ChatGPT Brand Visibility: Measuring and Raising Mentions
Being cited by ChatGPT and being named by ChatGPT are not the same thing. Many answers recommend businesses by name with no link, and many links point to reviews or articles about a business rather than to its own site.
For most brands the question that matters is simpler: when a potential customer asks ChatGPT for help, does it mention you, where in the answer, and does it describe you correctly? This guide shows how to measure that properly, how to tell searched answers from answers based on what the model already knows, and what you can realistically do to improve each.
Visibility, not just citations
Other guides in this series deal with getting ChatGPT to reach your site at all and with earning more citations to your pages. This one is about the brand: whether your name appears in answers, how prominently and how accurately. That wider view matters because of how ChatGPT answers look in practice.
- A recommendation list may name five businesses and link to none of them, or link only to a comparison article that mentions them.
- A citation can point to a third party. A review site or local news story about you counts as a mention of your brand but not a citation of your site.
- A mention can be unhelpful. Being named with the wrong location, an outdated service or as an example of what to avoid is worse than being absent.
So measure three things for every tracked prompt: whether you are named, where you appear relative to competitors, and whether what is said is right. The AI Visibility Tracker records brand presence as cited, mentioned, recommended or invisible, which maps directly onto this.
For crawler access problems, see fixing invisibility in ChatGPT. For earning more citations to your own pages, see optimising for ChatGPT.
Two kinds of ChatGPT answer
ChatGPT can answer a question in two broad ways, and the difference changes both how you measure and what you can influence.
| Searched answer | Answer from model knowledge | |
|---|---|---|
| How to recognise it | Shows sources or inline citations linking to web pages. | No sources shown. |
| Where the brand information comes from | Pages retrieved at the time of the question. | What the model learned during training, up to its cut-off. |
| How quickly it can change | Can change once the relevant pages are updated and fetched again. | Generally changes only when a newer model is released. |
| What influences it | Your pages being reachable and indexed, and what the retrieved pages say about you. | How widely and consistently you were described on the web before the training cut-off. |
ChatGPT decides for itself when to search, and users can also ask it to. Questions about recent events, local businesses, prices and product comparisons are more likely to trigger a search, but the choice can differ between runs of the same prompt. OpenAI does not publish the rules.
At launch, OpenAI said ChatGPT search uses third-party search providers alongside its own crawling and content from partners. It has not published how it ranks or chooses sources. Treat any precise claim about ChatGPT's source selection, including one you read on a vendor's site, with caution.
Testing ChatGPT without skewing the result
ChatGPT can personalise answers for signed-in users. Memory, custom instructions and previous conversations may all shape a reply, which is why the owner of a business often sees their brand recommended when customers do not. Before tracking, remove as much of that as you can.
- Use a temporary chat or a signed-out session. OpenAI describes temporary chats as not using or creating memories. Signed-out use avoids your account history altogether, though it may use a different model and have fewer features.
- Do not use the account you work in. An account used for months of conversations about your own business is the worst possible test environment.
- Keep the same product and plan. Free and paid tiers may use different models and different search behaviour. Pick the one closest to your customers and stay with it.
- Write down the model shown. When the model changes, results across every prompt can move at once.
- Keep location in mind. For local questions, name the town in the prompt rather than relying on ChatGPT to infer where you are.
The fix, step by step
Step 1
Set up clean, repeatable test sessions
Decide how you will test ChatGPT and keep to it every time. Use a temporary chat or a signed-out session so memory and past conversations do not colour the answer, avoid any account used for work on your own brand, and stay on one product tier. Record the model name shown in the interface with every run. Write these conditions into a short method note so that anyone repeating the checks, or the AI Visibility Tracker, works under the same set-up.
Step 2
Record mention, position and description
Run each prompt several times and, for every run, note whether your brand is named, where it appears relative to competitors in any list, and exactly what ChatGPT says about you. Save the full answer text. Score the description as correct, incomplete or wrong. Over a few weeks this gives you a mention rate per prompt, an average position and a list of inaccuracies, which is far more useful than a single yes or no.
Step 3
Separate searched answers from unsearched ones
For every run, record whether ChatGPT showed sources. Report mention rates for searched and unsearched answers separately. If you appear in searched answers but not in unsearched ones, your current web presence is working and the gap lies in older, training-era coverage that you cannot change quickly. If you appear in neither, begin with access and with the third-party pages that retrieved answers draw on.
Step 4
Read the sources behind searched answers
For prompts where competitors are named and you are not, open every cited source and note its type: comparison article, review site, directory, forum thread, news story or a competitor's own page. Check whether you are mentioned there at all, and how. Patterns appear quickly. If the same few comparison pages and directories keep being cited, your absence from them is likely to explain much of your absence from the answers.
Step 5
Confirm OpenAI's search crawler can reach you
Check robots.txt and your firewall for rules affecting OAI-SearchBot,
which OpenAI documents as the crawler used to surface sites in ChatGPT search, and look in your access
logs for its requests and their status codes. Blocking GPTBot is a separate decision about
model training. If the search crawler is refused, fix that first, since no amount of reputation work
helps searched answers link to a site ChatGPT cannot read.
Step 6
Improve how others describe you
Work through the third-party pages you found in step four. Claim and complete business directory and review profiles, correct wrong details, and ask genuine customers for reviews on the platforms that appear as sources. Where a comparison article is cited, find out whether the publisher accepts corrections or updates for businesses it omits. Keep every approach honest; paid placements disguised as editorial and fake reviews can harm you and may breach consumer protection law.
Step 7
Make your own description consistent everywhere
Write one plain description of what you do, for whom and where, and use it on your
About page, home page, business profiles and social bios. Use the same business name, address and service
list everywhere, and add Organization or LocalBusiness markup with sameAs
links to those profiles. When every source describes you the same way, a summary drawn from several of
them is more likely to be accurate.
Step 8
Report monthly by prompt group
Summarise results monthly for brand, category, comparison and problem prompts, split into searched and unsearched answers. Show mention rate, average position, share of voice against a fixed competitor list and the number of inaccurate descriptions. Mark each change you made and any model update you noticed on the same timeline. Look for movement that repeats across several prompts in a group rather than reacting to single answers.
Reading a ChatGPT answer for visibility
A single answer contains more evidence than a yes or no. When you store answers, capture these details:
- Order. In a list of suggestions, earlier positions are more visible. Record your position and who comes before you.
- Framing. "A popular choice for small teams" and "an option if budget is not a concern" are both mentions, but they send different people your way.
- Facts stated. Location, prices, services, founding date. Any error here is a correction job, and the tracker surfaces wrong-attribution incidents for this purpose.
- Sources. Which pages were cited, and whether you are named on each.
- Caveats. Phrases such as "I could not find recent information about" suggest ChatGPT lacks current sources about you.
An example record for a fictional firm, Kestrel Print Studio in Sheffield:
Prompt: Where can I get short-run wedding stationery printed in Sheffield?
Run: 3 of 5, temporary chat, sources shown
Named: yes, position 3 of 4
Said: "Kestrel Print Studio offers letterpress invitations" (correct)
"open Saturdays" (wrong: closed Saturdays since 2025)
Sources: example-weddings.co.uk/sheffield-stationers (names us)
example-directory.co.uk/printers/sheffield (old hours listed)
Action: update directory hours; add hours to contact page as textWhat is unlikely to help
- Asking ChatGPT to remember your business. Memory applies to the account that set it, not to other users.
- Repeatedly prompting your own brand. There is no public evidence that the number of times people ask about you changes what ChatGPT tells others.
- Publishing a custom GPT. A GPT can be useful in its own right, but OpenAI has not said that having one affects whether ChatGPT mentions you in ordinary answers. ChatGPT plugins, which some older guides still mention, were retired in 2024.
- Stuffing pages with your brand name or with phrases like "best in the UK". This makes pages worse for readers and there is no evidence it helps.
- Assuming an llms.txt file does the job. OpenAI has not said ChatGPT uses llms.txt when choosing sources. It is harmless, but not a substitute for the steps above.
Edge cases
A name shared with another business
If another company, product or place shares your name, ChatGPT may blend the two. Always include a distinguishing detail such as town or trade in your own descriptions, and in tracking, score mentions by hand to check which business was meant.
A recent rebrand
Answers from model knowledge may use your old name for some time. Keep a clear statement of the change on your About page, redirect old URLs, and update profiles, so searched answers can pick up the new name even if unsearched ones lag behind.
Very new businesses
A business founded after a model's training cut-off will not appear in unsearched answers from that model at all. Focus measurement on searched answers, and expect unsearched mentions only after newer models are released.
Regulated sectors
For health, legal and financial questions, ChatGPT may favour official and professional bodies and avoid naming individual firms. Being listed accurately on the relevant professional register may matter more than your own site.
Keeping ChatGPT visibility on track
- Keep one description of the business and change it everywhere at once when something changes.
- Check key third-party profiles quarterly for outdated hours, prices and services.
- Watch OAI-SearchBot in your logs after any firewall, CDN or robots.txt change.
- Keep test conditions fixed and note model changes on your timeline.
- Look at AI referrals in analytics. OpenAI has said links from ChatGPT search carry
a
utm_source=chatgpt.comparameter, so visits from those links can be seen in your reports, though not every visit is labelled.
For how to run the measurement consistently across every assistant, see fixing AI visibility tracking.
1. Why does ChatGPT recommend my business to me but not to my customers?
Because your own account is probably personalised. If you have discussed your business in past conversations, or it is saved in memory or custom instructions, ChatGPT can draw on that when answering you. Customers asking the same question have none of that context. Test in a temporary chat or signed out, with the questions customers ask, and repeat each prompt several times before concluding whether ChatGPT recommends you. A handful of clean runs is far more reliable than one answer in your own account.
2. How can I tell if a ChatGPT answer used web search?
Searched answers show sources, either as inline citations or as a list of links you can open. Answers without any sources were most likely written from what the model learned in training. Record this for every run you track, because the two kinds of answer respond to different work. Searched answers can change once the pages they draw on change, while answers from training data usually only change when a newer model is released.
3. Can I pay to appear in ChatGPT answers?
OpenAI publishes no programme that lets businesses buy or request a place in ChatGPT's answers, so be wary of anyone who promises guaranteed placement. What you can do is make your site reachable to OpenAI's search crawler, describe your business clearly and consistently, and improve the third-party pages ChatGPT draws on when it searches. These are slower routes, but they are the ones that the way ChatGPT gathers information actually supports.
4. Does blocking GPTBot remove my business from ChatGPT?
According to OpenAI's documentation, GPTBot collects content that may be used for training, while OAI-SearchBot is the crawler used for ChatGPT's search features, and they are controlled separately in robots.txt. Blocking GPTBot is therefore a training decision rather than a search one. Blocking OAI-SearchBot is what OpenAI says keeps a site out of search answers. Check OpenAI's current bot documentation before changing either, since names and behaviour can change, and check your firewall as well as robots.txt.
5. Why does ChatGPT describe my business wrongly?
Usually because a source it relies on is wrong or out of date, such as an old directory entry, a stale review profile or a page on your own site that nobody updated. For answers without sources, the model may be repeating what it learned in training. Record exactly what is wrong and which sources were cited, correct the sources you control or can influence, and state the right facts clearly on your own site with a date.
6. How often should I check my ChatGPT visibility?
Collect results weekly or fortnightly, running each prompt several times, and review them monthly. Checking every day mostly shows random variation between runs, while checking once a quarter makes it hard to connect changes with anything you did. Keep a dated log of your own changes and of model updates you notice, so that when the monthly figures move you can see what happened in the weeks before and judge whether the movement is real.
7. Is being mentioned without a link still useful?
Yes. A recommendation that names you can send people to search for your business directly, and it shows that ChatGPT associates you with the topic. It is worth tracking separately from citations, though, because the causes differ. Mentions without links often come from what reviews, articles and directories say about you, while citations of your own pages depend on those pages being reachable, relevant and clear about the question asked.
The AI Visibility Tracker queries ChatGPT, Claude, Perplexity, Gemini, Copilot and Brave Leo with your target prompts, records whether your brand is cited, mentioned, recommended or invisible, tracks frequency over time and identifies gaps against competitors.
Open AI Visibility Fixes →