS | Lifting the ceiling more than quadruples the share of keywords entering the auction: 5.1% to 21.9% | Switch to a conversion strategy — but first check that the account records conversions at all |
| Size of the list of phrases people actually search | WORKS | Every 100 phrases with real demand means, on average, +40 clicks and +650 impressions a month. What counts is the number of phrases, not the total demand behind them | Keep adding phrases above 10 searches a month. Count the target on phrases with demand only |
| The keyword in the page title | WORKS | A keyword’s chance of starting rises from 22% to 37%. Today 63% of pages in ad groups have no keyword in the title at all | Rewrite titles and H1 headings so they carry the phrase that points to that page |
| Whether the page is on the phrase’s topic — even without the words in the text | WORKS | A page that answers the phrase in different words also raises the odds — less than the phrase written in directly, but independently of it | First check whether the page answers the phrase at all. If it does, fix the title. If it does not, the page has to be written |
| Phrase competition | THE OPPOSITE OF WHAT YOU EXPECT | Medium-competition phrases start most often and bring six times more clicks than high-competition ones — but also more than low-competition ones. An “easy” phrase usually means “nobody finds it worth paying for” | Do not filter phrases by competition. Watch out only for a list built entirely from low-competition phrases |
| Keyword match type | WORKS | On broad match 5.9% of keywords produce impressions, on phrase match 0.7%, on exact match 0.1% | Do not narrow. Cut junk traffic with negative search terms, not with match types. One exception: keep brand ad groups on exact match — broader matching starts serving generic queries and eats the click-through buffer |
| Account structure: pages and ad group size | WORKS | Median spend: $51 with 1–2 pages promoted, $6,867 with more than forty. An ad group points to one page — above 50 phrases in a group only 4.1% of them start instead of 10% | Aim for 16–40 landing pages, 10–25 phrases per page, and never more than 50 phrases in an ad group |
| Campaign theme: local versus fundraising | WORKS | Local ad groups start twice as often as any other and win 64% of comparisons inside the same account. Fundraising groups lose to every theme — 2.8% of their keywords start. Service, informational and brand themes are indistinguishable | Add local ad groups — only 45% of organizations run them. Do not build the account on fundraising and do not judge the grant by it. Keep service groups for volume: they start less often, but deliver 16.5 impressions per keyword against 5.9 for local ones |
| Performance Max | WORKS | 4.2× the spend and 3.0× the clicks when accounts are compared. It reaches where a keyword list cannot | Turn it on once you have assets: copy and signals about who you are looking for. Without them PMax has nothing to build from |
| Dynamic ad groups (DSA) Google picks the queries from your site content | WORKS | Dynamic groups click worse than keyword-based ones. What decides is the proportion — while they make up to 45% of account impressions the loss is 1.1% CTR, above 70% it is already 4.4% CTR | Add them only when keyword-based groups hold at least 6% CTR and dynamic ones stay under 45% of account impressions. Turn them off when they pass 70% of impressions, or when account CTR drops below the required 5% |
| Amount of content on the page | WORKS | The share of keywords rated “weak page” falls from 73% at 2–4k characters to 42% at 13–16k; the whole jump sits between 4k and 6k. Length does nothing to account spend at all — nor to how many phrases in the account start | Get above 4k characters, aim for around 8k — and aim for relevance |
| Phrases the site already ranks for in organic results | WORKS | Those phrases produce impressions eight times more often than the average phrase on the list. And 89% of them are missing from the ad account entirely (median 19 phrases per account) | Add them to the account — the only phrase list in this study that needs no writing |
| Consent Mode behind the cookie banner | WORKS | Consent Mode delivers more conversions — on average about 25% of users refuse cookie consent — which is what lets you run maximize-conversions bidding and makes campaign analytics usable. A banner on its own, without Consent Mode, is no different from no banner | Check whether Consent Mode is wired in — in the page code or in Tag Manager. Conversions still count without it; Consent Mode recovers the part lost to refusals, so the bidding strategy has more data to learn from |
| Judging the account by search terms, not by the keyword list | WORKS | 71.5% of the list’s impressions and 65.4% of its clicks come from queries that appear on no account’s keyword list | Read the search terms both ways: cut what has nothing to do with the organization, and add the queries that fit it as phrases — they are the only phrases with a proven entry into the auction, before anyone put them in the account |
| Excluding search terms that never get clicked | WORKS | On a healthy account this is cosmetic. On an account under the required threshold it lifts CTR by 0.13% and costs 2% of clicks | Once a month cut the search terms with impressions and zero clicks |
| How long a new campaign takes to ramp up | WORKS | Half of campaigns reach full speed in three weeks, but some need fourteen. A new ad group and a new keyword take as long to ramp up as a whole campaign | For a fixed date (a gala, an intake) start 11 weeks ahead; for a flexible one, 7 |
| Google’s ad quality rating (Ad Strength) | DOES NOT WORK | With the same number of headlines and descriptions inside the same account, the rating does nothing to impressions or to click-through rate. In practice it counts filled-in fields | Fill it up to 15 headlines and stop worrying about it |
| Daily budget | DOES NOT WORK | Of all the lost impression share, only 1.2% falls on budget — the rest is auctions lost | Do not raise the daily budget expecting growth |
| Quality Score | DOES NOT WORK | Google does not report it for 97.5% of the keywords in the sample — only a handful of phrases that already have traffic get one | Do not make it a goal. It is a consequence of having traffic, not the road to it |
| Account age and domain age | DOES NOT WORK | Both disappear once you compare accounts of similar size. An old account has no advantage over a new one | Do not explain weak results with a young account or a young domain. Campaign ramp-up is a different thing and has to be allowed for — a new organization will not show results on day one. One exception works immediately even on a young account: phrases the site already ranks for in organic resultsdo not wait for account history |
| Page speed (PageSpeed) | DOES NOT WORK | No effect on the result | Fix it for the user. Do not budget for it as a way to improve campaign results |
| Number of campaigns | DOES NOT WORK | The same 400 phrases in three campaigns or in twelve give the same result. | Do not split an account “for tidiness”. Ad group size is a different thing and it does matter |
| Meta description, heading hierarchy, images, internal links | DOES NOT WORK | Of nine on-page SEO elements, three pass — and two of them come down to one thing: whether the page names the phrase’s topic | Drop the classic SEO checklist. Move the work to the title and to on-topic content |
| Domain authority (backlinks) | DOES NOT WORK | Accounts with a stronger link profile do better only because they are bigger. Compared at the same account size — no difference | Do not build links for campaigns |
| Which conversions the campaign optimizes for | DOES NOT WORK | 113 real configuration changes across the portfolio’s history; the effect on spend is zero in both directions | Set a real event anyway — for the report and for the algorithm, not for the spend |
The order of the work — and why it is not free choice
The list above says what works. It does not say where to start, and that is a separate question with a hard answer. Every earlier stage is what makes the next one worth doing. This is not a belief but a measurement: fixing page titles on an account under the $2 cap gives an effect indistinguishable from zero, while the same work after the cap is lifted gives five times more. Work done out of order is not less effective. It is ineffective.
| Stage | Co | Threshold | What happens if you skip it |
|---|---|---|---|
| 1 | Entry condition a conversion strategy instead of the $2 cap · uninterrupted conversion tracking | every search campaign on a conversion strategy — the $2 cap applies per campaign, so one left behind stays under the ceiling; no gaps in conversion tracking | 5.1% of keywords enter the auction instead of 21.9% |
| 2 | Pages number of pages · phrases per page · topics from the organic gap | 16–40 pages · 10–25 phrases per page · content above 4k characters | Adding phrases to ad groups that all point to one page changes nothing |
| 3 | Titles and H1 headings the ad group’s phrase written into the page title and the H1 | ≥50% of the account’s pages | a keyword’s chance of entering the auction is 22% instead of 37% |
| 4 | Phrases | 400 phrases with demand ≥10 searches/mo · broad match · up to 50 phrases per group | Without stages 2 and 3, adding phrases raises the count and not the result |
| 5 | Structure split oversized groups · retire dead ones · add local groups | no group above 50 phrases; at least one local group when the organization works locally | Two thirds of the list sits in groups that never served once |
| 6 | Performance Max | No numeric threshold — it goes in once you have assets: copy, a logo and signals about who you are looking for | The account stays on phrases alone. PMax reaches where a keyword list cannot — a median of +$721 in monthly spend |
| 7 | Negative search terms | Always, in the narrow version: a search term with at least 20 impressions and zero clicks | Negatives you skip cannot be made up later — the account slides under the required 5% click-through rate and only then has to be rescued with a rebuild |
| 8 | Dynamic ad groups (DSA) | Once keyword-based groups hold at least 6% CTR and dynamic ones stay under 45% of account impressions | Switched on with no condition they can hurt: above 70% of impressions they pull account click-through rate down by 4.4% |
What we changed on our side
We now know what works and in what order. The remaining question is who does it — because the bottleneck is the page content, not the campaign settings. This part is about what we therefore take on inside our services.
The bottleneck is the page, not the campaign
That is why we extend the service and not only the report. Telling a client they are short of pages is worth nothing if nobody is there to write them.
We read the effect of landing page changes after two months at the earliest, and we settle it after three. An earlier reading measures ramp-up, not effect: a new keyword delivers 30% of its eventual volume in the first week and reaches its own level only around the tenth. A reading after one month therefore shows about half the effect — and it is easy to call that “barely worked” and turn back from a road that does work.
What we added to the services so this can actually be fixed
Measurement alone will not build anyone a page. Since the study points at page content and structure as the bottleneck, we widened what we do inside the subscription — instead of handing the client a list of things to do themselves.
| Package | What the price includes |
|---|---|
| PING Recurring account audit |
The report only — we change nothing in the account. This study adds five things to it that were not there before:
|
| Assistant Fixing errors and making the changes that get the budget working |
Core of the package
The report plus fixing the errors in the account:
Proactive work
Every change from both blocks goes to you for one-click approval — nothing enters the account or the site without your consent. |
| VIP A dedicated account manager and direct support from the Webup! team | Everything above included and done by us — no task list on your side and no approving each change one by one |
| SMART (add-on) | Hosting and domain plus site management: the content we write and the titles we fix go live immediately, with no waiting for a developer on the organization’s side |
Most keywords never show once
The conclusions and the decisions are above; from here on come the measurements they rest on. We named the problem at the start — accounts hold keywords that never enter the auction. Here is the breakdown of what that number is made of.
We started by taking the set apart: what exactly happens to every keyword in the accounts we measured. 83% rows are dead structure — paused campaigns and groups left over from years past. Take them out and what remains is the live list, and that is what tells the whole story. The average account holds 158 keywords today; let us see what happens to them. Every keyword is in one of four states:
On the first two states the bars are the same height. An account that does not spend has no worse phrase selection: the same share with no demand, the same share rejected by Google. Those two states take about the same slice of the list in both groups — 26% i 25%.
The whole difference is in how the rest divides. On stalled accounts, 73% losing the auction sit against 0.5% getting impressions. On accounts that spend the grant — 55% against 21%. The same pool of keywords, two completely different auction outcomes.
An account that spends and an account that stalls have equally well-chosen keywords. What separates them is what share of those keywords enters the auction at all.
List size goes along with it: a weak account has a median of 110 live keywords, a strong one 390. But the count alone is not what makes the difference — the same 110 phrases would produce impressions on the strong account and not on the weak one.
Five keywords. That is how many actually work in a typical Ad Grants account, and they account for practically all of its traffic.
That moves the question from “which keywords to pick” to “what makes the keywords you picked start at all”, and that is a question about the landing page: whether it exists, whether it is on topic, and whether its title names that topic the way the user names it.
It also shows in where accounts lose impressions. Google reports how much of the available impressions a campaign lost and why: too small a budget, or the auction lost. Those two reasons add up to the whole loss.
Accounts do not lose impressions for lack of money. They lose them because they lose the auction, while the money sits unused.
Where those keywords are: two thirds sit in groups that never started
The above says how many keywords do not start. It does not say where they are, and that turned out to matter more. We broke 387k keywords from live campaigns down not by keyword but by the ad groupsthey sit in.
| Accounts by share of keywords that start | median groups | median keywords | % of keywords in groups that never served once |
|---|---|---|---|
| bottom quarter | 11 | 303 | 99% |
| middle (322) | 17 | 401 | 68% |
| top quarter (161) | 19 | 407 | 36% |
That changes the order of thinking about the fix. Adding keywords and pages to an account where 99% of the keywords sit in dead groups is pouring into the same bucket: the problem is not a lack of material but where the material sits.
What actually moves the result
The whole study comes down to one question, so here is the answer stated plainly: a keyword produces a result once it passes three gates. Below — how many impressions and clicks are collected by 100 keywords at each level.
| Gate | impressions per 100 keywords | clicks per 100 keywords | % of keywords that start |
|---|---|---|---|
| Starting point every phrase in the sample for which we have complete data | 390 | 41 | 5.8% |
| 1. A phrase with real demand the 200–10,000 searches/mo band | 940 | 102 | 11.3% |
| 2. A landing page on the phrase’s topic the phrase’s words appear in the page content | 1,171 | 127 | 13.8% |
| 3. The phrase in that page’s title the title names the topic the way the user names it | 1,944 | 195 | 21.4% |
A hundred random phrases from the list collect 41 clicks a month. A hundred phrases that passed all three gates — 195. Almost five times as many, on the same budget and the same number of phrases.
One map: what an account looks like at each level of spend
Before we take the levers apart — the average account at every level of grant use, from zero to the ceiling:
An account that spends the whole grant does not do one thing differently. It has more keywords, more ad groups and more pages at once — the three grow together and in fixed proportions.
In numbers: an account using the grant has twelve times more keywords than a stalled one, five times more ad groups, ten times more pages, and targets demand an order of magnitude larger. Which of those columns is the cause — is the point to have many phrases, or for them to add up to large demand? The answer is clear and the opposite of the intuition.
| The account’s list | accounts | total demand for the phrases | spend | clicks |
|---|---|---|---|---|
| Many phrases, each with small demand | 122 | 234,790 | $407 | 144 |
| Few phrases, each with large demand | 122 | 279,295 | $66 | 33 |
| Many phrases with large demand | 191 | 1,852,830 | $450 | 172 |
| Few phrases with small demand | 191 | 53,340 | $81 | 37 |
What decides is the number of phrases, not the demand behind them. An account with many niche phrases collects 144 clicks a month; an account with a few popular ones, 33. And it is the second one that aims at larger total demand.
The same shows when the two factors are separated statistically: the link between phrase count and spend barely weakens, once total demand is taken into account, which means it works on its own. The reverse does not: average demand per phrase, once phrase count is taken into account, means practically nothing. The mechanism is the same as everywhere in this study: every phrase is a separate entry into a separate auction. A phrase with ten thousand searches that loses the auction gives zero, while 100 niche phrases of which 20 produce impressions give traffic.
So the number of clicks is a consequence, not a cause — it grows because the phrase count grows. Two ratios hold whatever the scale: one ad group carries 8–11 phrases, and one landing page carries 2–3 ad groups. On stalled accounts that second ratio is completely different — five and a half groups per page, meaning everything points at the same two pages.
1. Lifting the $2 cost-per-click cap
Ad Grants imposes a $2 cost-per-click cap — against commercial competition in many sectors that is a bid too low to reach the results page at all. The cap has an exception, though: it does not apply on conversion-based strategies.
| Bidding strategy | median CPC | highest CPC | % of clicks above $2 | Does the cap apply |
|---|---|---|---|---|
| Maximize conversions | $1.20 | $22.23 | 33.5% | no |
| Maximize conversion value | $5.94 | $18.75 | 88.6% | no |
| Maximize clicks | $0.80 | $2.00 | 0.0% | yes |
| Target impression share | $0.91 | $2.00 | 0.0% | yes |
| Manual CPC | $0.22 | $0.77 | 0.0% | yes |
This is the single strongest change we found — and, as we show below, not enough on its own. The ceiling says how much you may pay, not how much the account gains from it; we measured the effect on accounts where both strategies ran side by side, each compared against itself.
| Campaigns | share below 10% | above 10% |
|---|---|---|
| with the $2 cap 302 campaigns | 94% | 6% |
| without the cap 1,265 campaigns | 95% | 5% |
Lifting the cap does not raise impression share. The account stays in the few-percent range of available auctions — it just enters up to five times more of them.
At account level the difference looks like this: an organization running only capped campaigns collects 62 impressions and 7 clicksa month; an organization with at least one conversion campaign — 302 impressions and 34 clicks. Impression share cannot tell them apart, because both groups sit below the reporting threshold.
We also checked what happens to an account right after the change — on 132 cases of crossing the $2 threshold in the portfolio’s history. In the first quarter after the change the median change in clicks is +10 and in spend +$15. In other words: for a typical account the strategy change is neutral in the short run, and the effect comes from what you do with phrases and pages afterwards.
The $2 cap still on: first check whether there is anything to optimize
Switching to a conversion strategy looks like one click in the interface, and sometimes it is — but on some accounts it is not, and it is worth knowing why. 40% of the accounts we measured recorded not one conversion in the whole period. A conversion strategy on such an account has nothing to optimize for, and Google usually will not start it.
So this diagnosis runs in two steps: check whether conversion tracking works at all — whether the events are recorded and whether they are attributed to campaigns — and only then switch the strategy. If conversions are there and the cap is still on, that is plain neglect and the change takes a few minutes. If there are no conversions, switching the strategy without starting them will stop the account instead of starting it.
2. Phrases with real demand
“A phrase with demand” is too vague to work with, so we measured two things against monthly search volume: the chance that a phrase starts at all, and how many impressions a hundred such phrases bring.
The first band does not work at all and that is a stronger statement than usually assumed. We checked it directly: we took the impressions and clicks of every phrase in the sample and set them against its volume from Keyword Planner.
| Phrase volume | phrases in the sample | how many of them ever got an impression | impressions per phrase |
|---|---|---|---|
| under 10 searches/mo | 6,640 | 1.6% | 1.5 |
| 10–49 | 149,329 | 3.2% | 2.0 |
| 50–199 | 70,584 | 7.0% | 2.8 |
| 200 and more | 61,204 | 11.4% | 11.9 |
Both curves keep rising, but not in the same way. The chance of starting flattens out around a thousand searches, while the number of impressions keeps growing — because a more popular phrase, once it starts, simply delivers more. Hence two different conclusions for two different situations:
| The account’s situation | Where to take phrases from | Why |
|---|---|---|
| An account starting from zero | the long tail included | The 1–50 band produces the most impression-generating phrases by count — 5,540 across the whole sample, more than any other band — simply because there are so many of them. |
| An account that has to use up the grant | the 200–10,000 band | A hundred phrases from that band give 1,168–3,587 impressions against 507 from the tail. The tail alone will not get you to full grant use. |
| A phrase above 10,000 turns up | always add it when it fits the topic | The highest return per phrase of any band. The caveat is about planning, not about value: there are 1,746 such phrases in the sample, so they are an addition, not a foundation. |
Phrase competition: the easiest band comes out worst
Keyword Planner assigns every phrase a competition level: low, medium or high. The reflex is obvious — the budget is free but limited, so take the low-competition phrases, because they are easier to win. The data says the opposite.
| Phrase competition | phrases in the sample | % of phrases that start | impressions per 100 phrases | clicks per 100 phrases |
|---|---|---|---|---|
| low | 75,490 | 6.2% | 543 | 58 |
| medium | 17,166 | 7.9% | 847 | 95 |
| high | 23,015 | 4.7% | 155 | 16 |
Why low competition comes out worse than medium. Because that measure does not say how hard the auction is to win — it says how many advertisers bid at all. A phrase nobody bids on is usually not an opportunity but a phrase with no commercial value: people type it, but it does not pay anyone to bid for it. Google’s own suggested bids show it directly: at low competition the median suggested top-of-page bid is zero — Google does not report one at all, because there is nothing to compute it from.
High competition is more interesting, because the stock explanation — “the grant loses because of the $2 cost-per-click cap” — does not survive the data. High-competition phrases are cheaper than medium ones: the median suggested top-of-page bid is $1.74 against $2.56 at medium, and 44% of them come out above the required $2 versus 58% at medium. So the most expensive band is the one that works best. The control settles it: on accounts that no longer carry the $2 cap, because they moved to a conversion strategy, the order of the bands is identical — medium starts most often, high least often. Lifting the cap raises all three bands and does not change their order.
What to do about it. When you build the list, do not filter phrases by competition in either direction. The one thing worth watching: do not build an account only from low-competition phrases, because that is usually a sign the list is made of terms nobody searches with any intent to do anything.
There is no band to avoid — other than phrases with no demand, which will never start.
The cheapest phrases to add are sitting in your organic results
Since every phrase needs a page on its topic, start with the phrases whose page already exists. There is a simple test for them: if the site shows up for a phrase in organic results, Google has judged one of its pages to be the answer to that query — for free and without our involvement. So we set the phrases organizations rank for against what they have in their ad accounts.
| Phrases the organization ranks for in organic results | phrases | share |
|---|---|---|
| already in the ad account | 2,494 | 10.7% |
| ranking, but not advertised | 20,859 | 89.3% |
| of those, in the 200–10,000 search band | 16,909 | 81.1% of the gap |
| of those, in the organic top ten | 7,596 | 36.4% of the gap |
| of those, in the organic top three | 2,885 | 13.8% of the gap |
Four fifths of these phrases fall exactly in the 200–10,000 band, the one that comes out best in this study. It is the only phrase list in the whole material you can add to an account the same day — without writing one character of content.
The difference from ordinary Keyword Planner suggestions is fundamental: the Planner only tells you how many people search for something, and then you still have to build a page for that phrase. A phrase from organic results passes both gates at once — it has demand i it has a page. Phrases from position eleven and below are also the best list of topics for new content: the site already brushes against them, so the work pays twice, in ads and in organic results.
Why this changes the result at all: ranking is not the same as collecting clicks
It is easy to assume the organization already has traffic on those phrases, so an ad adds nothing. The opposite is true, and that is the heart of this section. Click-through rate in search collapses with position: the first organic result collects on average 28.5% of clicks, the second 15.7%, the third 11%, the tenth only 2.5%, and every position on page two under 1% (a study across more than 80M phrases and billions of results).
And now our side of that equation: 13.8% of the gap phrases sit in the organic top three, and 63.6% sit outside the top ten. So the on-topic page exists and Google has already judged it — it just stands in a place where practically nobody clicks. This is not traffic we have; it is traffic we do not have, even though we have the content.
And that is exactly what the grant changes in one move: the ad puts the same page above the organic results, with none of the ranking climb that takes months. Cannibalization on phrases where you rank first organically is real, but in Ad Grants it is free — the budget we pay with would go unused anyway.
Does it really work — we checked
The above is reasoning, not measurement, so it had to be checked. There is no way to measure what the gap phrases will deliver after being addedwithout an experiment. What can be checked is the thing the whole idea rests on: some of the organically ranking phrases are already in the ad accounts. It is enough to compare them with the rest of the phrases in the same account.
| Phrases in the ad account | phrases | % with impressions | impressions per phrase | clicks per phrase |
|---|---|---|---|---|
| also ranking in organic results | 2,528 | 31.3% | 37.3 | 7.07 |
| all the rest of the list | 385,404 | 4.0% | 2.2 | 0.22 |
| the same, excluding phrases with the organization’s name | 2,114 | 29.0% versus 3.7% | — | — |
A phrase the site shows up for in organic results generates impressions in the campaign eight times more often than the average phrase on the list. It is the strongest phrase-selection rule we found in this study, and the only one that needs not one character of new content.
This does not contradict the on-page SEO elements that failed in this study. The ones that failed are exactly those that do not say what the page is about: speed, link profile, domain age. The two that passed are the ones that do — the phrase in the title and the phrase in the content. An organic ranking is that same signal read as a finished result: if Google judged this page to answer this query, it will judge the same in the auction. So the conclusion is not “SEO does not work” but “the part of SEO that describes the page’s topic works”.
One trap: the keyword counter is easy to inflate
Since the phrase count is what decides, there is an obvious shortcut: add three hundred and call it done. It will not work, because one fifth of a typical list is phrases with no measured demand — the band with zero chance of starting. Adding those raises the counter and changes nothing. (“No demand” here means either under 10 searches a month, or “Keyword Planner does not know this phrase at all”.) To settle it, we compared accounts with the same list size, differing only in the share of phrases with real demand — otherwise the result would just say that “more phrases with demand” means “a bigger account”.
| Phrases in the account | accounts | lower share phrases with demand | higher share phrases with demand | difference |
|---|---|---|---|---|
| 50–199 | 77 | $21 | $27 | 1.3× |
| 200–399 | 159 | $81 | $100 | 1.2× |
| 400–799 | 168 | $222 | $216 | no difference |
| 800–1,599 | 130 | $203 | $459 | 2.3× |
| over 1,600 | 83 | $582 | $2,190 | 3.8× |
The trap is conditional, though, and that has to be said plainly: up to eight hundred phrases the share with demand makes no difference to the result — on such an account almost everything in it is something people really search, so there is nothing to clean. It kicks in higher up, because the longer the list, the more terms collect on it that nobody searches.
And there it makes all the difference. Two accounts with a thousand phrases in the report can spend $203 or $459 — depending on how many of those phrases anyone actually searches. The same number in the counter, more than twice the result. Above 1,600 phrases the spread reaches fourfold, because in the weaker half half the list is dead: 49% phrases with demand against 91% in the better half.
So we set the target on phrases with demand, not on the raw keyword count, and we add an allowance: to have 400 phrases that can work, the account needs about 500 — a counter showing 400 really means 320 working.
Does this only work on stalled accounts — or on big ones too
This is the question every organization already spending something asks: “fine, but we already have four hundred phrases and twenty pages, will adding more still do anything?” We checked it directly, splitting the sample by spend level and comparing, within each level, accounts with a smaller and a larger phrase list.
| Accounts spending | accounts | median spend with the smaller phrase list | with the larger | difference |
|---|---|---|---|---|
| all | 571 | $31 | $218 | 7.0× |
| over $10/mo | 456 | $70 | $304 | 4.3× |
| over $100/mo | 266 | $296 | $716 | 2.4× |
| over $500/mo | 114 | $1,080 | $2,114 | 2.0× |
| over $1,000/mo | 73 | $2,391 | $3,689 | 1.5× |
The relationship does not disappear at any level. Even among accounts spending over a thousand dollars a month, those with a larger phrase list spend one and a half times more. The multiplier shrinks, not the direction.
3. Account structure: pages, ad groups and phrases per page
Every phrase needs a page, and every ad group points to one address, which is why the number of pages is a lever and the number of campaigns is not. How many pages you need — below, the median monthly spend against the number of promoted pages, over the last 21 months.
The relationship is steep and has no plateau: moving from 10–15 pages to 16–25 is a two-and-a-half-fold rise in spend, and an account above forty pages spends 134 times more than an account with one or two. A typical organization today promotes three pages.
What accounts that use the whole budget look like
You can ask it the other way round: what do the accounts that actually use the granted amount look like. Those spending over $8,000 a month have a surprisingly uniform profile:
| Median account profile | spend ≥ $8,000 | spend ≥ $4,000 | whole sample |
|---|---|---|---|
| Promoted pages | 23 | 23 | 5 |
| Keywords | 611 | 560 | 158 |
| Keywords per page | 19.4 | 20.7 | 28.5 |
The operational target: 16–40 promoted pages and 400 phrases with demand, that is 10–25 phrases per page. Accounts that use the full budget have a median of 23 pages, 611 keywords and 19 keywords per page.
The distribution of phrases per page confirms it from a third angle — median spend against density:
How many phrases belong in one ad group
The rule the industry repeats is “10–35 phrases per group”. We checked it on 8,820 ad groups and it looks different from the way it is told.
| Phrases in the ad group | groups | % of phrases that start | impressions per 100 phrases |
|---|---|---|---|
| 1–3 | 387 | 11.3% | 3,523 |
| 4–9 | 851 | 11.0% | 1,261 |
| 10–25 | 5,011 | 9.7% | 504 |
| 26–50 | 2,035 | 9.5% | 755 |
| over 50 | 536 | 4.1% | 415 |
We also checked it inside the same accounts, comparing one organization’s mid-sized groups with its overloaded ones. Groups up to 25 phrases start better than groups above 50.
The reason is the same as with pages: an ad group points to one address. Fifty phrases in a group is fifty phrases pointed at one page, and then most of them have little to do with it. The bottleneck is not the group itself, but that it grew beyond the page it serves.
Is there a ceiling above which adding phrases stops paying off?
We checked it directly: the median monthly account spend across bands of list size. If a ceiling existed, the curve would have to flatten somewhere.
There is no ceiling. Spend rises across every band — twenty-five-fold between the smallest and the largest list — and nowhere is there a point where it stops.
This is not because the thousandth keyword is worth more than the first. The list grows together with pages and titles: an account that reached a thousand phrases usually has a dozen or several dozen landing pages and titles matched to the phrases. So what grows is not the keyword count but the share of keywords entering the auction — the same list spread over three pages would give a many times worse result. The conclusion: there is no point at which adding phrases stops paying off, as long as you add them together with pages.
The curve has one clear threshold at the bottom: an account with a list under 200 phrases practically does not exist in the auctions — the most common diagnosis in the whole sample. Doubling the phrase list doubles the account’s spend (+101%), and raises clicks by 85%.
So it is worth setting the target from the amount rather than from the phrase count — how many phrases and pages a given amount works out to, we show further on, on the profiles of real accounts.
One rule does not depend on scale and always holds: 10–25 phrases per page, and one ad group points to one page. Everything else — how many pages, how many phrases — follows from how much the organization wants to spend.
One thing from campaign themes belongs to account structure: local ad groups. A phrase with a city or region starts twice as often as any other — 11.9% of keywords against 5.3–6.4% — and inside the same account local groups win 64% of comparisons against other themes, the most of all five. Yet only 45% of organizations run them, so on more than half of the accounts this is a ready lever nobody has touched. The full comparison of the five campaign themes is in the extra findings.
Zero impressions: this is not a keyword problem but a page-count problem
This diagnosis sounds the most cryptic, so let us take it apart. Accounts that have keywords and zero impressionslook very uniform in our data:
| Median profile | accounts with zero impressions | healthy accounts |
|---|---|---|
| Landing pages | 3 | 23 |
| Keywords | 101 | 611 |
| Keywords per page | 34 | 19 |
Three pages and 100 keywords. That is the whole diagnosis — the account has nowhere to send traffic, so no phrase has anything to win the auction with.
What to do: pick 8–10 topicsthe organization genuinely covers, build a separate page for each with the phrase in the title and in the content, and only then spread the phrases across them, 10–25 per page. The reverse order — phrases first, pages later — will not work, because the phrases will have nowhere to lead.
How many pages and phrases — depending on how much you want to spend
The chart at the start of the article answered “what does an account that spends this much look like”. Here we turn it into the question an organization actually asks: how much do we need to have in order to spend that much, and the numbers come out different. There is no single “400 phrases” threshold: the target follows from how much of the grant the organization wants to use, and the number of pages and phrases follows from the target. The conversion from our data:
| Target monthly | phrases in the account | of those, generating impressions | landing pages | phrases per page |
|---|---|---|---|---|
| ok. $300 | 400 | 80–100 | 16–20 | 20–25 |
| ok. $1,000 | 600–1,000 | 150–250 | 26–40 | 15–25 |
| full use ($8,000 and up) | 600+ | 250+ | 23+ | 19 |
Not “add 400 phrases”. Decide how much you want to spend and work the page count out from that, because pages are the bottleneck, not phrases.
4. Landing page quality: topic and title
Content: it is not about how much there is, but whether it is on topic
We pulled the content of 5,482 landing pages and checked whether the keyword we advertise on appears at all on the page it leads to — comparing within the same account, and often the same ad group. 24% of phrase → page pairs have no coverage in the content at all (353k pairs); 11% at the account median, but on some accounts it is most of the list.
| Compared inside the same account | phrase fully covered by the content | covered partly or not at all |
|---|---|---|
| Chance that a keyword picks up traffic at all | 32.7% | 11.6% |
| Impressions per keyword | 16.4 | 7.0 |
| Poor landing page ratings from Google | 45.6% | 68.3% |
| Quality Score | 6.32 | 5.65 |
Length works up to a point, then stops. Below: impressions collected by one landing page against the length of its text:
This chart has a flaw: impressions per page also rise because a longer page has more phrases pointed at it. To measure quality itself, we took the page rating for each keyword separately — a measure with a denominator, independent of the phrase count:
The longer the page, the less often Google rates it as poor. And it does not stop at eight thousand characters. The share of poor ratings falls from 73% at 2–4k characters to 42% at 13–16k.
Comparing longer and shorter pages inside the same organization gives the same direction, so we are not simply measuring bigger websites.
This concerns the page rating, though, not the account’s spend, and that difference must not be lost. We measured the effect of length on spend separately, controlling for PMax share, campaign count, account age, keyword count and group count: text length is not related to spend in any way. A medium or strong relationship is ruled out; a very weak one never can be, but even that would be no reason to order characters.
Relevance takes length’s place — whether the phrase’s words appear in the content — and relevance passes the test at account level too: after the same controls the relationship with spend holds. The lowest relevance quarter spends a median of $26, the higher ones from $187 to $1,172. That is a floor, not a scale — the top quarter spends less than the third, so relevance gets an account off the ground rather than lifting it higher the higher it goes. At keyword level:
| Are the phrase’s words on the page | impressions per keyword | “poor landing page” ratings | chance that a keyword starts at all |
|---|---|---|---|
| every word of the phrase present | 16.1 | 43.9% | 29.1% |
| partly — at least one | 7.1 | 53.5% | 15.9% |
| not one word of the phrase appears | 3.0 | 62.2% | 8.3% |
Five times more impressions comes not from the length of the text but from whether the phrase appears in that text at all. And one in four phrase → page pairs has not a single word of its phrase on the landing page.
Three conclusions for the brief, and none of them says “write long”. The whole big jump sits between 4 and 6 thousand characters (73% poor ratings falls to 56%); above that the rating improves slowly and unevenly (52% · 55% · 47% · 42%) and does not translate into spend. Hence: get above 4k characters, aim for around 8k and stop there. Order a list of phrases to cover, not a character count — every phrase in the content, the most important one in the title and the H1. And on a limited budget several pages of 8k beat one of 20k: each page is a separate place for a separate group of phrases, and 20k characters on one page is still one page.
The page title — the strongest of the elements that pass the test
We showed the effect of the title alone above, in the gates table. Here — what the title no replaces, and how common that mistake is.
We tested every on-page optimization element separately: whether a title exists and how long it is, the meta description, one H1 versus none or many, heading hierarchy, number of internal links, images, text-to-code ratio. All comparisons inside the same account. One passes: whether the keyword is in the page title.
| Page element | how many more phrases start generating impressions |
|---|---|
| The phrase in the page title | 25% → 43% |
| The phrase in the H1 | the same as the title |
| The phrase in the URL | the same as the title |
Classic SEO hygiene does not carry over to ads. One thing does: whether the page title names what the user is looking for.
The scale of the neglect is enormous, but it has to be stated precisely, because it is easy to count something other than what you say. Counting phrase → page pairs the account actually points at: 63% of pairs have not one meaningful word of their phrase in the page title (350,280 pairs, 5,311 pages, computed on the full set). Counting by page, the criterion is looser and so is the result: 14% of pages miss every one of their topics. And counting most strictly — how many of an account’s pages carry one of their phrases in the title in full — the median is 33% and 347 of 516 organizations (67%) are below half.
This is the cheapest fix in the whole study. It needs no new content, no new page and no budget — it needs one line rewritten in the content management system.
Does the phrase have to be in the title literally, or is it enough that the page is about it
All the numbers above are literal — they compare the phrase’s words with the title’s words. They cannot see that a page about “care for older people” is exactly on the topic of the phrase “help for seniors”. And that is what decides whether you have to write unnatural titles. So we separated the two: a language model assessed 75,720 phrase → page pairs purely on topic, regardless of the words used, and next to it we put the literal measure and Google’s own landing page rating — the only judge outside our definitions.
| Page title | pairs | how many phrases start generating impressions | Google rating “above average” |
|---|---|---|---|
| has the phrase literally and is on topic | 1,673 | 33.5% | 12% |
| is on topic, but without the phrase | 7,085 | 15.5% | 9% |
| has the phrase, but is not on topic | 108 | 15.7% | 0% |
| neither on topic nor with the phrase | 8,514 | 9.5% | 4% |
The simplest reading is two steps, which have to be done in this order:
The last bar answers “maybe the title alone is enough”. It is not: the phrase in the title of a page about something else gives 15.7%, exactly the same as step one alone, and not one “above average” rating from Google in 108 cases. Topical relevance is the entry condition, the phrase in the title is the amplifier — the amplifier gives more, but with no entry there is nothing to amplify.
| Criterion | % more phrases gets impressions | in how many accounts it works |
|---|---|---|
| phrase in the title, LITERALLY | +20.6 | 92% |
| page ON THE TOPIC of the phrase | +9.8 | 76% |
| phrase in the H1, LITERALLY | +21.5 | 96% |
| H1 ON THE TOPIC of the phrase | +9.8 | 75% |
5. Consent Mode behind the cookie banner
A consent banner blocks measurement tags until “accept” is clicked, and on average one user in fourrefuses. Consent Mode lets Google model conversions without cookies: instead of vanishing from the data, some of those people come back into measurement as modelled conversions.
The practical effect is simple. There are more conversions, so the account reaches the threshold for turning on maximize conversions sooner — and that strategy then has something to learn from and works better. Without Consent Mode the same campaign collects that threshold more slowly, or never.
This applies to markets where consent is required: the European Union, the United Kingdom and Switzerland. What counts is where the ads are shown, not where the organization is — an account sending traffic only outside those markets does not have this problem.
6. Dynamic ad groups (DSA) — the only lever that works with one switch
These are groups where Google picks the queries from the site’s content — with no keyword list — and writes the ad headline itself. On 10 of the 32 accounts under the required threshold they produce most of the impressions and click two to four times worse than keyword-based groups.
The damage is not spread evenly, and that is the most useful part of this measurement — it grows along with how much of the account the dynamic groups take over. Two measures say the same thing from two sides: how much CTR the account loses, and on what share of accounts dynamic groups come out worse than keyword groups.
While dynamic groups stay an addition, they cost about 1% CTR and on two accounts in five they even come out better than keyword groups. Once they take over the account, the loss grows to 4.4% CTR — and against the required 5% threshold that is the difference between a healthy account and one that needs rescuing.
Turning them off is not free, though. On accounts where dynamic groups serve, one click in four passes through them at the median (24%) and one in three on average (32%) — so the account raises CTR and loses traffic at the same time. And every second account runs them.
From that comes a gate we apply in both directions, because a dynamic group sometimes has to be switched off and sometimes added. A dynamic group is a sensible addition exactly when the keyword-based part of the account clicks at 6% or better (so there is a buffer to give away) and dynamic groups are not already producing more than 45% of account impressions (so they are not taking it over). Outside those conditions we do not propose one at all, and where one already runs — we propose switching it off. The gate is deliberately one-sided: it never encourages, it only blocks.
5% CTR — in what order we rescue it
Hence our sequence, now in three steps: first we check the dynamic groups, because that is the only lever that works in one move — but always showing the organization how much traffic it gives away for it; then the narrow rule everywhere, because it costs 2% of clicks and there is no reason not to apply it; and wider negatives and pausing keywords — only on accounts at risk, after showing the organization what it costs in traffic.
7. Performance Max
Performance Max — available in the grant since 2024 — is the strongest single reach lever once the cap is lifted. Accounts with PMax spend 4.2× more and collect 3.0× more clicks than accounts without it, and switching it on where it was absent gives a median +$721 in monthly spend.
The mechanism is different: PMax does not bid on specific keywords, it picks audiences and placements itself from the materials you give it. That is why it reaches where a keyword list cannot, and why its result depends mostly on the quality of the assets — the copy and the signals about who we are looking for.
We also checked where Performance Max actually shows up, because there was a claim that in the grant it only ever runs on search. The claim holds, and more strongly than we expected. Performance Max on a grant account appears in practically only two places: search and Maps. On grant accounts with Performance Max, search took 25.06M impressions (200 accounts, CTR 5.19%), Maps 709k (16 accounts, CTR 0.30% — Maps barely get clicked), and 739k Google booked on 57 accounts in the catch-all “multiple networks” bucket, which it does not break down. In proportion: 94.5% search, 2.8% the catch-all bucket, 2.7% Maps.
The Display network, YouTube and Discover took not one impression. Not “few” — zero, across every grant account and the whole history, against 25.06M impressions in search.
The number you have to know is different from what we expected: the click-through rate of all Performance Max on grant accounts is 5.05%, which sits exactly on the required 5% threshold and crosses it in both directions depending on the month (above the threshold in 14 of the 22 months measured). That is not a reason to avoid PMax. It is a reason to switch it on in an account that has headroom above the threshold, because PMax itself adds no headroom.
And it is spend with no extra budget, because it is still the same grant. An account without PMax leaves that part unused.
The same caution applies to dynamic ad groups — mechanically both tools do the same thing: they hand Google the decision about what the ad shows on, in exchange for reach a keyword list will not catch, and both pay in the same currency: a lower click-through rate measured against the required 5% threshold. So both are an addition to a working keyword side, not a replacement for it.
8. Match types and negative search terms
Neither is a lever — neither will get an account moving — but both cost you when ignored.
- Do not narrow the match typeAmong accounts without the cap, broad match has 5.9% of keywords producing impressions, phrase match 0.7%, and exact 0.1%. Compared inside the same ad group, narrowing to exact lowers a phrase’s chance of starting by 0.64% in absolute terms. Narrowing looks like tidying the account up; it is switching it out of the auction.
- Exclude search terms that do not get clicked — but one at a time30.7% of impressions come from queries far from the phrase’s intent, and this is not material for a wholesale cut: that tail has a 9.0% click-through rate, so cutting all of it would cost roughly a quarter of all clicks. What is free is the narrow rule — “at least 20 impressions and zero clicks” — which across 269 accounts covers 1,608 queries and takes away not one click. We set this out below, with the numbers.
What do working accounts look like?
So far these were medians. Below are specific accounts from the top of the sample, described by every factor this study found to matter. Two rows are there on purpose: they contradict the rules of this study — one the page count, the other the list size — and they are the ones that say the most.
| Account | spend | CTR | price clicks | landing pages | phrases with demand |
|---|---|---|---|---|---|
| An industry association the model case: everything done | $7,599 | 20.0% | $0.73 | 83 | 2,116 8.4M in demand |
| A foundation — food aid | $9,644 | 6.9% | $0.89 | 27 | 1,032 1.1M in demand |
| A regional tourism organization with Performance Max | $9,679 | 15.9% | $1.33 | 21 | 725 631k in demand |
| An information portal for NGOs warning: CTR below the 5% threshold | $5,772 | 4.4% | $0.81 | 28 | 1,181 1.4M in demand |
| A digital education foundation the cheapest clicks in the table | $9,565 | 13.8% | $0.64 | 7 | 43 57k in demand |
| A second tourism organization counter-example: 3 pages | $8,540 | 18.3% | $1.56 | 3 | 513 489k in demand |
| A relief foundation counter-example: 116 phrases, with Performance Max | $8,199 | 23.6% | $1.30 | 4 | 109 225k in demand |
The grant can be spent on three pages — the three-page account hits the ceiling, paying $1.56 per click against 73 cents on the account with 83 pages. But that is two accounts, not a pattern: in the same table the account with seven pages has clicks cheaper still than the one with eighty-three.
Those two rows do not refute the rule about pages — they show its limit. Without pages on the topic of its phrases an account usually does not move at all, which is why the median spend at three pages is a dozen or so dollars. There is one way out of it: having a niche so narrow and so well covered that a few pages are enough to win its auctions — then the account spends the grant, but pays twice as much per click, because it enters fewer auctions and has to win the more expensive ones. So pages decide both whether an account starts and what it pays.
Total demand shows the same as the quarters above: the digital education foundation has 43 phrases with demand and 57 thousand in total demand — the least in the table — and pays the least per click of all seven. A narrow, well-chosen niche with seven pages beats a wide list with no pages. Demand is not the cause of anything here; the cause is how many phrases have somewhere to lead.
Two accounts show the limits of that strategy. A relief foundation has 116 phrases, 109 of them with real demand — a tiny list, but cleaned to zero waste, and a 23.6% click-through rate. It works as long as the niche is deep; it does not scale further, because there simply are no more phrases. The NGO portal, in turn, has everything this study teaches — 28 pages, 1,181 phrases with demand — and at the same time a 4.4% CTR, below the threshold the Ad Grants rules require.
Account click-through rate: not through ads, through search terms
This is where the work most often goes in the wrong place. The reflex is: the click-through rate dropped, so we have to fix the ads — rewrite headlines, add variants, raise the ad quality rating in the interface. We measured it and that road does not work.
We compared pairs of ads from the same account, with the same number of headlines and descriptions, differing only in Google’s rating — 899 pairs. The difference in click-through rate: zero. The same at account level: accounts where 5% of the ads are rated poor have an 11.4% click-through rate, and accounts with 43% poor ads — 11.1%.
Click-through rate is governed above all by which queries the account shows up on at all. The ad answers the query — if the query is off topic, no headline will turn that around. This does not mean the wording of the ad is irrelevant: we measured Ad Strength — the rating Google gives only to responsive ads, and which mostly says whether every field was filled in. Two thirds of the serving ads in the sample do not have one at all. That is something other than the Quality Score, which Google computes for a keyword, not for an ad.
And search terms are the tool we do hold. Google shows the ad not only on the phrase in the account but on everything it treats as a variant of it — and that is where the loss happens. The question is what rule picks the queries to exclude: not “cut the worst 20%”, because nobody knows where that line runs in their own account, but a condition that settles a single query on its own numbers. We tested five such rules on 269 accounts with traffic.
| Rule — exclude the search term if… | click-through rate (median) | search terms in the sample | clicks lost | accounts that lose something |
|---|---|---|---|---|
| nothing — the current state | 12.41% | — | — | — |
| it has at least 20 impressions and zero clicks | 12.57% | 1,608 | 0 | 0 z 269 |
| it has at least 100 impressions and zero clicks | 12.41% | 101 | 0 | 0 z 269 |
| it has at least 30 impressions and a click-through rate below a quarter of the account average | 12.84% | 1,096 | 733 | 73 z 269 |
| it has at least 30 impressions and a click-through rate below half of the account average | 13.00% | 1,697 | 4,424 | 122 z 269 |
At the sample median none of these rules looks impressive — the best free one adds 0.16% CTR. Because the sample median is the wrong place to look. An account with a 20% click-through rate needs none of them. What counts is what happens on accounts sitting below the suspension threshold:
| On paid accounts with a click-through rate below 5% in August 2026 | accounts come out above the threshold | how much traffic it costs |
|---|---|---|
| literal negatives — 20 impressions and zero clicks | 2 | 2.0% of clicks |
| wider negatives, with the paid layer | 4 | more |
| pausing weak keywords | 7 | up to 39% of clicks |
| switching dynamic ad groups off | 6 | one switch |
This rule stays silent on most accounts. Those 1,608 search terms from the table above are not spread evenly across the sample: 140 of the 269 accounts have not one query that meets it. So this is not a monthly clean-up of the whole account but a tool that speaks up where something is actually happening.
How to read search terms — five steps
- Look at search terms, not at keywordsThese are two different lists. Keywords are what you want to show up on; search terms are what Google actually showed you on. The search terms report is in the Google Ads interface, and that is the material to work with.
- Sort by impressions, not by click-through rateA query with a 0% click-through rate and three impressions does nothing. A query with a 0.5% rate and four thousand impressions eats your metric. Start from the top of the impressions list among the ones that do not get clicked.
- Look for three recurring kinds of lossInformational queries whose answer is an article, not your offer (“what is…”, “how much does… cost”). Names of other organizations and companies the phrase happens to match. And the second meaning of the same word — “care” catches both care for an older person and website maintenance.
- Add negatives at campaign level, not at group levelA negative in the campaign applies to every group inside it, so one list serves the whole account instead of drifting apart across twenty groups. Keep group level for cases where a term should be excluded in one context only.
- Check once a month, not once a quarterOn a growing account new queries arrive all the time — and the 5% CTR threshold is computed from the last 30 days.
Extra findings: campaign themes, ramp-up and phrases with no demand
Three things came out along the way, off the main line of the study, and each changes something in day-to-day work: one in what the campaigns should be about, the second in planning dates, the third in what we no do.
Which campaign theme works at all
We classified every ad group in the sample’s live campaigns — 14,797 groups and 384k keywords — into five themes by topic and search intent. The most important part is the last column: the themes are not spread randomly, because organizations that run fundraising are different from those that do not — only a comparison inside the same account says something about the theme rather than about the organization.
| Ad group theme | share in the sample | % of keywords that start | impressions per keyword | clicks per keyword | wins inside the account |
|---|---|---|---|---|---|
| Local a phrase with a city or region | 9.5% | 11.9% | 5.9 | 0.90 | 64% |
| Service specific help, a programme, therapy | 34.3% | 5.3% | 16.5 | 2.16 | 53% |
| Informational a guide, symptoms, an article | 42.2% | 6.4% | 9.0 | 0.90 | 53% |
| Brand the organization’s name, “about us” | 7.2% | 5.4% | 3.4 | 0.73 | 52% |
| Fundraising: donations and support donate, support, help | 5.4% | 2.8% | 0.8 | 0.15 | 35% |
| Fundraising: the 1.5% tax share 1.5%, tax return, registry number | 1.4% | 1.4% | 1.3 | 0.08 | 15% |
Of the five campaign themes the market treats as equals, two really differ. Local beats every other. Fundraising loses to every other. Service, informational and brand are indistinguishable inside accounts — all three sit at 52–53%.
And the best theme is the least used. 45% of organizations run local campaigns. 47% advertise fundraising — so more organizations advertise donations than advertise their own area.
The 1.5% tax campaign: the exception among fundraising
In the table above fundraising loses to every other theme, and the 1.5% tax campaign comes out weakest of all five. That is how it looks across the sample as a whole — and it is the one place in this study where the average misleads this badly. We audit many accounts of public-benefit organizations and we see both sides at once: most of these campaigns really do stall, and at the same time there is a group of organizations where that same campaign turns out to be the strongest thing in the account. Below we show that second group — accounts where the 1.5% campaign is run the way this article describes, and where it really works.
| Month | clicks per account | spend per account | price clicks |
|---|---|---|---|
| November | 3,028 | $2,514 | $0.83 |
| December | 2,326 | $1,865 | $0.80 |
| January | 4,011 | $3,239 | $0.81 |
| February | 6,871 | $4,150 | $0.60 |
| March | 5,360 | $5,236 | $0.98 |
| April | 4,774 | $5,002 | $1.05 |
| May | 2,316 | $2,416 | $1.04 |
Across the whole season that is a median of 14,189 clicks and $11,722 per account, at 85 cents a click. At the peak, in March and April, one campaign spends $5,000 a month. February is the cheapest: 60 cents a click, because the auction is won cheaper then than in the last weeks of the filing season.
These accounts do not differ in luck but in a state you can reproduce: they run two tax campaigns on average instead of one, they have 36 landing pages instead of 20, and their phrases enter the auction twice as often — 4.8% against 2.3%. And the thing that matters most for a budget conversation: across the season they spend $35,633 on the whole account, of which 65% goes to the tax campaign. This is not a trick on a dead account but a working account aimed at one goal.
So the conclusion is the opposite of what the sample average suggests. The 1.5% tax campaign is worth having and worth preparing before the season — with eleven weeks of ramp-up that means starting in early November. Fundraising as a campaign theme does lose; this one, done properly, is the exception.
How long a campaign takes to ramp up, or how far ahead to launch it
The question is practical: an organization has a gala, a fundraiser or an intake on a specific day and wants the campaign running at full speed on that day — not just starting to warm up. We measured it on 236 campaigns, arranging their results into weeks counted from the launch day and comparing each campaign against itself: what percentage of its own settled level it reaches in each following week.
Half of campaigns reach 80% of their level in three weeks. That number is no use for planning an event, though, because campaigns behave very differently: 36% are ready in the first week, but some need fourteen. When you plan a date you have to allow for the slower ones, not for the average:
| How many campaigns are ready by then | lead time |
|---|---|
| half | 3 weeks |
| three quarters | 7 weeks |
| nine in ten | 11 weeks |
| all | 14 weeks |
An event with a flexible date — start 7 weeks ahead. A fixed date, like a gala or the opening of an intake — 11 weeks, about three months. The cost of being late is real, the cost of starting early is none.
One suspicion fell away along the way. You might assume part of that time goes on ad approval and setup — it does not: 92% of campaigns start showing in the same week they were launched. The whole ramp-up happens on the auction side. What a campaign lacks is not Google’s approval but the history the system learns from about when to show it.
We also checked the obvious shortcut: since an event usually gets a new ad group added to a campaign that already runs, maybe such a group inherits its momentum and goes faster. Measured on 327 ad groups — it does not. A group added to a campaign that has run for months needs exactly as long as a group starting with a brand-new campaign.
Finally we measured the same for a single keyword, and only the three numbers side by side show what this is really about:
| What ramps up | half ready in | nine in ten by |
|---|---|---|
| a campaign | 4 weeks | 11 weeks |
| an ad group | 4 weeks | 11 weeks |
| a single keyword | 4 weeks | 12 weeks |
All three ramp up the same way, because the smallest unit of learning is the keyword–query pair. A campaign and a group have no ramp-up of their own; their curve is the sum of the curves of the keywords inside. Which is why an account’s momentum cannot be lent to anyone.
For a keyword the curve is also the steepest: in the first week it delivers 30% of its eventual level, and reaches it only around the tenth.
Phrases with no demand do not hurt the click-through rate — so we do not clean them out
It seemed obvious: if phrases with a volume under ten searches a month get clicked less, then on an account close to the 5% threshold they are worth removing, and on a healthy account worth keeping. The direction held — on accounts below the threshold such phrases do get clicked less than the rest of the account, and on accounts with a 10–15% rate clearly more. Except there is nothing to remove.
On a typical account those phrases produce not one impression — the median of their share of impressions is zero. Removing 99,386 phrases with no demand from the whole sample moves the median account CTR by 0.007% and costs 8,114 clicks. That is a zero result at a real cost.
Cleaning out phrases with no demand is busy work: visible in the interface, invisible in the result. The click-through rate is watched on search terms, not on keywords.
What remains of it is a filter on the way in, not a clean-up on the way out: we do not propose phrases under ten searches and we do not count them towards the target — but not because they do harm, rather because they produce impressions seven times less often than popular ones. The ones already in the account stay.
The hypotheses we tested and that failed
The other half of this study’s value is here. Every item below sounds sensible and we have heard every one of them — from Google support or from industry guides. None of them translates into grant use, and every one costs time that should go to the four things that do work.
| Hypothesis | How we tested it | Why it fails |
|---|---|---|
| Rewriting ad headlines to match the phrase | 61,669 ads; pairs of ads running side by side in the same group — the same keywords, the same page, the same bid, differing only in the text | The difference in click-through rate between two such ads is 2.7% of the whole variance — the remaining 97.3% depends on which keywords and which page the ad group has. In low-traffic pairs there is still +1.7% CTR in favour of better phrase coverage, but in pairs above a thousand impressions, where the rate is measured precisely, the effect drops to zero. The reason: 90.5% of ads have the full 15 headlines and on their own cover 88.6% of their group’s keywords |
| Google’s ad quality rating (Ad Strength) | Ads across the whole sample, then 899 pairs from the same account with the same number of headlines and descriptions | “Poor” ads have a median of 11 headlines, all the others have 15 — it is a counter of filled-in fields. At the same number of headlines the difference in click-through rate is zero: 124 pairs to 107 on impressions and 19 na 17 on click-through rate, a coin flip. Fill it up to 15 and stop looking at it |
| Page speed (PageSpeed) | 40 tests on pairs of pages from the same account with almost identical content length (1.9% apart), differing in speed score by ≥15 points | After correcting for multiple testing none survives. A fast page is better for the user and worth taking care of — all we question is that it will raise campaign results |
| Domain authority (backlinks) | The link profile for every domain in the study, measured separately across the whole set and separately on accounts that actually show ads | Across the whole set the relationship looks strong, but among accounts that show ads it falls to zero. Dead accounts have a median of 45 referring domains, live ones 73 — that is a sign of how active the organization is, not a lever. The Google auction does not read links at all |
| Domain age and account age | Domain age counted from the first observed link (median 5.6 years), compared at the same account size | The same mechanism as with links: among accounts with traffic it falls to zero. A 5–10-year-old domain has three times more links and more keywords in the account — we are measuring how long the organization has existed, not an advantage in the auction |
| Six of the nine on-page SEO elements meta description · heading hierarchy · images and alt text · internal links · text-to-code ratio · title length | All nine in one model, always compared within the same account; recomputed on the pages of every account whose code we pulled | Three pass, and two of them say the same thing — whether the page names the phrase’s topic: the phrase in the title and the phrase in the content. The third, content length, works differently: not on whether a phrase starts but on the page rating Google gives. The other six come out in 39–57% of accounts, which is a coin flip — they help half the organizations and hurt the other half |
| Which conversions the campaign optimizes for | Accounts where someone actually switched the campaign to a narrower, more real conversion event — three months before against three after | Nothing changes: rises and falls split evenly on every measure, and the median stands still. Set a real event anyway — for the report and for the algorithm, but do not expect a jump in results. Conversions themselves cannot be skipped: without them there is no strategy that lifts the $2 cap |
| Daily budget | Breaking the lost impression share down into auctions lost and budget shortfall | Budget accounts for 1.2% — the rest is auctions lost. The exception: 16 accounts hitting the grant ceiling |
| Number of campaigns | The same 400 phrases spread over three campaigns or twelve | The same result. Ad group size is what matters: above 50 phrases a group stops working, because it leads to one page |
| Quality Score | An attempt to read it across the whole keyword sample | Google does not report one for 97.5% of keywords — only a handful of phrases that already have traffic get one. It is a consequence of having traffic, not the road to it. Do not set it as a goal and do not hold anyone to it |
What grant work gives you in organic search
Since ads and organic results use the same page, it is natural to ask whether work done for campaigns also leaves a mark in search. We checked it on the accounts for which we have both page measurements and organic positions.
| What we do to the pages for the campaigns | visibility in organic results | entering the top three |
|---|---|---|
| More promoted pages | a clear relationship very clear | no relationship |
| The phrase in the page title | the relationship is clear | no relationship |
| Longer content | the relationship is weaker, but visible | no relationship |
Work done for the campaigns is what makes an organization start appearing in organic results at all. The same content works in two channels at once.
What it does not do: it does not get you into the top three. Which makes sense — the top places are decided by domain authority and the links pointing to it from outside, things that writing content does not change. Content opens the door to the ranking; the podium is played for with something else.
We also measured how much of that authority organizations actually have: the median is 61 referring domains, and only 4% have fewer than ten. That is not much next to commercial sites, but it is not zero, so there is material to work with on organic results.
And does the ad not take traffic away from organic results?
There is nothing to take away. We checked the organic positions of the domains in the study: the median number of ranking phrases is 7, and the median number in the top three — zero. Two thirds of organizations have not one phrase in the top 3. Overlaying paid phrases on organic ones gives a median coverage of 0.4%, and 0.0% in the top three.
The ad is practically the only route to almost every phrase the organizations advertise on. There is no cannibalization here, because there is nothing to cannibalize.





