AI Visibility: Does ChatGPT Recommend Your Brand?

To find out whether ChatGPT recommends your brand, ask it the questions your buyers ask, run each one several times in a fresh session, and record whether you were named, where you ranked in the list, and which sources it cited. One answer proves nothing. A structured set of prompts run repeatedly is a measurement.

This guide sets out how to run that test properly, what to do with the result, and how to fix the three things that most often keep a brand out of AI answers.

Why a single check is misleading

AI assistants do not return a fixed result the way a search engine does. The same prompt can produce different brands on different runs, in different accounts, and with browsing switched on or off. Personalisation, memory of your past chats and the model version all change the answer.

That means three rules for any test worth acting on. Use a logged out or fresh session so your own history does not bias the result. Run each prompt at least three times. Record the answers rather than trusting your memory of them.

Build the prompt set

A useful test covers four categories. Write five to ten prompts in each, in the language your customers actually use.

The category and problem prompts are the commercially important ones. Brand prompts tell you whether your own description is accurate. Category prompts tell you whether you are in the consideration set at all.

Score the answers

Record four fields for every run. Whether the brand was mentioned. Its position in any list. Whether a link to your site was cited. Whether the description was accurate.

Two scoring rules save a lot of self-deception. First, count position as well as presence. Being eighth in a list of ten is not the same as being second. Second, treat a citation of your own site as weaker evidence than a citation of an independent source. A model that only knows about you from your own marketing copy has no corroboration, and corroborated brands survive model updates better.

Test more than ChatGPT

ChatGPT is the question people ask, but it is not the whole surface. Gemini, Claude, Perplexity and Google’s AI features all draw on different data and update on different cycles. A brand can be strong in one and invisible in another. Any conclusion drawn from a single assistant is a conclusion about that assistant, not about your AI visibility.

Google documents how AI features use content from the web, and it is worth reading alongside your results, because the eligibility rules for AI surfaces are the ordinary indexing rules.

The three things that usually keep you out

1. The crawlers cannot reach you

An AI assistant cannot cite what it cannot fetch. Check your robots.txt against the published crawler lists before assuming a content problem. OpenAI documents GPTBot, and Google lists Google-Extended and its other agents in the crawler overview. Many sites block these by inheritance rather than by decision, having copied a robots.txt from a template.

Blocking them is a legitimate commercial choice. Blocking them accidentally while asking why ChatGPT never mentions you is not.

2. There is nothing quotable to cite

Models cite specific, checkable statements. Pages built entirely from adjectives give them nothing to lift. Pages with clear definitions, named figures, dated claims and plain answers to direct questions give them a great deal.

The practical fix is structural. Answer the question in the first hundred words of the page. Use headings that match the question form. Keep one idea per sentence so a passage can be extracted without losing its meaning. Google’s people-first content guidance describes the same qualities for a different reason, and the overlap is not a coincidence.

3. Nobody else says it about you

This is the hardest and most valuable one. Models weight independent corroboration. If your claim to be a leading supplier appears only on your own site, it carries little weight. If it appears in trade press, directories, review platforms, comparison articles and forum discussion, it becomes something the model treats as established.

Practically, that means the off-site work that earns AI mentions is the same work that earns links. Digital PR, expert commentary, being included in third party comparison pieces, and having accurate, consistent entity information across the places that describe your market.

Fix the description as well as the ranking

A frequent and underrated finding is that the model names you but describes you wrongly. Old positioning, a service you dropped, a location you left, a founder who moved on. That is a correction problem rather than a visibility problem, and it is usually caused by stale third party pages ranking above your current ones.

Fix it by updating your own canonical descriptions first, then by getting the stale third party sources corrected or outranked. Consistency across your site, your profiles and your listings is what makes a description stick.

Turn the test into a tracker

A one-off audit gives you a baseline. It does not tell you whether you are gaining or losing ground, and models change constantly. Re-run the same prompt set on the same schedule, monthly at minimum, and chart mention rate and average position over time.

Two metrics are worth separating. Overall mention rate across all prompts, and non-brand mention rate across category and problem prompts only. The second is the honest one. A brand can score well overall purely because it appears when its own name is in the prompt, which proves very little.

Terrier runs this as a repeatable programme. The one-off baseline is the AI visibility audit, and the ongoing measurement is AI visibility tracking. GEO retainers start from £1,250 a month with no minimum term and 60 days’ notice, and the bands are set out on the GEO pricing page.

A thirty minute version you can run today

1. Open a logged out ChatGPT session.
2. Ask “best [your category] in [your city]” three times, in three separate chats.
3. Record every brand named and the order.
4. Ask “what is [your brand]” once and read the description carefully for errors.
5. Check your robots.txt for GPTBot and Google-Extended rules.
6. Repeat steps two and three in Gemini.

That takes half an hour and tells you which of the three problems you have. If you are not named anywhere and the crawlers are blocked, you have an access problem. If the crawlers are open and you are still absent, you have a corroboration problem. If you are named but described wrongly, you have a correction problem.

What to do with a bad result

Absence from AI answers is rarely fixed by writing more blog posts. Work through the causes in order of cost.

Start with access, because it is free. Read your robots.txt line by line and confirm every AI crawler you want is allowed. Check that your key commercial pages are actually indexed in Google, since the same crawling and indexing rules govern eligibility for AI surfaces.

Move to extractability next. Take your five most commercially important pages and rewrite the opening of each so it answers the obvious question directly, in plain sentences, before any positioning language. Add the specifics a model can quote: what you do, where, for whom, at what starting price, with what evidence.

Then work on corroboration, which is the slowest and most valuable. Get your entity information consistent everywhere it appears. Earn coverage in the trade press and comparison sites your buyers already read. Contribute expert commentary under a named author. These are the same activities that build links, and they compound.

Common mistakes when testing

• Testing while logged in. Your own history biases the answer towards brands you have discussed.
• Running each prompt once. A single sample from a probabilistic system is an anecdote.
• Using prompts nobody would type. Test the language buyers actually use, not your internal category names.
• Only testing brand prompts. Being named when your name is in the question proves almost nothing.
• Celebrating a self-citation. A model quoting only your own site has no independent corroboration to fall back on.
• Comparing runs across different dates without recording the date. Models change. Undated results cannot be compared.

Frequently asked questions

How do I check if ChatGPT recommends my brand?
Run a fixed set of category, comparison and problem prompts in fresh sessions, at least three times each, and record mentions, position and citations.

Why does ChatGPT give different answers each time?
Model responses are probabilistic, and browsing, memory and account personalisation all change the result. That is why repetition matters.

Does blocking GPTBot hurt my visibility?
It removes your pages as a source for that crawler. Whether that is a cost or a protection depends on your business, but it should be a decision rather than an accident.

Is AI visibility just SEO?
It overlaps heavily. Crawl access, indexing, clear answers and independent corroboration all matter to both. The measurement is different because there are no rankings to check.

How often should I re-test?
Monthly. Models update frequently and competitors are working on the same problem.

Related pages

• AI visibility audit
• AI visibility tracking
• GEO pricing and packages
• Schedule a call to baseline your brand across ChatGPT, Gemini and Claude.