Why your brand isn't in AI Overviews — check your Google rank

If your brand is missing from AI answers, check your Google organic position for that exact query before you touch the page.

In our measurement, that one variable separated every cited page from every uncited one. Across 12 queries where Google returned an AI Overview, the brand was cited 6/6 times when it held an organic position and 0/6 times when it did not. Fisher's exact test, one-sided, p = 0.00108, zero exceptions in either direction.

We had already done the content work — schema markup, statistics, tables, author bylines, first-party numbers. The brand's appearance rate in non-branded AI answers was still 0.0%.

Key takeaways

• Organic rank was the only variable that separated cited from uncited. 6/6 with a position, 0/6 without, across 12 queries carrying a Google AI Overview. Fisher's exact test, one-sided, p = 0.00108, zero exceptions ◐.
• Indexing, country, language, brand fame and slot scarcity were each tested and each failed to explain it. All five can be checked on your own site without touching a page, and the first four take minutes — do that before committing to a rewrite.
• Being retrieved and being named are two different events. In one cell the brand's page was pulled into context 6/6 times and still appeared in only 3 of 12 answers ◐. Rewriting a page you fail to retrieve and building links to a page already being ignored are opposite jobs.
• The engines are not running the same mechanism. Across 154 head-term citations, ChatGPT gave 66% to vendor product pages and 0% to video; Gemini gave 38% to vendor listicles and 26% to video ◐. A single playbook moves one surface and leaves the other flat.
• Check what kind of question you are asking before you believe any number. On a separate 336-answer dataset the same brand scored 100% on branded questions, 57.1% on commercial and 0.0% on unbranded ◐ — a headline of 50.0% carried entirely by questions that already said its name.
• The counter-evidence is real and it is about surfaces. On ChatGPT, 44.2% of cited pages did not rank in Google's top 20 (AirOps, 82,108 citations, grade △▲). Our rank result is measured on Google AI Overviews and has not been tested on ChatGPT by us or, as far as we can find, by anyone.

Why isn't my brand showing up in AI search results?

In our data, the strongest single predictor was Google organic rank for the same query. Pages holding an organic position were cited in the AI Overview every time; pages with no organic position were cited zero times. Indexing, country, language, brand fame and answer-slot scarcity were each tested and each failed to explain the split.

That is one brand, one category, one week. Treat the direction as the finding and the values as local.

What we measured

Marks used below: ◐ our own first-party measurement, no peer review, no control group, and we sell a product in this space. ○ preprint with a control or statistical test. ◎ peer-reviewed with a control group. ▲ the author sells a related product.

Six candidates to check, and how to knock each one out

These are the six stories that could explain a missing brand, and they are separable — each one can be tested on its own site without touching the page. Run them in this order. The first four cost minutes.

The middle column is what to do on your site. The right-hand column is only what each candidate returned for us — one brand, one category, one week ◐. Yours can come back differently; that is the point of checking rather than assuming.

Ruling a candidate out is worth as much as confirming one, because each candidate implies a different quarter's work. "Nobody has heard of us" and "we don't rank for this query" lead to entirely different budgets.

The fame row is worth a sentence on its own. Obscure tools do get cited. Published work reports that visibility correlates strongly with brand prominence (Jack et al. 2026, ~37,000 runs, 533 brands, arXiv 2605.27439 — grade △▲, the authors sell AI-visibility consulting), and even their lowest tier shows 3% exposure. Our observation does not contradict their numbers. It contradicts the story that unknown brands never reach the candidate set.

Retrieval failure and selection failure are different problems

A retrieval failure is when the engine never pulls your page into its context window. A selection failure is when your page is in the context window and the engine writes the answer without you. They look identical on a dashboard and they need opposite responses.

We had both at once, on different engines ◐. In the English/Gemini/head-term cell the brand's page was retrieved 6/6 — every single time — and the brand still appeared in only 3/12 answers. The document was there. It was not chosen.

Rewriting a page you are failing to retrieve cannot work. Building links to a page that is already being retrieved and ignored is a different job — and on the largest off-site dataset we have read, links are not the strongest thing you could be doing either: across 75,000 brands, branded web mentions correlated with AI Overview presence at 0.664 against 0.218 for backlinks (Brand mentions beat backlinks, Ahrefs, grade △▲).

There is a phrasing effect underneath this, and it is not the one usually claimed: the same underlying question asked as a head term retrieved the page 6/6, while the conversational version produced 0 citations out of 242. Why that happens is in Long-Tail Prompts vs Head Terms in AI Search.

How do I check which stage is failing?

Four checks, in this order. The first two cost minutes; skipping them is how a quarter gets spent on the wrong stage.

1. Can an engine fetch and parse the page at all? A page that blocks AI crawlers in robots.txt, or that only renders its content after JavaScript, fails before ranking is relevant.
2. Does the page hold an organic position for the query you care about? Nothing inside the page tells you this. You have to look at the result set.
3. Is the page being retrieved but not selected? Where the engine exposes its sources, look for your domain in the citation list of answers that never name you.
4. Are you even asking unbranded questions? A question containing your own brand name always produces a mention and measures nothing.

Step 4 is the one people skip. In a separate dataset on the same brand, the headline appearance rate was 50.0% — and split by question type it was 100% branded, 57.1% commercial, 0.0% unbranded ◐. The full breakdown is in How to track your brand in AI answers.

That was one brand. We have since run the same split inside each of five brands on a later dataset, and four of them were named in 100% of the answers to questions containing their own name against 4.2% to 37.1% on their category questions ◐ — Branded vs unbranded prompts.

The quickest-looking version of step 4 — opening a chat and asking whether the model knows you — is the one that does not work at all. We measured it separately: Does ChatGPT Know My Brand?

How long does it take to show up in AI search?

Nobody has published a timeline for a brand's appearance rate. One study has measured how fast a page gets cited: on an established domain publishing about its own category, 36% of newly published pages were cited in Google AI Mode within 24 hours, while ChatGPT Search reached 42% by day 30. Read that as the fast end of the range, not the typical case.

Semrush, published 10 December 2025. Method as stated: 81 FAQ-style pages newly published on the Semrush blog in September, tracked for 30 days across ChatGPT Search and Google AI Mode. Grade △▲ — Semrush sells the tracking product, it is one domain, there is no control group, and the domain carries a stated Authority Score of 84.

Read the Google column twice. It rises and then falls: in their words, "by day 14, the number of pages being cited had dropped to 39 (48%)" and "by day 30, only 21 pages (26%) were still being cited." These are counts of pages cited at that moment, not a running total — a page cited in week one can be gone by week four. The two engines produced opposite shapes from the same 81 pages: Google cited new pages quickly and shed most of them; ChatGPT accumulated slowly and held.

Four boundaries on that study, which its authors are partly explicit about:

• It measures pages, not brands. "Our page was cited" and "our appearance rate rose" are different measurements, and only the first one is in there.
• It is the ceiling case. A domain that already ranks, publishing FAQ content about the category it already ranks in. A new site publishing into a category it has no position in is the situation this post is actually about, and it is not the situation they measured.
• No control group, and they say citations in AI Mode "fluctuate a lot."
• A large share of the pages were never cited at all. Across the whole window the highest readings reported are 59% on Google AI Mode and 42% on ChatGPT; every other reading is lower. Even on a domain with an Authority Score of 84, publishing is not the same as being cited.

What we can add is the shape of the wait, not its length. The chain has three stages and they are observable to very different degrees:

We have not run this ourselves and will not put a number on it. What we can say is why a quick re-check next week won't settle it for you either: roughly 65% of cited sources change between consecutive days with nobody touching anything, and at a single sample the 95% interval on a brand-detection rate is about ±72 percentage points (Schulte et al. 2026, arXiv 2604.07585, grade ○▲). A before and an after that are each one observation cannot resolve any change smaller than that noise. Fix the sample count and the observation window in advance, or the date you happen to look on will decide the result.

The two engines are not running the same mechanism

Splitting our citation data by engine produced two different pictures from the same week ◐.

Denominators: 154 citations in the head-term set, split by engine; AI Overview citations across 12 queries. These columns come from different query sets and must not be averaged.

One playbook run against both engines moves one and leaves the other flat. A dashboard that averages them shows a number describing neither. The YouTube row is unpacked separately in Does YouTube help you get cited in AI search?.

Does this contradict the GEO research?

Less than it looks. The most-quoted GEO result — Aggarwal et al. 2024, Princeton, arXiv 2311.09735, grade ○ — reports large gains from adding quotations (+42.6%) and statistics (+32.8%). That experiment measures a document already sitting in a five-document context. It is a conditional-on-retrieval effect.

The peer-reviewed replication attempt (Puerto et al. 2025, C-SEO Bench, NeurIPS D&B 2025, arXiv 2506.11097, grade ◎) found 3 significant positive results out of 54 technique-by-domain combinations, and 0 on QA tasks, and found the gains collapse as more competitors adopt the same technique — independently reproduced by Chu and Hou 2026 (+0.802 for a lone adopter, +0.007 when all nine adopt).

Our result is the same shape from the other end. We had every content attribute those papers recommend and got 0/120, because we were failing at a stage those experiments hold constant.

Where the rank finding breaks down

Our 6/6 was measured on one surface — Google AI Overviews, which are tied to Google's index by construction. The largest measurement we have found on a different surface points the other way, and it deserves to be in this post rather than in a footnote.

AirOps (grade △▲, they sell AI-visibility software) report 82,108 ChatGPT citations drawn from 15,000 original queries that expanded into 43,233 fan-out queries across 548,534 retrieved pages. In that set:

• 55.8% of cited pages ranked in Google's top 20 — which means 44.2% did not.
• 84% of all citations went to domains outside the top 100, with the single largest domain taking 2.36%.
• About 0.78% of citation events involved a page that appeared in no search result for that query at all — the model citing from memory.

Taken at face value, "check rank first" is at best a partial rule on ChatGPT's surface, and our headline is a Google-AI-Overview result wearing a general coat.

Two things let both results stand, and the distinction is worth more than either number:

A citation set can be overwhelmingly composed of pages nobody has heard of, on domains outside the top hundred, while each individual brand's citations still track that brand's own rank. The marginal describes the shape of the whole corpus. The conditional is the only one of the two that a single brand can act on, and it is the one almost nobody publishes, because it requires following one page rather than counting a corpus.

One figure in the same dataset cuts the other way, in our favour, and we would be selecting if we left it out: inside ChatGPT's own retrieval list, position 1 was cited 58.4% of the time and position 10 14.2% △▲. Order still decides citation there — it is just an order Google has nothing to do with.

What we have not done: tested the conditional on ChatGPT. We have not found anyone who has tested it on any surface other than the one we used. If it turns out to be weak there, the rank check still costs minutes — but for a ChatGPT-first audience it stops being the obvious first move, and the engine-split table above becomes the more useful part of this post.

What this cannot tell you

• n = 12. Twelve queries, one brand, one category, one week. The direction is worth believing; the values are not.
• Each AI Overview was measured once. At a single sample, the 95% interval on a brand-detection rate is roughly ±72 percentage points (Schulte et al. 2026, arXiv 2604.07585, grade ○▲, vendor-affiliated authors).
• Correlation, not proof. "Ranks organically" and "gets cited" could both be downstream of something we did not measure. Nobody has published a before-and-after where a page moved up and citations followed — including us.
• The ChatGPT half is unconfirmed. We could not obtain Bing SERP data we trust; our SERP provider returned status_code: 20000, "Ok." alongside a Bing result set that was silently wrong.
• We did not test, review or benchmark any product named here. Every domain above is a record of what was cited.

Frequently asked questions

Why do smaller brands sometimes appear in AI answers instead of larger ones?

Because the retrieval stage is not a popularity contest. Among the domains cited in our set were pexo.ai, socialync.io, quso.ai and reap.video — services with no meaningful public recognition ◐. Published work does find visibility correlating with brand prominence (Jack et al. 2026, ~37,000 runs, 533 brands, arXiv 2605.27439, grade △▲), and even its lowest prominence tier shows 3% exposure. Obscurity is a disadvantage, not a lock.

How is AI search visibility different from ordinary SEO?

It adds a stage. Ordinary SEO ends when you rank. Here, ranking gets your page into the engine's candidate set, and a second, separate decision determines whether the answer names you. We watched a page get retrieved 6/6 times and still be named in only 3 of 12 answers ◐. Those two failures look identical on a dashboard and need opposite responses.

Why does AI cite my site but never recommend my brand?

They are separable events, and we saw the split. In the AI Overviews we measured, the brand appeared as a recommendation in the answer's prose only on the two queries where it held first organic position ◐ — being used as a source was considerably more common than being advised. That is n = 12 on one brand; treat it as direction, not as a threshold.

How often should I re-measure?

Sample count matters more than calendar frequency. At one sample per question the 95% interval on a brand-detection rate is roughly ±72 percentage points; at seven samples it is ±15.8 (Schulte et al. 2026, arXiv 2604.07585, grade ○▲). A monthly measurement at seven samples per question tells you more than a daily one at a single sample. Worth knowing before you buy a cadence: of the eight vendor pages we read for 8 AI visibility tools compared, four publish a refresh cadence and all four say daily, while none of the eight documents how many times it asks each question.

Will publishing more content fix it?

We cannot tell you that it will, and nobody has published a controlled before-and-after where pages were added and citations followed — including us. What this case shows is that content work alone was not sufficient: the pages carried schema, statistics, tables and author bylines, and the unbranded appearance rate was still 0.0% ◐, because the failure was at a stage the page cannot reach.

How we run this check

The order that falls out of the data is: find out which stage is failing before you rewrite anything. That is how peekr is built. The free check takes one URL, reads that page plus the robots.txt, llms.txt and sitemap.xml at the same address, and scores three axes we deliberately never merge — AI crawler access, AI legibility, on-page SEO. The fourth axis, whether AI names you, is left blank on purpose: no amount of page inspection can fill it in.

Run the free check on your own URL One site, no account. One question asked once is a demonstration of the mechanism, not a measurement — the ±72 point interval above is why.

Method

Figures marked ◐ come from our own stored measurements on one brand in the AI video-editing category, July 2026: Google SERPs and AI Overviews via a SERP API across 13 queries with an overview on 12, and engine responses via the OpenAI and Google APIs with retrieval enabled, producing 154 citations in the head-term set and 242 in the natural-language set. Raw responses retained. We do not crawl third-party sites.

External figures are attributed inline with publisher, date, method as stated by the publisher, and a conflict-of-interest mark. We read the Semrush write-up and the AirOps report pages directly; we did not receive or inspect either underlying dataset, and neither appears to be public.

If you can replicate this on a different brand or category, please publish it — including if you get the opposite result.