Every agency in the country currently has an opinion about what AI is doing to search. Opinions are cheap. So we did the thing opinions are supposed to be made of: we pulled sixteen months of Google Search Console data, the period in which AI answers went from novelty to default, for every client property we could measure, and we looked.
This is what we found, including the parts that are complicated, because the complicated parts turn out to be the most useful.
How we measured, so you can judge it
The window runs from March 2025 to July 2026, sixteen months of daily search data per property, aggregated monthly. To measure change fairly we compared each property’s first three full months against its most recent three. Seven client properties had the full window; one more had only four months of history (a newer property, included for its level but excluded from trend claims), and one client’s property has no Search Console history at all because our relationship there is hosting and platform rather than search, so it is excluded and we are saying so. We also looked at two websites we rebuilt and then, deliberately, did nothing further with, because they answer a different question. No property that had data was left out. That matters: a dataset you curate after the fact is an opinion wearing a lab coat.
The context the window sits in
To read our numbers properly, hold the industry’s numbers beside them, because the sixteen months we measured were the roughest weather organic search has seen. Across the same period, SparkToro’s clickstream research found 68% of Google searches ending without a click to any website, up from 60% two years earlier. Pew Research Center tracked real users and found clicks nearly halving when an AI summary appeared, 8% against 15%. Seer Interactive measured organic click-through on AI-answered searches falling 61%. This is the backdrop every claim below stands against: not a calm sea, the storm itself.
US Google searches; excludes the Google app, so the true figure is likely higher
View the data behind this chart
| Measure | Value |
|---|---|
| A decade ago | ~45% (approximate) |
| 2024 | 60.45% |
| Jan to Apr 2026 | 68.01% |
The headline: rankings improved everywhere. Everywhere.
Across every property with the full sixteen months, average search position improved. Seven out of seven. Not one client lost ranking through the period in which AI Overviews rolled out across UK search, AI Mode launched, and Google shipped every algorithm change in between.
The improvements were not marginal. The smallest was 4.6 places; the largest was 18.5. One education platform moved from average positions in the mid-teens to sitting between 4.6 and 6.7 for the whole of 2026. A trades business we took on a year ago, a full engagement covering the website, the copy, conversion and local search, saw its average position nearly halve, from 37 to 19, in the engagement’s first year, with impressions up 22%.
seven out of seven improved
View the data behind this chart
| Measure | Value |
|---|---|
| Property A | +4.6 places (17.6 to 13.0) |
| Property B | +5.3 places (12.9 to 7.6) |
| Property C | +8.4 places (14.2 to 5.9) |
| Property D | +8.7 places (32.5 to 23.8) |
| Property E | +9.2 places (16.8 to 7.5) |
| Property F | +18.5 places (37.7 to 19.2) |
And one result we want to show you precisely because we were not working on it: a client whose active SEO engagement ended more than a year ago improved its average position by another nine places anyway. The foundations kept compounding with nobody touching them. We think that is what foundations are for.
The honest layer: rankings and clicks are no longer the same story
If we stopped there, this would be a brochure. Here is the rest.
Clicks rose on every property we actively manage. But on one large account, a specialist distributor with a deep library of genuinely useful technical resources, something else happened: impressions rose 41%, average position improved five places, and clicks fell 21%.
sixteen months, one account: the whole AI story in one chart
View the data behind this chart
| Measure | Value |
|---|---|
| Impressions | +41% |
| Average position | improved 5 places |
| Clicks | -21% |
Read that again, because it is the whole AI story in one account. The site became more visible and better ranked, and received fewer clicks. Where did they go? Into the answers. This client’s strongest pages answer exactly the kind of practical, informational questions that AI Overviews now answer directly at the top of the page, citing sources that many people no longer need to visit. The visibility is real, the authority is real, the influence on buyers is real, and a growing share of it no longer arrives as a click.
This is the shift the industry calls zero-click search, and we can now show it to you in our own client data rather than quoting someone else’s study, though the studies agree: this account is a single-business replica of the pattern Pew and SparkToro measured across the whole web. It is also, plainly, why AI-search visibility is now part of the job: if a machine is going to answer your buyer’s question, the difference between being its cited source and being invisible is the difference between being in the conversation and not existing. Seer’s research puts a number on that difference: cited brands earn 35% more of the clicks that remain. Rankings still matter enormously, our seven-for-seven says so, but the definition of winning has widened.
The experiment we did not mean to run: what a build does on its own
Two of the websites in our data answer a question clients ask us constantly: does the work keep working if you stop?
Both were rebuilds where the search fundamentals were built in, on-page optimisation as part of the build itself, and then no ongoing work of any kind was done. No retainer, no content programme, nobody tending anything. Six months against the previous six: the first grew its impressions by 140%, appearing for more than five hundred search terms it had never appeared for before, with its homepage moving from page four to page two. The second multiplied its search footprint seventeen-fold, with 87% of the search terms it now appears for being entirely new.
We are not telling you those sites do not need ongoing work; their clicks are still catching up with their visibility, and that catch-up is exactly what ongoing work accelerates. We are telling you the build itself did measurable search work with nobody touching it, because a website built properly is not a picture, it is infrastructure. The full rebuild-only story is here.
What a business owner should take from this
Three things, plainly.
First, the AI era has not broken good search work. Every property we measured ranks better than it did sixteen months ago, through every change Google shipped. The clear, useful, well-structured websites Google’s own guidance now explicitly rewards are precisely what held and grew.
Second, stop measuring only clicks. Visibility now works through answers as well as visits, and a business that only counts sessions will conclude it is shrinking while its influence grows. Watch your rankings, your impressions, and increasingly, who the AI answers name.
Third, foundations compound. The account nobody touched improved. The rebuilds left alone grew. The work is not a subscription to a result; it is an asset that keeps earning, which is the only kind of marketing spend worth defending in a board meeting.
See your own sixteen months
If you want to know how your own site has really fared through the same sixteen months, our free AI-search visibility check will show you where you stand in ordinary search and in the AI answers, reviewed by a person, with you within a working day.
Questions answered
Why should we trust an agency's own dataset?
Judge it by the method, which is why we stated it in full: every measurable property included, the exclusions named with reasons, the windows fixed before the results were read, and the inconvenient findings published alongside the flattering ones. A dataset that shows its clicks falling on one account is not a brochure. Check our numbers against the independent studies we cite; they tell the same story.
Does seven for seven mean SEO guarantees rankings?
No, and we would never claim it does; nobody controls Google. What the seven-for-seven shows is narrower and more useful: through the most disruptive period in modern search, properties built on clear structure, genuine content and real authority did not just survive the changes, they improved through them. The foundations are the resilience; the guarantee does not exist and never did.
How do we run this measurement on our own site?
Open Search Console, take your first three full months against your latest three across the longest window you have, and read positions, impressions and clicks side by side for your commercial terms. Our guide to judging SEO honestly walks the whole method, including the traps. Or ask us to run it: the free check below does exactly this.
Method note: Google Search Console, Web search type, 3 March 2025 to 2 July 2026, monthly aggregation, first three full months vs last three. Seven client properties measured over the full window; one newer property (four months of history) included for level only; one client untracked (hosting and platform relationship) and stated. The rebuild comparison uses matched six-month periods. All figures from our own client properties, shared with permission. Industry context: SparkToro/Similarweb (June 2026); Pew Research Center (July 2025); Seer Interactive (September 2025 update).
