POST: Mission Reports
ETA: 10 min
PUBLISHED: 18/11/2026

AI Overviews, AI Mode and ChatGPT search: what 2026 data says about SEO

Verified 2026 data on how AI Overviews, AI Mode and ChatGPT search reshape organic clicks, plus a practical playbook for getting cited instead of summarised away.

// SHARE_THIS_REPORT
01 / Context

The click is no longer the default outcome of a search

For two decades the deal was simple: rank well, get the click. In 2026 that deal has a large asterisk. Google's AI Overviews sit above the organic results for a big share of informational queries, AI Mode offers a full conversational results page, and ChatGPT has become a search habit for hundreds of millions of people. OpenAI said in February 2026 that ChatGPT had reached 900 million weekly active users.

None of this means organic search is dead. Google is still, by a wide margin, the largest source of referral traffic on the open web. What has changed is the shape of demand: more questions are answered on the results page, fewer visits are "free", and the visits that do arrive tend to come later in the decision.

What this article covers

  • What independent studies and Google itself have published about clicks, from 2025 to November 2026.
  • What Google's own May 2026 guidance says you should and should not do for generative AI features.
  • How ChatGPT search decides which sites it can cite, and what you control.
  • A practical playbook: content, entities, structured data, crawl access and measurement, plus an honest look at llms.txt.

The aim is not to sell another acronym. GEO, AEO and "AI SEO" are useful labels for a set of habits, but as you will see, most of those habits are good SEO and good content strategy done with more discipline than before.

02 / The data

What the verified numbers say about lost clicks

The most cited dataset is Ahrefs'. Its first study, published in April 2025, found that the presence of an AI Overview correlated with a 34.5% lower click-through rate for the top-ranking page. In February 2026 Ahrefs re-ran the analysis on 300,000 keywords, comparing December 2023 with December 2025, and the figure rose to 58% (Ahrefs, 2026 update). The study also found lower CTR across the whole first page, not just at position one. Ahrefs is clear that this is a correlation, not proof of causation.

Pew Research Center, using the browsing data of 900 US adults in March 2025, found that people clicked a traditional result in 8% of visits when an AI summary appeared, against 15% when it did not. Clicks on the links cited inside the summary happened in about 1% of visits. Google disputed the methodology, saying the query set was not representative.

Seer Interactive's 2026 update, covering 53 brands and 5.47 million queries, shows CTR on AI Overview queries bottoming out in late 2025 and ticking up in early 2026. Seer itself warns against forecasting a recovery from two months of data. It also reports that brands cited in the overview get noticeably more clicks than uncited brands on the same queries, though still fewer than on queries with no overview at all.

How to read this

The studies disagree on magnitude, use different methods and mostly measure US desktop data. They agree on direction: on informational queries, an AI Overview takes clicks, and being cited inside it is the best available consolation. Budget and forecast accordingly, and stop treating last year's top-of-funnel traffic as a baseline.

03 / What Google says

Google's May 2026 guide: AI search is still search

On 15 May 2026 Google published "Optimizing your website for generative AI features on Google Search", its first dedicated guide to AI Overviews and AI Mode. Its central message is blunt: optimising for generative AI search is optimising for search, and therefore still SEO. The features rely on Google's core ranking and quality systems, and a page has to be indexed and eligible for a snippet to appear.

The guide is just as useful for what it tells you to ignore:

  • No special files. Google says you do not need new machine-readable files, AI text files, markup or Markdown versions to appear in Google Search.
  • No chunking. There is no requirement to break content into tiny pieces, and there is no ideal page length.
  • No AI-specific writing style. The systems understand synonyms and meaning, so chasing every long-tail variant is pointless.
  • No inauthentic mentions. Manufacturing "mentions" across the web is, in Google's words, not as helpful as it might seem.
  • Structured data is not required for generative AI features, though it remains useful for rich results.

Google also addresses query fan-out, the technique where the model issues many related sub-queries to assemble an answer. Creating pages for every fan-out variation primarily to manipulate AI responses, it warns, falls under its scaled content abuse policy.

What it does recommend is unglamorous: original, non-commodity content based on real expertise, crawlable and fast pages, sound JavaScript rendering, supporting images and video, and accurate Merchant Center and Business Profile data. If your site's engineering blocks any of that, no amount of "GEO" will compensate.

05 / Playbook

How to earn citations: content, entities and structured data

Strip away the marketing and the pages that get cited share a few traits. They answer a specific question early and plainly, they contain something the model cannot get elsewhere, and they come from a source that is easy to identify and trust.

Content that cannot be commoditised

  • Lead with the answer. State the conclusion in the first paragraph of a section, then support it. This is good writing, not a trick.
  • Add information gain. Original data, prices, test results, screenshots, worked examples and first-hand experience are what an AI answer needs to cite. Rewritten consensus is not.
  • Date and maintain. Show when a page was last reviewed and actually review it. Stale numbers are a reason to cite a competitor.

Entities and consistency

Models and search systems work with entities: your organisation, your people, your products. Make them unambiguous. Use the same name, description and key facts across your site, Google Business Profile, Merchant Center, LinkedIn and industry directories. Give authors real bio pages with verifiable credentials. Link related topics so your expertise reads as a coherent cluster, which is the logic behind a signal architecture map.

Structured data, used for what it does

Google says no special schema is needed for AI features, and you should take that at face value. Organization, Person, Product, Article and LocalBusiness markup still help Google understand who you are and keep rich results eligible. The rule Google repeats is the one that matters: markup must match the visible content. Schema that claims things the page does not show is a liability, not an AI hack.

06 / Reality check

llms.txt and other shortcuts: what the evidence supports

llms.txt is a proposed convention: a Markdown file at the root of a site that summarises its content for language models. It is cheap to create, which is why so many plugins now generate one automatically. The question is whether anyone reads it.

For Google Search the answer is documented. Google's May 2026 guide states that you do not need AI text files or Markdown versions of pages to appear in its generative AI features, and that Google Search ignores them. Google's John Mueller had earlier compared llms.txt to the keywords meta tag. The picture is less tidy elsewhere in Google: Chrome's developer documentation describes llms.txt as an emerging convention, and an experimental Lighthouse audit for agentic browsing checks for it. That is about browser agents, not search rankings.

No major assistant has publicly confirmed that it uses llms.txt for retrieval or ranking. Our recommendation: if you already have one, keep it accurate; if you do not, it is a low-priority, low-cost task for documentation-heavy products such as APIs or developer tools. Do not expect it to change visibility in Google.

Other tactics to retire

  • Spinning out hundreds of near-duplicate pages to match fan-out queries. Google names this as scaled content abuse.
  • Buying or seeding "mentions" on low-quality sites to look authoritative.
  • Hidden text or schema aimed only at bots.
  • Trusting third-party "AI visibility scores" as ground truth. Google's guide notes that no third-party tool has access to its internal ranking or AI systems.
07 / Measurement

Measuring AI visibility without fooling yourself

Until mid-2026, AI Overview and AI Mode performance was folded into the standard "Web" search type in Search Console. On 3 June 2026 Google announced Search Generative AI performance reports, separate beta views for Search and Discover, initially for a subset of sites. Google's note on the post says that as of 31 August 2026 these insights had rolled out to all websites worldwide. The reports show impressions by page, country, device and date; Google's announcement does not describe click data, so plan around impressions.

The same June announcement introduced a Search Console control that lets site owners opt out of appearing in, and grounding responses for, Google's generative AI features. Google states that opted-out sites will not receive traffic or impressions from those features, and that the control is not used as a ranking signal elsewhere. For almost every commercial site, opting out costs more visibility than it protects.

A workable reporting setup

  • Track generative AI impressions per page group in Search Console, alongside classic clicks and CTR.
  • Segment GA4 referrals from chatgpt.com, perplexity.ai, copilot and gemini into an "AI assistants" channel.
  • Watch branded search and direct traffic: assistant exposure often shows up there rather than as a referral.
  • Run a fixed prompt set monthly for your key topics and log who gets cited, accepting that results vary by run.

AI search surfaces compared

The main surfaces behave differently, so the levers and the measurement differ too.

SurfaceHow sources are chosenWhat you controlHow to measure
Google AI OverviewsCore ranking and quality systems, with query fan-out across sub-topicsIndexability, snippet eligibility, content quality, nosnippet and max-snippetSearch Console generative AI report (impressions), Web search type
Google AI ModeSame systems, more fan-out, conversational follow-upsSame as AI Overviews, plus Merchant Center and Business Profile dataSearch Console generative AI report; clicks counted in Web traffic
ChatGPT searchLive retrieval via OAI-SearchBot and search partnersrobots.txt and WAF access for OAI-SearchBot, crawlable HTMLGA4 referrals from chatgpt.com, server logs, prompt sampling
Perplexity, Copilot, Gemini appOwn crawlers or Bing and Google indexesBot access, Bing Webmaster Tools, clear entitiesGA4 referrals, logs, prompt sampling

Based on Google Search Central and OpenAI documentation as of November 2026. Referral attribution for assistants is incomplete; expect some traffic to appear as direct.

A 30-day AI search checklist

Work through these in order; the early items unblock everything after them.

  • [+] Confirm Googlebot, Bingbot and OAI-SearchBot can reach key pages through robots.txt, CDN and WAF, and verify it in server logs.
  • [+] Make sure primary content is present in the rendered HTML and is not hidden behind client-side JavaScript or interaction.
  • [+] Check that priority pages are indexed and not restricted by nosnippet or a very low max-snippet value.
  • [+] Rewrite the opening of each key section so it answers the question directly in one or two sentences.
  • [+] Add one piece of original evidence per priority page: data, pricing, test results or a worked example.
  • [+] Align organisation, author and product facts across the site, Business Profile, Merchant Center and major profiles.
  • [+] Audit structured data so every property matches visible content; remove markup that does not.
  • [+] Create an AI assistants channel in GA4 and add generative AI impressions to your monthly SEO report.

AI Overviews and AI Mode SEO: common questions

[01]

Do I need special schema markup to appear in AI Overviews or AI Mode?

No. Google's May 2026 guide states that structured data is not required for generative AI search and that there is no special schema.org markup to add. Keep using Organization, Product, Article and similar types for rich results and entity clarity, and make sure every property matches what is visible on the page.

[02]

Should I create an llms.txt file?

It will not help in Google Search: Google says it ignores AI text files and Markdown copies for its generative features. No major assistant has confirmed using llms.txt for ranking. It is a cheap, optional extra for documentation-heavy products, but it should never come before crawl access, content quality or measurement.

[03]

How much organic traffic do AI Overviews really take?

It depends on the query mix. Ahrefs' February 2026 update found a 58% lower CTR for the top result on keywords with an AI Overview, and Pew found clicks nearly halved in its 2025 panel. Transactional and navigational queries are affected less. Measure your own page groups rather than applying an industry average.

[04]

Can I opt out of Google's AI features without hurting rankings?

Since mid-2026 Search Console offers a control to opt out of generative AI features. Google says it is not a ranking signal for regular results, but opted-out sites receive no traffic or impressions from AI Overviews, AI Mode or AI features in Discover. For most commercial sites that is a poor trade.

[05]

How do I get my site cited in ChatGPT search?

Start with access: allow OAI-SearchBot in robots.txt and make sure your CDN or firewall does not challenge it. Then focus on clear, answer-first pages with original evidence and a consistent brand entity across the web. There is no paid or technical shortcut that OpenAI documents for inclusion.

Find out what AI search actually sees on your site

We audit crawl access, rendering, entities and content against Google's and OpenAI's documentation, then give you a prioritised fix list.

// EDITORIAL_ROUTE

AI Overviews & AI Mode SEO in 2026: data and playbook | Git-Force