Why Your Google Rankings Don’t Match Your ChatGPT Mentions, and What to Do About It

If you are on page one of Google and ChatGPT has never named you, nothing is broken. It also is not a coincidence, and it is not fixed by more of the same SEO. Teams that report on rankings alone are watching one ledger while buyers increasingly use another. This guide explains the gap with sources, shows how the major engines differ, gives you a way to diagnose it and a test to measure it, and ends with a 90-day plan. It builds on my guides to AI citation versus AI recommendation and how SEO, AEO and GEO fit together. Figures are as of October 2026, and every study is labeled with its date, method and limits.

Why can a page rank on Google and never be mentioned by ChatGPT?

Because ranking and being named answer two different questions. Google ranks pages for a query. An assistant writes an answer, and for a buying question it often chooses a short list of businesses. Five differences explain most of the gap:

1. The assistant searches more than once. Google’s documentation says its AI features may “issue multiple related searches across subtopics and data sources.” Ahrefs describes the same behavior as query fan-out and uses it to explain why AI citations rarely match the ranking for the original prompt. A page can be absent from the results for your exact words and present in the results for a related sub-question.
2. The assistant selects facts, not just pages. A page needs to hold something an answer can use: a price range, a delivery time, who something is for. A page can rank on authority and say nothing quotable.
3. The assistant names businesses. For a buying question the answer is often a short list. Being one of the pages Google ranks does not put you on that list.
4. Access is controlled separately. To appear in ChatGPT’s search features, OpenAI’s crawler OAI-SearchBot has to be allowed in. OpenAI’s documentation says sites that block it will not be shown in ChatGPT search answers, though they can still appear as navigational links. A Google ranking tells you nothing about whether that door is open.
5. The answers are not stable. SparkToro’s study found that asking the same question repeatedly returns different brand lists. A single ChatGPT result is a sample, not a rank.

What does query fan-out look like?

Here is an illustration of how one buyer prompt can become several searches. The sub-searches are invented to show the mechanism, not taken from a real log. Notice that your page might win on one of them without ranking for the sentence the buyer typed.

What does the research say about the overlap?

Four studies matter, and they measure different things, so read the limits as carefully as the headline.

Does low overlap mean ChatGPT ignores Google?

No, and this is the most common misreading. The Ahrefs figure measures where a cited URL ranks for the original prompt. An assistant also runs related searches, and the cited page may rank well for one of those. So “only 12% rank in the top 10” does not prove that the other 88% rank nowhere. It proves that your position for the question as the buyer typed it is a weak predictor of whether you are cited. The same logic explains Google’s own shift from 76% to 38%: the model pulls from more searches than the one on the screen.

It also does not mean rankings are worthless. Google says a page must be indexed and eligible to show in Search with a snippet to appear as a supporting link in AI Overviews or AI Mode, and that there are “no additional requirements” beyond that. In Ahrefs’ March 2026 data, 37.9% of citations still came from the top 10. Ranking is a way in. It is not the whole door. See Google Search Central: AI features and your website.

How do the engines differ?

“AI search” is several products with different habits. The overlap with Google varies a lot, and so does the crawler you must let in.

The practical reading: do not optimize for “AI” as one thing. If your buyers use Perplexity, rank-led SEO takes you much further. If they use ChatGPT, a page that states the answer plainly, on a site the crawler can read, matters more than your position on the exact query.

What does the gap look like on a real site?

Two of my client cases show it from opposite sides. In both, the queries are not word-for-word the same on Google and in ChatGPT, so treat them as illustrations, not a controlled test.

Case 1: ranked 4th and 5th on Google, ChatGPT’s first pick

Lens & Co by Tiara is a young Lagos eyewear store whose content began going live on September 27, 2026. By October 10, 2026, four Google searches I tracked had it on page one of Google Nigeria, but not at the top: 5th, 4th, 5th and 4th. The same site was listed first in 5 of 6 ChatGPT answers and named in all 6 Google AI Overviews, first in 4 of them. The full evidence is in the Lens & Co case study.

Why would a store at 4th and 5th beat the sites above it in the answers? I cannot prove the cause, but the evidence points one way. ChatGPT repeated facts from the store’s own pages almost word for word: delivery in 24 to 48 hours in Lagos and 3 to 7 business days elsewhere, a virtual try-on, prescription lenses through partner opticians. Each page was built to answer one buyer question. Pages built that way hand an assistant something to quote.

Case 2: first on Google, but not first in the answer

LAGE Lightings, a Lagos luxury lighting brand, shows the reverse. On October 12, 2026, in a signed-in Google search with the location set to Lagos, the first organic result for “where to buy droplights in Lagos” was the store’s droplights page. Google’s AI Overview above it still led with markets in Alaba. Separately, in July 2026 ChatGPT listed the brand seventh for luxury chandeliers in Lagos, and by October it was the first of five picks. The products and queries differ, so this is not a clean comparison, but the pattern is the same: a Google position did not decide the order of the answer. Details are in the LAGE Lightings case study.

What did Claude read instead?

For Lens & Co, AmasoAI recorded which sites Claude consulted across its checks between September 29 and October 10, 2026. Marketplaces dominated, and the store already appeared ahead of its closest brand rival. That is the other half of the gap: you are not only competing for a ranking, you are competing with whatever the assistant chooses to read.

Which gap costs you more: the ranking or the mention?

Both, but in different ways, and the data shows why neither ledger can be ignored. Think of two ledgers. The clicks ledger is Google: still the bulk of the traffic, but each result earns fewer clicks when an AI summary appears. The shortlist ledger is the assistants: small in referral traffic, but they decide which businesses a buyer considers.

• Fewer clicks when a summary appears. Pew Research Center tracked 900 US adults in March 2025: users clicked a traditional result link in 8% of visits when a Google AI summary appeared, against 15% when it did not, and clicked a source inside the summary in 1% of visits. Ahrefs’ February 4, 2026 update, on 300,000 keywords, found a 58% lower average click-through rate for the top result when an AI Overview was present (correlational, desktop). SparkToro reported that 68.01% of US Google searches in January to April 2026 ended without a click. See Pew, Ahrefs and SparkToro.
• AI referrals are small but growing fast. Similarweb counted over 1.13 billion AI referral visits to the top 1,000 websites in June 2025, up 357% on June 2024, with ChatGPT responsible for more than 80% of them. Google sent about 191 billion referrals in the same month, so AI referrals were roughly 0.6% of that by my arithmetic. See Similarweb.
• The visitors who do arrive may be worth more. Semrush’s AI search study estimated an AI search visitor at 4.4 times the value of an organic one by conversion rate, but states that its projections are extrapolations. See Semrush.

What this means for the gap: rankings still bring most of today’s clicks, so do not abandon them. But the mention decides whether you are on the shortlist, and visitors arriving from an answer have already seen your name. A business that is strong in only one ledger is leaving money on the table. For the full statistics and caveats, see AI search statistics for 2026.

How do you diagnose your own gap? The Rank–Mention matrix

This is the working model I use with clients. It is my own framework, not a published standard, and the causes are hypotheses to test. For each buyer question that matters to you, place yourself in one of four boxes by asking two things: do I rank in Google’s top 10, and does an assistant name my business?

Most businesses I audit have a mix, and the useful finding is which questions fall in box 2, because those are the cheapest wins: the page already ranks, so the audience and authority exist. Box 3 is the most dangerous to misread, because a lucky mention feels like success and disappears when the answer changes.

Why does a page that ranks go unnamed? 7 checks

When a question lands in box 2, I run these seven checks in order. It is a practitioner’s checklist from my own audits, not a study, and each one is quick to test.

What does a rewrite look like?

Here is an illustration of checks 2 and 3 with placeholder facts. It is example wording, not a real client page.

The second version states who, what, how much and how long. An assistant can lift any of those facts into an answer. The first gives it nothing to lift.

What makes an assistant name a business that does not rank top 10?

I would not claim a single cause, but four signals recur in the data and in my cases:

• Independent mentions of the brand. In Ahrefs’ study of 75,000 brands (December 12, 2025, brands with Domain Rating above 40, across ChatGPT, AI Mode and AI Overviews), YouTube mentions showed the strongest correlation with AI visibility, at about 0.74, and branded web mentions followed at 0.66 to 0.71. Domain Rating (0.27 to 0.33) and backlinks correlated far more weakly. The authors stress that correlation is not causation, and the sample is larger, more established brands than most small businesses. Read it in full at Ahrefs: brand visibility in AI.
• Quotable, specific facts on the page. Prices, times, who it is for, how it works. The Lens & Co pages were quoted back almost verbatim.
• Consistent facts across the web. In one Lens & Co answer, ChatGPT placed the store in Ojodu; in the others it said Ogba. A single consistent address, phone number and description on every profile removes that uncertainty.
• Access. The crawler must be allowed in. My guide to AI crawlers and robots.txt lists which bots to allow.

How do you measure the gap in 30 minutes?

You need a spreadsheet, a logged-out browser window and ten buyer questions. This is a shortened version of my repeatable test for what ChatGPT says about your business.

1. Write ten questions a buyer would really type, one per row: “best [what you sell] in [city],” “is [your category] genuine,” “who delivers [product] to [city].”
2. Record your Google position. Search each in a signed-out window with the country set to your market. Note your organic position, whether an AI Overview appears and whether it names you.
3. Ask ChatGPT, Gemini, Claude and Perplexity the same ten questions. Note whether you are named, where you are in the list and which sources are shown.
4. Repeat each question three times on different days. SparkToro’s study shows single answers are unreliable, so judge the pattern, not one screenshot.
5. Place each question in the matrix above and score it.

How do you score it?

Use two rates and one gap. These are my working definitions, not an industry standard.

• Top-10 rate: the share of your questions where you rank in Google’s top 10.
• Mention rate: the share of questions where an assistant names you in at least two of your three runs.
• Gap: top-10 rate minus mention rate. A large positive gap means many box 2 questions: you rank but are not named. A negative gap means box 3: you are named without ranking.

Illustration with invented numbers: you rank in the top 10 for 7 of 10 questions (70%) but are named in 3 of 10 (30%). The gap is 40 points, so start with the seven checks on the questions where you rank but are not named. If instead you rank for 2 and are named for 5, the gap is minus 30 points: you are being named on the strength of other pages or mentions, and the priority is to build the pages that make it repeatable.

A few caveats. Signed-in accounts can be personalized, so use a logged-out window where you can. Ten questions are a snapshot, not the whole market. And a tool that automates this is only as good as its sampling, a point I make in my AmasoAI review.

Which tools can automate the test?

If you already pay for Ahrefs or Semrush, start with the AI features you own. Ahrefs Brand Radar covers Google AI Overviews, ChatGPT, Perplexity, Copilot, Gemini and Claude. Semrush’s AI Visibility Toolkit starts from $99 a month per domain as an add-on to a paid plan. AmasoAI is a WordPress plugin with a free plan and paid plans from $6.99 a month, which I compare in my review. On Google’s side, Google says traffic from its AI features is counted in the Web search type of Search Console’s Performance report, and in Google Analytics 4 you can look for an AI assistant channel in your traffic acquisition report. Whatever you use, keep a manual check as your control.

If you sell software, the layer-by-layer version of this advice is in SEO vs AEO vs GEO for B2B SaaS.

How do you close the gap?

1. Open the door. Allow OAI-SearchBot, Claude-SearchBot, PerplexityBot and Googlebot in robots.txt, and check your host’s firewall and CDN are not blocking them.
2. Write one page for each buyer question, answer-first. The first lines should state the answer in plain words, with the price, time or condition an assistant can quote. My guide to how to get cited by ChatGPT shows the structure.
3. Make every fact match everywhere. Name, address, phone, opening hours and one-line description should be identical on your site, Google Business Profile and social profiles.
4. Earn independent mentions. Guest articles, expert quotes, lists and podcasts in your niche. Do not buy placement and present it as editorial.
5. Add structured data only where it is true and useful. Google says it is not required for AI features, but accurate markup helps machines read the page. See my guide to schema markup for AEO and GEO.
6. Re-test monthly and track direction. Judge by the share of questions where you are named across several runs, not by any one position.

What does a 90-day plan look like?

On timing, Lens & Co’s first ChatGPT captures came about two weeks after content started going live, and LAGE Lightings moved from seventh to first over about three months. Those are two examples, not a promise. Nobody can guarantee a position in an AI answer.

Fixing a box 2 page usually needs a writer with specific skills. I list them, with a hiring scorecard, in who can write content that gets cited by ChatGPT, Gemini and Perplexity.

What should you not conclude from this?

• “SEO is dead.” No. Google says a page must be indexed and snippet-eligible to be a supporting link in its AI features, and in Ahrefs’ latest data more than a third of AI Overview citations still come from the top 10. Rankings are still the base layer for Google and a route to Perplexity.
• “Rankings do not matter for AI.” Also no. They matter less than most people assume for ChatGPT and more than zero for others. Treat them as one signal among several.
• “One ChatGPT screenshot proves we are fine.” It does not, because the answers vary run to run.
• “Ahrefs’ 12% is a law.” It is one study of long-tail prompts with data from July 2025, and it measures rank for the original prompt only.
• “My evidence proves a method.” Two client cases do not make a study. Mine are small, early, partly signed in and partly my own work. Use them as worked examples and run the test on your own questions.

If you want a map of your own gap, I run this exact test as part of a free AI visibility audit. See the pricing page for what happens after: foundation builds start at $390 one-time and growth retainers at $290 a month. Or tell me what you want to be found for.

Keep reading: To see the gap between ranking and being named, read why your business isn’t showing up in ChatGPT or Google. In that test, ChatGPT named 50 businesses and only 5 had their own page on Google’s first page. For software buyers, I ran ten “best software” questions through ChatGPT in how SaaS products get named in ChatGPT’s best-software answers. For a security buyer’s version of this, I tested ten vendor questions in how AI recommends cybersecurity vendors. For retail and e-commerce, the Black Friday plan is in how to get products recommended by ChatGPT, Gemini and Perplexity. For regulated finance brands, see why NerdWallet gets cited instead of your fintech.