---
title: "Your Brand Doesn’t Rank in ChatGPT. It Has a Mention Rate."
description: "One ChatGPT screenshot cannot tell you where your brand stands. Follow a Dallas service prompt through retrieval and sampling to see why AI visibility needs repeated checks, a defined scope, and a mention rate."
url: "https://groas.com/post/your-brand-doesn-t-rank-in-chatgpt-it-sh"
image: "https://pub-87da24ecbbfc4c3bad6875f3aa013712.r2.dev/generated-images/fc4db853-8122-4259-9700-3e55fcf82f80.png"
published: "2026-10-26T05:34:39.541Z"
modified: "2026-10-26T05:34:39.611Z"
---

October 26, 2026 · 10 min read

# Your Brand Doesn’t Rank in ChatGPT. It Has a Mention Rate.

[Alexander PerelmanHead Of Product @ groas](https://groas.com/author/alexander-perelman)[LinkedIn](https://www.linkedin.com/in/alexander-433793253/)

![A raffle tumbler of white balls, some stamped with a red garage-door icon, drops one ball mid-air, suggesting AI brand mentions are a random draw, not a fixed rank.](https://pub-87da24ecbbfc4c3bad6875f3aa013712.r2.dev/generated-images/fc4db853-8122-4259-9700-3e55fcf82f80.png)

In this article

1. [One screenshot is one draw](#one-screenshot-is-one-draw)
2. [The prompt branches before you see an answer](#the-prompt-branches-before-you-see-an-answer)
3. [Four reasons the brand list moves](#four-reasons-the-brand-list-moves)
4. [Treat visibility like impression share, with one important limit](#treat-visibility-like-impression-share-with-one-important-limit)
5. [Ask which buying topic you miss](#ask-which-buying-topic-you-miss)
6. [Give the rate enough runs to mean something](#give-the-rate-enough-runs-to-mean-something)
7. [Make the dashboard answer three questions](#make-the-dashboard-answer-three-questions)
8. [Put the model to work in a new market](#put-the-model-to-work-in-a-new-market)

A Dallas garage door company checks ChatGPT once for “best garage door company in Dallas for same-day spring repair.” A competitor appears; the company does not. Within minutes, someone forwards the screenshot with a subject line in all caps.

I used to get that email about Google rankings. At least then we could argue about position three versus position five. Here, **there is no fixed brand position to argue about**. An AI answer is generated anew on each run. The pages retrieved and the words chosen can change, so the same question can name different brands on different runs, engines and days. In one volunteer test with 2,961 repeated runs, there was [less than a 1 in 100 chance of getting the same brand list twice](https://pages.citepoint.ai/articles/why-ai-answers-change), and about 1 in 1,000 for the same list in the same order. Your brand does not hold a ChatGPT rank. It appears some percentage of the time.

## One screenshot is one draw

One check feels decisive because Google trained us to look for a position, get a number and act. But an AI answer is a draw from a changing set of possible answers, not a standing assigned to your company.

In one repeat-run study of 968 runs across 42 prompt-engine pairs, [42.9% flipped between mentioned and not mentioned at least once](https://www.airanklab.com/blog/ai-answer-stability-study). Consecutive runs shared a median of only 41% of cited domains, and a single run captured about 37% of the total source pool observed in the study. A screenshot can be genuine and still show you very little of what repeated checks would reveal.

I used to tell PPC clients the same thing about a single auction: losing one impression out of thousands is no reason to rewrite every ad. I was wrong to think they would listen, but right about the math. If the Dallas spring-repair question names you on Monday, skips you on Tuesday and names you alongside two competitors on Wednesday, “we checked and we’re not there” is not a finding. **It is one observation.** The job is to measure how often it happens.

## The prompt branches before you see an answer

Follow that Dallas question through the process. What you typed does not necessarily go into one search and come back as one result. ChatGPT can [rewrite a prompt into narrower sub-queries, search them and build an answer from the pages it finds](https://mavek.ai/blog/chatgpt-query-fan-out/). Google says AI Overviews and AI Mode [“may use a query fan-out technique, issuing multiple related searches across subtopics and data sources”](https://www.airanklab.com/blog/ai-answer-stability-study).

The same-day spring-repair question might branch into questions about repair cost in Dallas, same-day availability, replacement time and the risks of doing the job yourself. A page that answers one of those questions may enter the answer even if it does not target the exact phrase you typed. I unpack that rewrite step further in [_Your Buyer Asks One Question. AI Searches for the Parts._](https://groas.com/post/chatgpt-doesn-t-answer-your-buyer-s-ques)

Each sub-search can pull a different shortlist. A local service page might answer availability; a Reddit thread might discuss pricing; a manufacturer guide might explain the repair. The model then compresses what it retrieved into a response. This is the step a lot of visibility advice skips. Your Dallas company did not necessarily lose a contest against one competitor’s page. It may have been absent from the pages retrieved for several parts of the question.

Those sub-queries can also vary with wording, location, conversation, memory and the run itself, so [two people asking the same question can receive different sources](https://mavek.ai/blog/chatgpt-query-fan-out/). **Cover the natural sub-questions of the topic**, rather than trying to guess and target one exact hidden query. You cannot count on that query being used next time.

### Retrieval sets the candidates; generation chooses the words

Suppose retrieval does return the same pages twice. You still should not expect identical wording. The model generates its response from a probability distribution, and output can vary even under settings intended to reduce variation. In a Thinking Machines Lab test of 1,000 completions at temperature 0, there were [80 unique answers; the most common appeared only 78 times](https://pages.citepoint.ai/articles/why-ai-answers-change). The reported quirk was traced to server batching. Anthropic and OpenAI also acknowledge that [zero temperature or a fixed seed does not make output fully deterministic](https://pages.citepoint.ai/articles/why-ai-answers-change).

For the Dallas prompt, one run might name a competitor whose pricing guide was retrieved. Another might name your company when a same-day-service page informs the response. **Retrieval affects which brands can readily be supported; generation affects which ones appear in the answer you receive.** Neither step gives you a permanent position. Ask what share of runs name you, not which rank you hold.

## Four reasons the brand list moves

Once you separate retrieval from the final wording, the changes between runs become less mysterious. Four things can move the answer:

1. **Sampling.** The generated response can vary even when the question looks identical.
2. **Context.** Location, conversation history, memory and small wording changes can alter the sub-queries and sources.
3. **Engine.** Different systems can retrieve different pages and assemble different answers.
4. **Date.** Indexes and models change, so last week’s sources need not be this week’s.

When someone tells me a competitor “ranks” in ChatGPT, I ask which of those they held constant. Usually, none.

The engine distinction matters more than a combined score suggests. [ChatGPT leans on the Bing index plus its own crawler, while Google uses its own index](https://pages.citepoint.ai/articles/why-ai-answers-change); Perplexity runs retrieval-first with its own crawler. Ahrefs found that [AI Mode and AI Overviews cited the same URLs only 13.7% of the time across 540,000 pairs](https://pages.citepoint.ai/articles/why-ai-answers-change), though 86% were semantically similar. Similar subject, different source list. **Measure each engine separately** before you roll anything into an average.

![Illustration of a slot machine showing different garage door brands for the same prompt](https://pub-87da24ecbbfc4c3bad6875f3aa013712.r2.dev/generated-images/c80c33ca-0145-4e59-bad4-474e21a31d73.png)

## Treat visibility like impression share, with one important limit

I lived inside impression share for years. It is a better mental model for AI visibility than organic position, as long as you know where the analogy breaks.

Google defines Search impression share as [impressions received divided by estimated eligible impressions](https://support.google.com/google-ads/answer/9244124). You never expected to win every eligible auction. A competitor could outbid you, your budget could run out at 4pm, or your ad could lose one auction and win the next. A 62% impression share did not mean you ranked sixth. It described how often you appeared in the auctions counted as eligible.

AI visibility is also about frequency, but **a prompt run is not an ad auction**, and its retrieval shortlist is not Google Ads eligibility. You have to define the questions and conditions you are counting. Then you can ask: across those runs, how often was your brand named?

Keep three measures separate:

- **Mention rate:** the share of runs in which your brand appears at least once.
- **Citation rate:** the share of runs in which one of your pages is cited as a source. [A brand mention and a page citation are different events](https://ahrefs.com/blog/ai-share-of-voice/).
- **Position within an answer:** among the runs that mention you, whether you appear first, third or with a caveat. This describes those answers; it is not a standing rank.

Ahrefs defines AI share of voice as [responses mentioning your brand divided by responses mentioning any tracked brand](https://ahrefs.com/blog/ai-share-of-voice/), then weights it by volume. That is another useful measure, but do not quietly swap its denominator for all prompt runs. Whatever you report, attach the scope: prompts, engines, location and dates. “30 out of 100 Dallas runs on ChatGPT last week” gives me something to work with. “30% visibility” without a scope is decoration.

## Ask which buying topic you miss

The Dallas company should not put every spring-repair question into one bucket. “How much does spring repair cost?” can pull pricing guides and Reddit threads. “Who does same-day spring repair?” can pull local service and availability pages. They sound closely related to a customer, but they can produce different shortlists. You might appear often for one and rarely for the other.

Group prompts by the question a buyer is trying to answer. I use five to eight buying topics and three to five phrasings per topic, with location and language fixed. For this company, the list could include:

- Garage door spring repair cost in Dallas.
- Same-day garage door repair in Dallas.
- Garage door company reviews in Dallas.

Run each phrasing more than once and keep the results by topic and engine. A tool worth paying for should let you inspect that split, not just hand you a grand total. The useful approach is to [run several prompts per topic and model, report the share of answers mentioning you, and track cited pages](https://searchengineland.com/monitor-brand-visibility-ai-search-channels-448697). The citations help you see which kinds of pages are informing the answers.

If both pricing phrasings rarely name you, investigate the pages retrieved around pricing. If one phrasing names you and a near twin does not, do not rush to declare your pricing content broken; wording and run-to-run variation may explain the gap. **The topic split tells you where to look.** It does not, by itself, prove why you are missing.

## Give the rate enough runs to mean something

A percentage looks precise long before it is useful. For a prompt that names you 30% of the time, the approximate 95% margin is [20 percentage points with 20 runs, 9 points with 100 runs and 4.5 points with 400 runs](https://pages.citepoint.ai/articles/why-ai-answers-change). Six mentions out of 20 and eight out of 20 are not grounds for two different strategies.

Even SparkToro’s practical suggestion of [60 to 100 runs per prompt](https://pages.citepoint.ai/articles/why-ai-answers-change) gets you only to single-digit precision. I would rather run a fixed prompt set on a fixed schedule, keep market and language consistent, and retain the full answer text alongside rolling seven-day rates. [_Build an AI Visibility Tracker That Measures Mention Rates, Not Screenshots_](https://groas.com/post/build-your-own-ai-visibility-tracker-in) lays out one setup: 20 prompts, five runs each per engine, counted as a rate.

That discipline changes how you read a jump. If a score moves 12 points week over week on 40 runs, [_Your AI Visibility Score Is a Sample, Not a Ranking_](https://groas.com/post/why-your-ai-visibility-number-changes-ev) is the piece to read before announcing a win. The change may be noise. Repeated movement in the same direction, under the same conditions and across more runs, deserves more attention than one exciting week.

You will still see advice to run 20 to 50 buyer prompts once across ChatGPT, Gemini and Perplexity, then log the mentions. That can help you assemble a prompt list. It cannot establish a dependable mention rate. **One run per prompt is still one draw per prompt**, however polished the dashboard looks.

## Make the dashboard answer three questions

Before you believe a visibility number, ask:

1. **What is the scope?** Which prompts, engines, location and dates does it cover? Ahrefs likewise says to [report prompts, platforms, competitors, location and period](https://ahrefs.com/blog/ai-share-of-voice/).
2. **How many runs sit behind it?** If there is one per prompt, the report is a collection of screenshots with better fonts.
3. **Is it a snapshot or a sustained trend?** A move from 28% to 34% on 60 runs a week is not, on its own, proof that anything changed. Keep watching under the same conditions before acting.

The point is not to make every report a statistics seminar. It is to stop treating a neat percentage as stronger evidence than the checks behind it. **Read the sample before you read the score.**

![Infographic showing 32 mentions in 100 runs and a margin that narrows as runs increase](https://pub-87da24ecbbfc4c3bad6875f3aa013712.r2.dev/generated-images/24956bd4-45f4-40f6-879a-f3a774c2b135.png)

## Put the model to work in a new market

Now say you open a plumbing shop in Fort Worth. Your first week of AI checks shows zero mentions out of 15. The Dallas screenshot lesson still applies: do not rewrite the site because of that one small set of checks.

Start with eight buying topics and four phrasings for each. Run each phrasing five times per engine on a weekly schedule, and pool the answers over two rolling weeks. Keep the engines separate. If the pricing topic sits at 8% while emergency repair sits at 41%, investigate pricing first: inspect the answers and cited pages, then see whether your own page answers the pricing question directly with local specifics a generic guide lacks. After a change, watch that topic’s rate over the next two windows under the same conditions.

That is how groas approaches the measurement: repeated checks across ChatGPT, Perplexity, Gemini and AI Overviews, counted as a share of voice rather than a rank. The first 15 checks did not tell the plumbing shop where it stood. A defined set of questions, repeated over time, can tell it **how often it shows up and where to investigate next**.

## Frequently Asked Questions

### Why doesn't my brand have a fixed rank in ChatGPT like it does in Google?

ChatGPT generates a new answer on every run, and the pages it retrieves and the words it picks can change each time. One study of 2,961 repeated runs found less than a 1 in 100 chance of getting the same brand list twice. Your brand appears some percentage of the time rather than holding a standing position.

### Can one ChatGPT screenshot tell me whether my brand is visible in AI answers?

No. One answer is a single draw from a changing set of possible answers, not a standing assigned to your company. In one repeat-run study of 968 runs, 42.9% of prompt-engine pairs flipped between mentioning and not mentioning a brand at least once, so a genuine screenshot can still show very little of what repeated checks would reveal.

### How does ChatGPT rewrite my prompt before answering it?

ChatGPT can rewrite a prompt into narrower sub-queries, search them, and build an answer from the pages it finds; Google says AI Overviews and AI Mode may use a similar query fan-out technique. A same-day repair question might branch into sub-queries about cost, availability, replacement time and DIY risks, each pulling a different shortlist of sources.

### What makes AI answers about the same brand change between runs?

Four things move the answer: sampling (the generated response can vary even for an identical question), context (location, conversation history, memory and wording), engine (different systems retrieve different pages), and date (indexes and models change). Holding none of these constant means the brand list can differ from run to run.

### Should I measure my brand's AI visibility separately for ChatGPT, Google and Perplexity?

Yes. ChatGPT leans on the Bing index plus its own crawler, Google uses its own index, and Perplexity runs retrieval-first with its own crawler. Ahrefs found AI Mode and AI Overviews cited the same URLs only 13.7% of the time across 540,000 pairs, so measure each engine separately before rolling anything into an average.

### What is a mention rate in AI visibility tracking?

A mention rate is the share of prompt runs in which your brand appears at least once. It should be kept separate from the citation rate (the share of runs where one of your pages is cited as a source) and from position within an answer, and always reported with a scope: prompts, engines, location and dates.

### How many runs do I need per prompt to get a reliable AI visibility number?

More than most dashboards use. For a prompt naming you 30% of the time, the approximate 95% margin is 20 percentage points with 20 runs, 9 points with 100 runs and 4.5 points with 400 runs. SparkToro suggests 60 to 100 runs per prompt, which still only gets you to single-digit precision.

### How should I group prompts when tracking my brand's AI visibility?

Group prompts by the question a buyer is trying to answer, using about five to eight buying topics and three to five phrasings per topic, with location and language fixed. Run each phrasing more than once and keep results by topic and engine, because closely related phrasings like pricing versus same-day availability can produce different retrieved pages and different mention rates.

## Paid Ads & SEO, Run For You

Businesses pay one flat monthly fee per service, agencies resell it under their own brand, and groas does the work either way.

[See Pricing](https://groas.com/pricing)

## Structured data

```json
{"@context":"https://schema.org","@graph":[{"@type":"Organization","@id":"https://graph.groas.com/entity/groas","name":"groas","url":"https://groas.com/","description":"groas is a done-for-you marketing service, not software: elite people working with proprietary technology systems run search ads, Meta ads, SEO and AI search for businesses, and white label for agencies, with a dedicated senior strategist owning every account.","logo":{"@type":"ImageObject","url":"https://groas.com/icon-512.png","width":512,"height":512},"sameAs":["https://www.linkedin.com/company/groas/"]},{"@type":"WebSite","@id":"https://groas.com/#website","name":"groas","url":"https://groas.com/","inLanguage":"en-US","publisher":{"@id":"https://graph.groas.com/entity/groas"}},{"@type":"WebPage","@id":"https://groas.com/post/your-brand-doesn-t-rank-in-chatgpt-it-sh#webpage","url":"https://groas.com/post/your-brand-doesn-t-rank-in-chatgpt-it-sh","name":"Your Brand Doesn’t Rank in ChatGPT. It Has a Mention Rate.","description":"One ChatGPT screenshot cannot tell you where your brand stands. Follow a Dallas service prompt through retrieval and sampling to see why AI visibility needs repeated checks, a defined scope, and a mention rate.","inLanguage":"en-US","isPartOf":{"@id":"https://groas.com/#website"},"about":{"@id":"https://graph.groas.com/entity/groas"},"primaryImageOfPage":{"@type":"ImageObject","url":"https://pub-87da24ecbbfc4c3bad6875f3aa013712.r2.dev/generated-images/fc4db853-8122-4259-9700-3e55fcf82f80.png"},"breadcrumb":{"@id":"https://groas.com/post/your-brand-doesn-t-rank-in-chatgpt-it-sh#breadcrumb"},"dateModified":"2026-10-26T05:34:39.611Z","mainEntity":{"@id":"https://groas.com/post/your-brand-doesn-t-rank-in-chatgpt-it-sh#article"}},{"@type":"BreadcrumbList","@id":"https://groas.com/post/your-brand-doesn-t-rank-in-chatgpt-it-sh#breadcrumb","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https://groas.com/"},{"@type":"ListItem","position":2,"name":"Blog","item":"https://groas.com/blog"},{"@type":"ListItem","position":3,"name":"Your Brand Doesn’t Rank in ChatGPT. It Has a Mention Rate.","item":"https://groas.com/post/your-brand-doesn-t-rank-in-chatgpt-it-sh"}]},{"@type":"BlogPosting","@id":"https://groas.com/post/your-brand-doesn-t-rank-in-chatgpt-it-sh#article","url":"https://groas.com/post/your-brand-doesn-t-rank-in-chatgpt-it-sh","mainEntityOfPage":{"@id":"https://groas.com/post/your-brand-doesn-t-rank-in-chatgpt-it-sh#webpage"},"isPartOf":{"@id":"https://groas.com/blog#blog"},"headline":"Your Brand Doesn’t Rank in ChatGPT. It Has a Mention Rate.","description":"One ChatGPT screenshot cannot tell you where your brand stands. Follow a Dallas service prompt through retrieval and sampling to see why AI visibility needs repeated checks, a defined scope, and a mention rate.","datePublished":"2026-10-26T05:34:39.541Z","dateModified":"2026-10-26T05:34:39.611Z","image":{"@type":"ImageObject","url":"https://pub-87da24ecbbfc4c3bad6875f3aa013712.r2.dev/generated-images/fc4db853-8122-4259-9700-3e55fcf82f80.png"},"author":{"@id":"https://groas.com/author/alexander-perelman#person"},"publisher":{"@id":"https://graph.groas.com/entity/groas"},"keywords":"runs, brand, chatgpt, mention rate, mention, rate, dallas, prompt","wordCount":2090,"inLanguage":"en-US"},{"@type":"Person","@id":"https://groas.com/author/alexander-perelman#person","name":"Alexander Perelman","jobTitle":"Head Of Product @ groas","description":"My background includes Goldman Sachs and Stanford Computer Science.","image":"https://groas.com/media/blog/4f17cac3a81acc224cc1b2cfab8f94bbc295a3089d13e8421125771ef8d4064f.jpg","worksFor":{"@id":"https://graph.groas.com/entity/groas"},"sameAs":["https://www.linkedin.com/in/alexander-433793253/"]},{"@type":"FAQPage","@id":"https://groas.com/post/your-brand-doesn-t-rank-in-chatgpt-it-sh#faq","mainEntity":[{"@type":"Question","name":"Why doesn't my brand have a fixed rank in ChatGPT like it does in Google?","acceptedAnswer":{"@type":"Answer","text":"ChatGPT generates a new answer on every run, and the pages it retrieves and the words it picks can change each time. One study of 2,961 repeated runs found less than a 1 in 100 chance of getting the same brand list twice. Your brand appears some percentage of the time rather than holding a standing position."}},{"@type":"Question","name":"Can one ChatGPT screenshot tell me whether my brand is visible in AI answers?","acceptedAnswer":{"@type":"Answer","text":"No. One answer is a single draw from a changing set of possible answers, not a standing assigned to your company. In one repeat-run study of 968 runs, 42.9% of prompt-engine pairs flipped between mentioning and not mentioning a brand at least once, so a genuine screenshot can still show very little of what repeated checks would reveal."}},{"@type":"Question","name":"How does ChatGPT rewrite my prompt before answering it?","acceptedAnswer":{"@type":"Answer","text":"ChatGPT can rewrite a prompt into narrower sub-queries, search them, and build an answer from the pages it finds; Google says AI Overviews and AI Mode may use a similar query fan-out technique. A same-day repair question might branch into sub-queries about cost, availability, replacement time and DIY risks, each pulling a different shortlist of sources."}},{"@type":"Question","name":"What makes AI answers about the same brand change between runs?","acceptedAnswer":{"@type":"Answer","text":"Four things move the answer: sampling (the generated response can vary even for an identical question), context (location, conversation history, memory and wording), engine (different systems retrieve different pages), and date (indexes and models change). Holding none of these constant means the brand list can differ from run to run."}},{"@type":"Question","name":"Should I measure my brand's AI visibility separately for ChatGPT, Google and Perplexity?","acceptedAnswer":{"@type":"Answer","text":"Yes. ChatGPT leans on the Bing index plus its own crawler, Google uses its own index, and Perplexity runs retrieval-first with its own crawler. Ahrefs found AI Mode and AI Overviews cited the same URLs only 13.7% of the time across 540,000 pairs, so measure each engine separately before rolling anything into an average."}},{"@type":"Question","name":"What is a mention rate in AI visibility tracking?","acceptedAnswer":{"@type":"Answer","text":"A mention rate is the share of prompt runs in which your brand appears at least once. It should be kept separate from the citation rate (the share of runs where one of your pages is cited as a source) and from position within an answer, and always reported with a scope: prompts, engines, location and dates."}},{"@type":"Question","name":"How many runs do I need per prompt to get a reliable AI visibility number?","acceptedAnswer":{"@type":"Answer","text":"More than most dashboards use. For a prompt naming you 30% of the time, the approximate 95% margin is 20 percentage points with 20 runs, 9 points with 100 runs and 4.5 points with 400 runs. SparkToro suggests 60 to 100 runs per prompt, which still only gets you to single-digit precision."}},{"@type":"Question","name":"How should I group prompts when tracking my brand's AI visibility?","acceptedAnswer":{"@type":"Answer","text":"Group prompts by the question a buyer is trying to answer, using about five to eight buying topics and three to five phrasings per topic, with location and language fixed. Run each phrasing more than once and keep results by topic and engine, because closely related phrasings like pricing versus same-day availability can produce different retrieved pages and different mention rates."}}]}]}
```