Google lets you opt out of its AI. On our panel, that switch would turn off 53% of mentions
There is no general answer. There is one per brand, and it ranges from 34% to 78%.
Since 31 August 2026 Google offers a single-checkbox control to stop appearing in AI Overviews, AI Mode and generative AI in Discover. Across 17,710 measured answers from a closed panel of 11 brands, Google AI Mode accounts for 53% of every time an AI search engine names the brand (95% CI 50.9–55). Per brand the split ranges from 34% to 78%: for 7 of the 11, flipping that switch would turn off more than half of their AI presence.
- Sample
- 17,710 measured answers over a closed panel of 11 brands
- Period
- 3 August – 7 September 2026
- Method
- Panel closed on both axes: only brands measured every week of the window and across all three engines. Every answer where the brand appears is counted and split by engine. Observations per engine are balanced within 10% (5,840 / 5,939 / 5,931), so the split does not measure our own measurement schedule.
- Language
- Spanish (Spain)
What changed out there
On 31 August 2026 Google confirmed that the Search generative AI control — in testing in the UK since 3 June — was rolling out to every site. With one checkbox, a site stops appearing in AI Overviews, AI Mode and generative AI in Discover. Google states two things: that opting out means no traffic and no impressions from those surfaces, and that the setting is not used as a ranking or inclusion signal elsewhere in Search.
One checkbox, whole consequences. The question it leaves on the table — "how much do I lose if I tick it?" — is not one Google answers, because its generative AI performance report gives impressions but no clicks and no queries. We cannot answer it in traffic either. We can answer it in mentions, which is what we measure.
How much Google AI Mode weighs in a brand's presence
The panel is 11 brands measured every week and across all three engines, with the same questions. Observations are split almost evenly — 5,840 on Google AI Mode, 5,939 on ChatGPT, 5,931 on Perplexity — so the split below is not measuring how often we happen to run each engine.
| Brand appearances over 6 weeks | How many | Share |
|---|---|---|
| Google AI Mode | 1,211 | 53.0% (95% CI 50.9–55) |
| Perplexity + ChatGPT | 1,076 | 47.0% |
| Total | 2,287 | — |
And the split per brand, most dependent first:
78% · 71% · 66% · 65% · 63% · 61% · 59% · 49% · 41% · 41% · 34%
That spread is the finding. There is no such thing as "what you lose by opting out": there is what each brand loses, and it goes from a third to four fifths. A company whose 34% of mentions come from AI Mode and one whose 78% come from there are making different decisions even if they tick the same box.
The counterweight: AI Mode names more brands
Before reading 53% as an achievement, look at this, because it explains part of the number:
| Engine | Sector brands named per answer |
|---|---|
| Google AI Mode | 1.10 |
| Perplexity | 0.66 |
| ChatGPT | 0.41 |
Google AI Mode names nearly three times as many brands per answer as ChatGPT. Part of that 53% share is therefore a property of the engine, not a merit of the brand: in a shop window that fits more names, more names fit. What does not change is the practical consequence, which is what is at stake when it is switched off.
And the window was closing on its own
The same panel, week by week, same questions and same method:
| Week | AI Mode answers naming the panel | Citations per answer |
|---|---|---|
| W32 (3 Aug) | 22.3% | 13.07 |
| W33 | 22.8% | 12.82 |
| W34 | 23.4% | 12.22 |
| W35 | 23.5% | 9.94 |
| W36 | 15.9% | 4.08 |
| W37 (7 Sep) | 16.7% | 3.98 |
In six weeks Google AI Mode went from 13.07 citations per answer to 3.98 — 70% fewer — and from naming the panel in 22.3% of its answers to 16.7%. Over the same window, ChatGPT and Perplexity together moved from 11.1% to 9.0%: flat within the noise. The citation collapse is measured separately, and the provider that transports those queries confirmed it on its own sample of roughly 2,378 paired daily questions.
Plainly, and without insinuating anything about anyone's intentions: the switch arrives weeks after the surface it turns off shrank on its own. Whoever decides today is deciding about a smaller shop window than July's — and that cuts both ways, because there is less to turn off, but what remains is still half of all mentions.
How to use this
The useful question is not "should I opt out?". It is "what percentage of the times an AI search engine names me comes from Google?", and that is a number of your own, not a sector statistic. You get it by measuring the same questions across several engines over several weeks and comparing. If it comes out at 34%, the checkbox is an editorial decision. If it comes out at 78%, it is a business one.
What this finding does NOT prove
- It does not measure how many sites have opted out. Google does not publish it and we have no way to detect it from outside: what is measured here is what is at stake, not what has happened.
- It does not measure traffic. We count mentions in answers, not visits. A brand can be named often and get few clicks — Google’s own report for this surface gives impressions but no clicks, so nobody has that figure in full.
- Part of Google AI Mode’s share is a property of the engine: it names 1.10 brands per answer against ChatGPT’s 0.41. More names fit there, so appearing more does not mean ranking better.
- Eleven brands is a small sample of brands, even at 17,710 answers. What is robust is the spread — that dependence varies enormously from one brand to another — not the exact magnitude for any of them.
- The panel is Spanish and drawn from specific sectors. In a market where ChatGPT weighs more, the split will be different.
- AI Overviews is not measured. This figure is AI Mode only, and the switch turns both off at once, so what is actually lost is equal to or greater than what appears here.
Frequently asked questions
So should I opt out of Google’s AI?
It depends on a number that is yours, not the sector’s: what percentage of your mentions in AI search engines comes from Google AI Mode. On this panel it ranges from 34% to 78%. At 34% the decision is editorial; at 78% it is a business decision. Anyone giving you a general recommendation without that number is guessing.
Does the switch penalise me in ordinary Google Search?
Google states the setting is not used as a ranking or inclusion signal elsewhere in Search. That is their statement, not a measurement of ours: we have not verified it and have no way to.
Does opting out of Google remove me from ChatGPT or Perplexity?
No: it is a Google control and only affects Google surfaces. Nor does it compensate, because engines do not read the same things — in our corpus, two engines answering the same question share between 4% and 13% of their sources.
How do I measure my own split by engine?
With the same questions sent to several engines over several weeks, counting in how many answers your brand appears and splitting that by engine. What matters is that the number of questions per engine is balanced: otherwise the split measures your measurement schedule, not your visibility.