Guide · AI search

How much control do we have when Google changes how search results look?

Very little over the change itself, and some over how exposed you are to it. When Google made AI Overviews the default in the US, search visits to English Wikipedia fell about 5% relative to the same articles in other languages, with nothing changing on Wikipedia’s side. Your control lies in how much of your business depends on one layout, and how quickly you notice when it moves.

The short version

  1. Google’s switch to AI Overviews by default cut English Wikipedia’s monthly search traffic by 5.45% relative to German and 4.82% relative to French, about 100.27 million visits a month (Khosravi and Yoganarasimhan (opens in a new tab)).
  2. Google’s AI layer shifts week to week: re-searching 96 keywords two days later, 16 changed whether they showed an AI Overview at all (our AI Overview frequency study).
  3. Google’s own rankings move too: measured against results from two days later, ChatGPT’s match with Google’s top 10 fell from 8.3% to 5.2% (our AI citations and Google rankings study).
  4. Each new Google surface rewards pages differently: AI Overviews cited 29.7% of top-10 pages and AI Mode 16.3% on the same searches (our AI Mode vs AI Overviews study).

Can a business control how Google designs its results?

No: Google sets the layout, and publishers can only respond to it. Khosravi and Yoganarasimhan (opens in a new tab), researchers at the University of Washington, put it plainly: search platforms “can reallocate traffic through product-design changes that publishers cannot directly control.”

That is a policy argument rather than a measured result, but the stakes behind it are measured. They cite Google’s share of the global search market at approximately 90%. When one company controls the front door for that much demand, a design decision on its side becomes a traffic event on yours.

The timeline shows how fast this moves. Google made AI Overviews, the AI summary at the top of its results, part of the default US experience in May 2024. By early 2025 the feature had reached more than 100 countries.

How much traffic can one design change move?

Enough to matter even for a site that AI summaries cite often. The researchers compared English Wikipedia articles with the same articles in German and French. Readers of those editions were mostly in Europe, where AI Overviews were not yet the default.

After the US rollout, monthly search traffic to English articles fell 5.45% relative to German and 4.82% relative to French. The authors estimate that as about 100.27 million fewer search visits to English Wikipedia per month. A third comparison, with Japanese, pointed the same way at 16.53%, over a shorter window.

The striking part is that Wikipedia is among the sources AI summaries cite most. Being cited did not offset the loss. And these figures likely understate the effect for searches that actually showed a summary, because only about 40% of English Wikipedia traffic comes from the US. Our guide on AI Overview citations and lost clicks weighs the wider evidence.

How often does Google’s AI layer change what it shows?

Often enough that a single check tells you little. In our study of US searches, we re-ran 96 keywords two days apart. 83.3% had the same outcome, but 16 flipped between showing an AI Overview and not.

The sources shifted more than the outcome. Where both days showed an AI Overview, fewer than half the cited pages were the same: the typical overlap was 0.43 on a scale where 1 means identical.

Google’s regular rankings moved as well. Between 26 and 28 September, Google’s top 10 for the same 80 questions changed substantially (our AI citations and Google rankings study). Measured against the later results, ChatGPT’s match with Google’s top 10 fell from 8.3% to 5.2%.

Does each new Google surface reward the same pages?

No, which means each new layout can reshuffle who gets seen. In our comparison of AI Mode and AI Overviews on the same 400 searches, AI Overviews cited 29.7% of the top-10 pages and AI Mode 16.3%.

AI Mode also leans on Google’s own pages in some searches. On searches with no AI Overview, 69.3% of AI Mode’s citations went to Google pages, mostly listing pages for local businesses and products. That is a design choice on Google’s side, and it leaves less room for outside sites in those answers. We cover that scenario in our article on AI Mode as Google’s default.

AI answers are also sensitive to small changes on the user’s side. In a small test by Wen and colleagues (opens in a new tab), rewording questions by only 13% on average changed which domains Gemini cited for every one of 30 question pairs.

What part of the risk can you actually manage?

You can manage your dependence on any one layout and how fast you react. Ranking still helps inside AI Overviews: the first organic result was cited in 49.5% of AI Overviews, the ninth in 15.5%. So strong rankings remain a hedge, just not a guarantee.

Spreading demand across channels is the other lever. On one site studied by Watanabe and Nakayashiki (opens in a new tab), Google clicks fell about 20% site-wide while ChatGPT referrals grew. Even pages nobody optimized grew 3.5 times in ChatGPT referrals over about five months. Our guide on whether ChatGPT citations lift Google rankings covers what that study means for Google traffic.

Position inside AI answers is partly outside your hands too. In a simulated hotel test, being placed higher in the list an AI assistant saw had the same pull as a price cut of about $11.7 per night (Baig and colleagues (opens in a new tab)). The authors advise “continuous measurement rather than a one-time fix.”

What should you do about it?

Treat Google’s layout as a risk you monitor and hedge, not a fixed condition. In practice:

  1. Map your exposure. List which revenue pages depend on informational searches, the kind AI summaries answer directly.
  2. Monitor weekly, not once. Track a fixed set of searches for whether an AI Overview appears and which pages it cites.
  3. Compare against a baseline. Judge traffic drops against similar pages or sections, so you can tell a Google change from your own mistake.
  4. Keep rankings strong. Top positions are still cited far more often inside AI Overviews.
  5. Build other routes to demand: AI assistants, email, direct and partner channels, so one layout change cannot cut off most of it.
  6. Plan for scenarios. Model what a further drop in search visits, at least the size Wikipedia saw, would do to your pipeline.

If you want a monitoring setup built around this, see our generative engine optimization service.

What does the research not tell us yet?

The research measures past changes, not the next one, so forecasts remain guesses.

  • The traffic evidence comes from Wikipedia, a non-commercial reference site. Commercial publishers may lose more or less.
  • The Wikipedia study cannot see visits lost inside the site after a search visit disappears, so the total effect could be larger.
  • Our frequency and ranking checks cover two dates in September 2026 and US searches only.
  • The rewording and hotel findings come from small or simulated tests.
  • No study yet measures how much diversifying channels actually protects revenue when Google changes its layout.

Frequently asked questions

Did Google AI Overviews reduce website traffic?

For English Wikipedia, yes. Its monthly search traffic fell about 5.45% relative to the German edition after AI Overviews became the US default.

Can we opt out of Google’s AI Overviews without losing rankings?

The research reviewed here does not answer that. It shows only that being cited, as Wikipedia often is, did not prevent a traffic decline.

How often do AI Overviews change?

Within days. In our check, 16 of 96 keywords changed whether they showed an AI Overview two days later, and the cited pages changed more.

Should we still invest in SEO if Google keeps changing the layout?

Yes, as one hedge among several. The first organic result was cited in 49.5% of AI Overviews in our study, against 15.5% for the ninth.

Sources

Free strategy call

Some questions are easier to answer about your own business.

Bring the one that matters most. On a free 30-minute call we’ll take a first look at it and send you a short written read afterward.