MEMO · MAP · AUGUST 2026

Where Google’s AI can’t reach

9 min read · every claim sourced below

The short version: AI Overviews cut clicks roughly in half on the searches they appear on. They do not appear evenly. They almost never appear on a search for someone’s name, just 0.9% of navigational queries. Every other blind spot on this map points the same way, which gives the whole thing one instruction: build a name people look for on purpose. Getting cited by the AI does not bring you clicks either, but it does build that name, so the two halves of this memo are the same job.

The doom posts are directionally right and useless. Nobody tells you which searches you lost, and the loss is nowhere near evenly spread. Below is the map, measured, plus the honest part about which side of it you are standing on.

First, the size of it

Pew Research followed 68,879 real Google searches from 900 US adults through March 2025. It is the cleanest number in this whole subject because it watched actual browsing rather than modelling it.

0%
clicked a website when an AI answer was on top
0%
clicked when there was no AI answer
0%
clicked a link inside the AI answer itself

Hold that 1% in mind, because it decides the second half of this memo. Being named inside an AI Overview is not a way to get visitors. It is worth something else entirely, and we will get to what.

Two numbers you will see everywhere are not in this memo: “up to 30% CTR collapse” and “60% drops overnight”. We could not trace either to a named study. The irony is that the real measured fall is larger than the first one: 15% to 8% is a 47% relative drop.

The map: which searches actually get an AI answer

This is the part nobody puts in a video. Ahrefs analysed 146.1 million desktop search results pages in September 2025 and measured how often an AI Overview appeared, split by what kind of search it was.

Query typeGets an AI Overview
Navigational (someone searching your name)0.9%
Transactional (about to buy or do)2.1%
Commercial (comparing before buying)4.3%
Very newsy / fresh6.3%
Local (overall)7.9%
Single-word queries9.5%
Branded13.1%
Non-branded24.9%
Informationalabout 36% the bullseye
Comparisons (X vs Y)95.4% the bullseye

Informational is Seer’s figure from roughly 50,000 queries; everything else is the Ahrefs set. Comparison queries are the single most exposed shape on the internet at 95.4%, which is worth knowing before you write another “X vs Y” post.

The one number the whole map turns on

0.9%. When somebody searches your actual name, Google almost never puts an AI answer in front of it.

That is why podcasts, newsletters and communities keep working, and it has nothing to do with them being more authentic or higher quality. They work because they manufacture people who search for you, on purpose, by name. The blind spot is not a format. It is being someone people look for.

The local contradiction, worked out in the open

One study says AI Overviews appear on 68% of local searches. The table above says 7.9%. Both are real, and the disagreement is more useful than either number alone.

It resolves as intent inside local rather than local itself. A simple transactional local query like “tacos San Francisco” leans on the local pack and rarely gets an AI answer. An informational one like “how long does an eye exam take near me” gets one about 92% of the time.

So the local blind spot is transactional-local only. If you are a local business publishing helpful explainer content about your field, you are not in the blind spot at all. You are in the bullseye with everybody else.

The part you will not enjoy

Read the map again and notice its shape. Almost every blind spot is either bottom-of-funnel or name-driven. Which means: if your business is informational how-to content, your entire category is the target.

There is no arrangement of keywords that fixes that, and any article telling you to “just start a podcast” is selling you a feeling. The name-search route below is a genuine response. It is not a rescue, and it is slower than what you lost.

The other half: being the source the AI quotes

If you cannot avoid the AI answer, the next question is how to be the site inside it. This is the highest-demand question in the entire subject, and it is also where the advice is worst.

Start by distrusting the headline statistic

Nearly every guide to AI Overview citations rests on one number: how much the citations overlap with the top 10 organic results. Here is that number, as reported:

Reported overlapSource
76%Ahrefs, July 2025
54.5%September 2025 (up from 32.3% in May 2024)
38%Ahrefs, current. They disclose part of the drop is improved parsing
~17%BrightEdge, February. Different methodology and dataset

A figure that moves between 17 and 76 depending on who counted cannot carry the strategy everybody has built on top of it. Ahrefs themselves are straight about it: part of their own drop from 76% to 38% is a parsing improvement, meaning they now detect citations they previously missed, not that the world changed underneath them.

We are not picking the convenient one. We are telling you the range, because anyone quoting a single figure here is quoting a measurement artefact as a law.

What follows from it anyway

One thing survives the disagreement: “just rank number one” is wrong. Ahrefs puts roughly 31% of citations at positions 11 to 100, and another 31% beyond position 100. Pages nobody would call ranking get quoted constantly.

What correlates instead is retrievability, meaning whether a machine reading your page can lift a complete answer out of it without the surrounding context:

  • Answer in the first paragraph. Completely, not as a tease. An answer buried under 400 words of preamble is an answer that cannot be extracted.
  • Headings that are whole questions, each followed by a self-contained answer that makes sense lifted out on its own.
  • Expertise signals at page level, not domain level. Author credentials, first-hand experience, original data, primary sources on the page itself.
  • Match what people actually ask, in their words, rather than a keyword target.

Treat that list as practitioner guidance rather than measurement. It is a well-argued account of what correlates, not a controlled result, and we would rather label it than dress it up.

What a citation is actually worth

Not clicks. The Pew figure at the top of this memo says links inside an AI answer get clicked 1% of the time, and we are not going to contradict our own opening section to sell you an optimisation.

The value is the lift on everything else. Seer measured brands appearing in AI Overviews at 35% higher organic click-through and 91% higher paid click-through than uncited competitors on the same page. Being named is recognition, and recognition is what eventually makes somebody search your name, which lands you back at 0.9%.

So the two halves are one instruction

The defence: the deepest blind spot on the map is a name search, at 0.9%. The offence: being quoted does not send visitors, it builds recognition. Recognition produces name searches.

Build a name people look for on purpose. Everything above is either a way to do that or a measurement of what happens if you do not.

And the switch, since you will ask

Google shipped a control in June 2026 that pulls your site out of AI Overviews entirely. In Search Console, under Settings, there is a generative-AI setting you can set to Exclude. It covers AI Overviews, AI Mode and AI Overviews in Discover, it is property-level, and it is still rolling out region by region so it may not be in your account yet.

Google is clear that it does not affect normal Search ranking or inclusion. It also does not stop model training, which is a different control entirely.

We are not recommending it. Flipping it does not return one visitor. Google writes the same answer from the next site instead, so you lose the citation as well as the click, and by the section above the citation was the part that was actually worth something. It is here because you deserve to know it exists, not because it is the move.

Before you rely on any of this

  • These are trigger rates, not traffic guarantees. A query type that rarely gets an AI answer is not thereby a growth channel. It is just a place the AI is not standing in front of you.
  • The walls are moving. Semrush measured commercial-intent coverage growing 71% over six months. Today’s blind spot is not promised to be next year’s.
  • The trigger data is September 2025 desktop. Checked 15 August 2026. Mobile may differ and we have not seen a set this large for it.

Questions people actually ask

How much traffic do AI Overviews actually cost you?

Pew followed 68,879 real Google searches from 900 US adults in March 2025. On results pages with an AI summary, people clicked a traditional search result in 8% of visits. Without one, 15%. That is a 47% relative fall. Links inside the AI answer itself were clicked in just 1% of visits, which is why being cited is not a traffic strategy.

Which searches almost never get an AI Overview?

Navigational ones, at 0.9%. That is somebody typing your name. Transactional queries sit at 2.1%, commercial at 4.3%, very newsy queries at 6.3%, local at 7.9%, and single-word queries at 9.5%. Those figures come from Ahrefs' analysis of 146.1 million desktop SERPs in September 2025.

Which searches get hit hardest?

Informational ones, at around 36% per Seer's sample of roughly 50,000 queries, and comparison queries in the X-versus-Y shape at 95.4%. If your traffic is how-to and explainer content, you are the target. No map changes that, and this memo does not pretend otherwise.

I read that AI Overviews now answer most local searches. Which is right?

Both, and the disagreement is the useful part. Ahrefs puts local at 7.9% overall. Whitespark's local study puts AI Overviews on 68% of local searches. It resolves as intent inside local rather than local itself: a simple transactional query like 'tacos San Francisco' leans on the local pack, while an informational one like 'how long does an eye exam take near me' gets an AI answer about 92% of the time. The blind spot is transactional-local, not local.

Do I have to rank in the top 10 to get cited?

No, and this is where most advice is wrong. Ahrefs puts roughly 31% of citations at positions 11 to 100 and another 31% beyond position 100. Ranking is neither necessary nor sufficient. Be careful with the headline number here though: see the section on why that statistic is unstable.

Is being cited in an AI Overview worth anything if nobody clicks?

Yes, but not as traffic. Seer measured brands appearing in AI Overviews at 35% higher organic click-through and 91% higher paid click-through than uncited competitors. The value is recognition that lifts everything else on the page, not visits from the citation itself.

Should I use Google's opt-out and pull my site out of AI Overviews?

Probably not, and understand what it does first. Search Console has a generative-AI control that removes your site from AI Overviews, AI Mode and AI Overviews in Discover while leaving normal Search untouched. It does not return a single click. Google writes the same answer from the next site instead, so you lose the citation as well as the visit. It also does not stop model training, which is a separate control.

How long do these numbers stay true?

Not indefinitely. Semrush measured commercial-intent AI Overview coverage growing 71% over a six-month window. The trigger rates in this memo are September 2025 desktop data, checked 15 August 2026. Treat them as the current shape of the map, not a permanent one.

Want this built for you?

We write these memos because we build this stuff every day. If you want it working in your business instead of sitting on your reading list, that is literally our job.