Short answer: if your competitors are getting recommended by AI while you still rank well on Google, it is because ranking and being named in an AI answer are decided by different processes, so a page-one ranking does not carry over on its own. An AI answer names only a handful of brands, and the pages it cites often do not rank in Google's top 10 at all; many sit on review sites, marketplaces and comparison pages. To get into those answers, make sure AI agents can fetch your pages, write facts that still make sense when lifted out of context, get the same facts confirmed on independent sites, and measure the answers themselves with questions that do not contain your brand name.

None of this means SEO stopped mattering. It means a second shelf has appeared next to the results page, and it is much shorter. Small businesses already feel it: in WordStream's 2026 survey of over 300 US small businesses, the most-cited GEO challenge was competitors' content appearing in AI summaries instead, at 24%, ahead of difficulty tracking AI referral traffic and declining organic traffic, both at 22%. Below is what has changed, why a good ranking does not guarantee a place on that shelf, and what to do this quarter.

What changed: the answer now arrives before the click

The queries people type just before they choose a product are exactly the ones that now get an AI-written answer first. In Seer Interactive's 2026 study of 30,842 tracked informational queries, comparison queries triggered a Google AI Overview 95.4% of the time (267 of 280), and question-format queries 85.9% of the time (1,214 of 1,413). "Which is better, X or Y" and "what is the best tool for Z" are the queries where an AI summary is now close to the default.

One buying question, two shelves"Which is better, X or Y?"Results page1 Ranked link2 Your page, ranking well3 Ranked link4 Ranked link5 Ranked linkThe reader scans the list and picks a link.AI answer, shown before the clickA short answer in prose, with roomfor only a few names:Competitor ACompetitor BCompetitor CCites review sites, marketplaces and comparison pages.Not one of the names? The buyer may not see you.
Figure 1. Comparison and question-format queries now get an AI-written answer before the results. The results page has room for ten links and lets the reader choose; the answer has room for a few names and chooses for them. Source: InsightWonder, based on the Seer Interactive study and our own benchmark cited in this article.

Clicks have not disappeared, and the picture is less dramatic than early commentary suggested. Across 53 brands and 5.47M tracked queries, Seer found organic click-through on queries with an AI Overview climbed from 1.3% in December 2025 to 2.4% in February 2026, while queries without one went from 2.8% in January 2025 to 3.8% in February 2026. What did shift is who gets the clicks that remain. Brands cited inside the AI Overview earned +120% more organic clicks per impression than when they were not cited, though that still trailed queries with no AI Overview by -38%. Seer is careful to say it cannot prove that citation causes the higher click-through; stronger brands may simply be more likely to be cited. Either way, the brand inside the answer is in a better position than the brand below it.

Outside Google the shelf is shorter still. Assistants such as ChatGPT, Perplexity, Gemini and Claude answer in prose, and prose has room for only a few names. In our own benchmark, one model named 3.5 providers per answer and another 2.1. If your competitor is one of those two or three names and you are not, a buyer who asked an assistant may never see you, however well you rank.

The clearest evidence comes from Google's own AI feature. Ahrefs analysed 863K keyword results pages and 4M AI Overview URLs and found that 37.9% of the URLs cited in AI Overviews also appeared in the first 10 result blocks. Of the rest, 31.2% came from positions 11 to 100 and 31.0% from beyond the top 100. Ahrefs compares this with roughly 76% in its July 2025 study, but also notes that it improved its parsing in between and that AI Overviews moved to Gemini 3 in January 2026, so the two figures are not a clean before-and-after. Its own reading is that AI Overviews now draw more on sources found through query fan-out than on the original results page. Google describes the same technique for AI Mode: it issues multiple related searches across subtopics and data sources at once, so the pages an answer draws on can come from searches you never ranked for.

Query fan-out: the answer can cite pages from searches you never ranked forBuyer's question"best tool for Z"RELATED SEARCHES ACROSS SUBTOPICS (ILLUSTRATIVE)Z tools: pricingAlternatives to the leaderReviews and complaintsFit for a specific use casePages the answer citesFound through any ofthe searches, often fardown the original resultsYour page-one rankingCovers the query as typed,one search among severalGoogle describes AI Mode issuing multiple related searches across subtopics and data sources at once.
Figure 2. A strong ranking for the typed query covers one search out of several, which is why the pages an AI answer cites are often not the pages that rank first. Ahrefs' reading of its AI Overview data points the same way. Source: Google's description of AI Mode query fan-out; InsightWonder illustration.

Three mechanisms explain why a strong ranking does not translate directly:

  • Models want corroboration. A model needs to see the same fact in independent places before it states it confidently. A claim that exists only on your own domain is the weakest evidence you can offer, however well it ranks.
  • Third parties take the citations. In the product categories we measured, a large share of citations went to marketplaces, review sites and publisher round-ups rather than to brand sites. A listing page that repeats your facts in a neutral voice can win the citation your product page deserved.
  • Assistants lift facts, not adjectives. They quote sentences that answer a constraint: a dimension, a material, a return window, a compatibility. Copy written for brand tone gives them nothing to lift.

How hard it is to break in depends on your category. In our Q3 2026 benchmark, three product categories already had a clear default answer: project management software at 38% of answers, skincare at 35% and electric bikes at 28%. In every local-service category we measured, the leading brand appeared in fewer than 13%. In a consolidated category you are displacing an incumbent; in an open one, the shelf is still being filled.

SEO and GEO, side by side

The two disciplines overlap more than the debate suggests, because the same clear, crawlable, well-structured page serves both. What differs is the target and how success is counted.

Traditional SEOGEO
What it optimises forA position on a results pageBeing named or cited inside an AI-written answer
Unit of contentA page targeting a keywordA fact, comparison or answer a model can quote and attribute
Success signalRankings, impressions, clicksMention rate, citations and share of voice in answers, per engine
Where content has to existMostly on your own domainYour domain plus the independent sites answers draw on
What earns trustLinks and authorityThe same facts confirmed across independent sources

For a longer treatment of where the two diverge and where they are the same job, see GEO vs SEO.

What to do about it

1. Make sure the agents that fetch for answers can reach you

Each AI vendor runs several agents with different jobs, and blocking the wrong one is the most common way sites remove themselves from answers. OpenAI's crawler documentation separates GPTBot, which gathers training content, from OAI-SearchBot, which builds the index behind ChatGPT search. Blocking GPTBot does not remove you from ChatGPT search; blocking OAI-SearchBot does. On Google, Google-Extended is a usage control rather than a crawler: disallowing it removes you from grounded Gemini answers while, according to Google's crawler documentation, leaving search inclusion and ranking unaffected.

Test with the real user agents rather than reading robots.txt. A permissive robots.txt in front of a firewall rule that returns 403 to those agents is a block in practice. Also check pages with JavaScript disabled: most AI crawlers do not execute JavaScript, so content that appears only after client-side rendering is an empty page to them. Our guide to AI crawler access walks through each check.

2. Write facts that can be lifted out of context

Put the answer first and make the first two sentences under each heading self-contained. State the specifics buyers filter on, such as size, material, compatibility, price logic, return window and who the product is not for, as sentences rather than only inside a spec table. Write comparison and FAQ content around the questions buyers actually ask, which carry a situation, such as a budget, a constraint or a use case, rather than a bare category term. Extractable writing covers the patterns that survive being quoted.

3. Get the same facts confirmed off your own domain

Find the marketplaces, review sites and comparison pages that answers in your category cite, and check that what they say about you is current: prices, availability, features, markets. Use one name everywhere. Spelling variants, a product line described as a separate company or a legal name that differs from your trading name can each appear as a separate brand and split your share. Third-party corroboration explains how to find and fix the sources that matter.

4. Measure the answers, not the rankings

Rank trackers cannot see any of this. Keep a fixed set of buyer questions, written as full sentences with a situation in them, and keep your brand name out of them: a question that names you will be answered with your name, which proves nothing. When we got this wrong ourselves, 13 of 15 generated questions contained the brand name and the score read 13% when the honest value was zero. Report each engine separately, count only sources marked in the answer text, and keep "cited as a source" apart from "recommended". Building a prompt set shows how to write questions that do not flatter you.

You do not need a tool to find out whether a competitor is being recommended instead of you. You need a list of questions, a consistent log and an hour. Most of the value is in step three, where the results sort themselves into three very different kinds of work.

  1. Write ten questions a buyer would ask before choosing in your category, without your brand or any competitor's name in them.
  2. Ask each one in the assistants your buyers use, in a fresh session, and record the brands named and the URLs cited.
  3. Sort the results. Questions where a competitor was named and you were not are your gaps. Questions where nobody was named are open ground, and usually cheaper to win than displacing an incumbent.
  4. For each gap, open the cited URLs. They tell you where the answer came from, and therefore where your facts are missing.
  5. Repeat on a fixed schedule with the same questions. Answers vary between runs even when nothing has changed, so act on patterns that repeat, not on a single run.
question | engine | date | brands named (in order) | URLs cited | you named? | you cited?
Each brand-freequestion, per engineWho did the answer name?Which URLs did it cite?A COMPETITOR, NOT YOUGapOpen the cited URLs: they show which facts you are missing, and where.NOBODYOpen groundUsually the cheapest to win: no incumbent to displace.YOUHeld, for nowCheck cited versus recommended; re-check on the same schedule.
Figure 3. How to sort the results of the quick check. Each bin calls for a different kind of work. Diagram of the procedure in this section, not measured data.

The three bins call for different responses. A gap where a competitor is named is a content and corroboration brief: the cited URLs show which facts the answer relied on and where they live. Open ground is the cheapest win, because no incumbent has to be displaced. And a question where you are already named still deserves a look, because being cited as a source and being the brand recommended are not the same outcome; your page can supply the evidence while the answer recommends someone else.

If you would rather not keep this log by hand, InsightWonder's competitor intelligence counts every brand an answer names, not only the ones on your list.

Sources

A note on this revision: this article was rewritten to meet our current content standard. Statistics that came from roundup sites have been removed or replaced with figures read at the primary source: click-through figures now come from Seer Interactive's own study pages, and citation-overlap figures from Ahrefs' own study, with its caveat about comparability. We removed a search-volume forecast we could not open at its primary source, a search market-share figure, AI Overview prevalence and industry figures quoted from roundups, and an earlier citation-overlap figure we could only find second-hand. We also removed a composite case study and its numbers, because a case that is not one verifiable company cannot be checked, along with the old charts and images. The charts now in the article are drawn only from the primary sources listed above.