Your SEO report can be green on the same morning ChatGPT recommends a competitor for the query you rank first for. Most teams only check one of the two.
Take a hypothetical payroll software company and call it Fernbrook. It holds position one on Google for “best payroll software for small business.” It earned that spot the usual way. Then a finance lead at a 40-person company opens ChatGPT and types: “We’re 40 people in Texas with a few remote staff and we’re moving off spreadsheets. What payroll software should we look at?”
The answer names three vendors. Fernbrook isn’t one of them. The sources under it are a review site, a Reddit thread and a competitor’s comparison page.
Nobody at Fernbrook sees this. Neither Search Console nor the rank tracker records it. The deal just never starts.
Rank and the answer measure different things
A ranking is a position on a list of pages, for one query string, on one engine. An AI answer is a paragraph that names brands, written for a longer question and assembled from whatever sources the engine chose to read.
Rank tells you where your page sits on a list. The answer tells you whether anyone says your name.
That gap is built into how these systems work, so the next core update won’t close it. ChatGPT has no rank to track. OpenAI says ChatGPT now serves more than 1 billion people a week, and that advertising doesn’t influence its answers (OpenAI, August 2026 (opens in a new tab)). There’s no paid shortcut into those answers. And a report that stops at position one is measuring the step before the one that counts.
Even on Google, position one has moved. BrightEdge found AI Overviews on about 48% of the queries it tracks, averaging more than 1,200 pixels tall, which pushes organic results below the fold on a standard screen (BrightEdge, February 2026 (opens in a new tab)). You can hold the top blue link and still sit under an answer that names someone else.
What the overlap data says, and where it disagrees
A top ranking doesn’t reliably earn a place in the answer. The studies just disagree on how wide the gap is.
- Across 863,000 keyword results, Ahrefs found that 38% of pages cited in AI Overviews also ranked in Google’s first ten results, down from about 76% in its study the previous July (Ahrefs, March 2026 (opens in a new tab)). Its parsing improved between the two, so the drop isn’t strictly like for like.
- BrightEdge puts the overlap at about 17% (BrightEdge, February 2026 (opens in a new tab)).
- For AI assistants it was lower still. In a 2025 study of 15,000 long-tail queries, only 12% of cited URLs ranked in Google’s first ten results for the original query, and 80% didn’t rank for it at all (Ahrefs, August 2025 (opens in a new tab)). Perplexity overlapped most at 28.6%. ChatGPT and Gemini sat around 8%.
- Google’s two AI surfaces don’t even agree with each other. Across 730,000 response pairs, AI Overviews and AI Mode cited the same URLs only 13.7% of the time (Ahrefs, December 2025 (opens in a new tab)).
I wouldn’t put any one of these numbers in a board deck as the truth for your category. Each uses its own keyword set and its own way of spotting a citation. Most of this data also comes from SEO vendors, some of whom sell AI visibility tools that compete with ours. It’s still the best public data we have, and the spread between vendors shows how much the method shapes the number.
What the studies agree on is direction. A large and growing share of what AI answers cite comes from pages that don’t rank on page one.
Ranking still helps. In that Ahrefs data, the ten page-one positions hold more than a third of AI Overview citations, which is a lot for ten slots. It just won’t carry you into the answer on its own.
And these studies count citations, which are links. Being linked isn’t the same as being recommended. An engine can cite your comparison page and still tell the buyer to pick the competitor you compared yourself against.
Why a top ranking doesn’t carry over
The engine isn’t answering your keyword. Ahrefs puts much of the shift down to fan-out: Google expands a query into related searches and picks sources from those (Ahrefs, March 2026 (opens in a new tab)). Assistants start from the buyer’s own words, which carry context like team size, region and the tool they’re leaving. You ranked for the head term. The answer was built from the sub-questions.
The most exposed queries sit closest to a decision. Seer Interactive found comparison queries triggered an AI Overview 95.4% of the time and question-style queries 85.9% (Seer Interactive, April 2026 (opens in a new tab)).
The answer judges companies, and your page is only evidence. If your ranking page explains the category well but never says plainly who you’re for and why you win, the engine can use your explanation and hand the recommendation to someone else.
Mentions elsewhere seem to matter more than links. In a study of 75,000 brands, branded web mentions correlated with AI Overview visibility at 0.664, against 0.218 for backlinks (Ahrefs, May 2025 (opens in a new tab)). That’s correlation, not proof. But it matches what you find when you read the sources under an answer: reviews, forums, directories and other people’s lists.
The answer moves faster than your rank. A ranking can hold for months. Ahrefs found AI Overview content changed every 2.15 days on average, even when the gist stayed the same (Ahrefs, November 2025 (opens in a new tab)). One check tells you about one moment.
A rank-to-answer gap audit you can run this week
You can do this with a spreadsheet and a couple of afternoons. The hard part is actually reading the answers.
Pull your proudest rankings. Export 10 to 20 queries where you rank in the top three and the buyer is close to choosing: category terms, “best X for Y” queries and comparisons. Skip branded queries. You’ll win those anyway.
Rewrite each one the way a buyer asks an assistant. Keep the intent and add context, like the Fernbrook question. There’s more on this in Build a question set that sounds like your buyers, not your keyword tool.
Ask every engine your buyers use. At minimum ChatGPT, Gemini, Perplexity and Claude, plus AI Mode if Google is your biggest channel. Use a clean session with no memory or custom instructions, and ask each question at least three times per engine. SparkToro found less than a 1 in 100 chance that ChatGPT or Google’s AI returns the same list of brands in any two responses (SparkToro, January 2026 (opens in a new tab)).
-
Record four things for every answer, plus the top two or three cited domains.
What to record for every answer Field What to write down Named Does the answer text mention your brand? Position If it does, were you first, second or later? Cited Is one of your URLs among the sources, and which one? Winner Which brand did the answer put first? -
Sort every query into one of four groups.
- Ranks and named. Defend it. Note which page got cited so the next redesign doesn’t break it.
- Ranks, not named. This is the gap and your priority list. Order it by what each query is worth to pipeline.
- Named without ranking. Study it. Something other than your ranking page is doing the work, usually a third-party source.
- Neither. If it’s a real buying question, it goes on the content plan. If not, drop it.
At 20 questions, that’s 240 answers. Tedious, but do it once anyway: reading them yourself teaches you things a score won’t.
Closing the gap on the queries that matter
For each query in the gap group, open the sources the engines cited and look for what they say that your page doesn’t.
If a review site or forum thread is carrying the answer, most of the fix happens off your site, in reviews, directory listings and the communities where your buyers talk. That gets a full piece in The sources AI trusts about your brand probably aren’t yours.
If a competitor’s page is cited, compare what it states plainly with what yours only implies. Engines tend to lift clear, specific sentences. “Built for companies with 10 to 200 employees, including multi-state payroll” can be quoted. “Payroll that grows with you” can’t. (Both lines are made up for this example.)
If your own page is cited and you still lose, it’s doing the category’s homework for free. Move the reasons to pick you near the top, in plain text, with the specifics a buyer asks about: who it’s for, how pricing works, what it replaces.
If nothing of yours shows up, check the boring things first. Can the page be read without running scripts? Is the key information in text, or stuck in an image or a collapsed tab? Is your robots.txt blocking crawlers you actually want? There’s a full check for this in Can AI crawlers actually read your site?
Then ask the same questions again in two to four weeks. Same wording, same engines, same way of recording. Change any of those and you’re comparing noise.
Keep the SEO program. Change the report.
None of this means you stop doing SEO. Similarweb reports that about 95% of ChatGPT users still use Google (Similarweb, December 2025 (opens in a new tab)). In a Gartner survey, 31% of consumers said AI summaries made them search longer, against 16% who searched less (Gartner, January 2026 (opens in a new tab)). People use both.
What should change is the report. Next to rank, add a column for whether the answer names you, for the same buyer question, on each engine. Rank is an input. Being named is the outcome.
Doing this by hand once is worth it. Doing it every week across four engines and ten competitors gets old fast, and that’s the job we built Contentstack Canoe for. It runs your buyers’ questions across ChatGPT, Gemini, Perplexity and Claude on a schedule and shows whether the answer names you, who it names instead and which sources it leaned on. It sits next to Search Console, not in place of it.
Start with the five queries you’d least expect to lose.