How to check your AI visibility for free

You do not need an AI visibility checker to find out whether ChatGPT, Gemini and Perplexity recommend your products. You can check by hand, today, for free. This guide gives you the method step by step: what to check first, which questions to ask, how to run them so personalization does not skew the result, what to write down, and how many runs it takes before the numbers mean something.

If you want the definition first, read our guide to what AI visibility is. This one is the hands-on part.

How can I check my AI visibility score?

For most assistants there is no official score to look up, so you measure it yourself. Ask ChatGPT, Gemini and Perplexity the questions your shoppers ask, several times each, and count how often your products are named. Your score is that share, read with its number of answers.

Google is the one exception with a built-in report. Its optimization guide points site owners to the Generative AI performance report in Search Console for Google’s own AI features. We found no comparable report from ChatGPT, Gemini or Perplexity, so the check happens from the outside, the way a shopper sees it.

The whole method fits in five steps:

1. Confirm the assistants can read your store.
2. Write the questions your shoppers ask.
3. Ask each question several times on each assistant, with personalization off.
4. Record every brand named, not only whether you were.
5. Count your share and read it with its sample size.

Plan on an afternoon for a first pass with ten questions. The sections below cover each step.

Can AI systems actually reach your site?

Check this first, because no amount of good content helps if an assistant’s crawler is blocked. Your robots.txt decides which AI crawlers may read your pages. OpenAI, for example, says sites that opt out of its OAI-SearchBot will not be shown in ChatGPT search answers.

OpenAI’s crawler documentation says OAI-SearchBot is used to surface websites in ChatGPT’s search features. Perplexity and Google run their own crawlers, with their own rules. An old line in robots.txt can block one of them without anyone noticing.

Our free checker reads your store’s robots.txt and shows, crawler by crawler, whether your home, product, collection and content pages are allowed:

AI crawler access checker

Check whether 15 AI crawlers, from GPTBot to PerplexityBot, may read your site's pages.

Your website's address

For example yoursite.com.

Check my site

A quick check confirms you are a person, not a bot.

Descriptions checked 2026-10-13. We do not save the address or the result.

Open the full tool

If a crawler you care about is blocked, fix that before you spend an hour asking questions. A blocked page is a result you already know.

What are good questions to ask an AI assistant?

Good questions are the ones your shoppers type when they need your product, in their words: the problem, the use, the budget, the person it is for. Leave your brand name out. A question that names you only tests whether the assistant knows you exist.

Start with ten questions. Where to find them:

• Your site search. What visitors type into your store’s search box is the closest thing to a shopper’s own words.
• Reviews and support emails. The way customers describe the problem your product solved.
• Search data. The questions in Google’s People Also Ask boxes for your product type.

Then shape the list so it covers more than one kind of intent:

Type of question	Example for a soap brand
Need	“Which bar soap lasts longest in a shared shower?”
Comparison	“Bar soap or body wash for a gym bag?”
Constraint	“Unscented bar soap under $15”
Audience	“Gentle soap for a teenager”

Write each question two ways, in different wording. In a May 2026 paraphrase study on OpenAI and Anthropic models, reworded versions of the same question overlapped in only 14 to 29% of the brands they recommended, against 50 to 61% when the same prompt was simply rerun. Two versions give you a fairer picture than one. Keep the final list fixed. If you change the questions between checks, you cannot tell a change in your visibility from a change in your test.

How do you measure AI search visibility?

Run every question in a fresh session with personalization off, several times, on each assistant your shoppers use. Memory and past chats shape which products an assistant picks, so a personalized session measures what the assistant thinks of you, not what a new shopper sees.

Personalization matters more than most merchants expect. OpenAI says ChatGPT picks shopping results by weighing your query and your context, such as Memory or custom instructions. Its Memory help page adds that Memory can shape the search query ChatGPT runs. If you have asked ChatGPT about your own store for months, it may favor you in a way no shopper would see.

How to switch it off on each assistant:

• ChatGPT. Start a temporary chat and choose Unpersonalized before your first message. A temporary chat on its own is not enough: if you choose Personalized, it can still use your existing memories.
• Gemini. Gemini can learn from your past chats when Memory is on. Turn Memory off in its settings, as the Gemini help page describes, before you run your questions.
• Perplexity. Use incognito mode. Perplexity says memory and previous searches are always off in incognito.

Then follow the same routine for every question:

One manual AI visibility check, question by question

1. 1 Open a fresh session

   A new ChatGPT temporary chat set to Unpersonalized, a Gemini chat with Memory off, or a Perplexity incognito window. One question per session, so earlier answers do not carry over.

2. 2 Paste the question as written

   Use the exact wording from your list. Do not add your brand, your city or follow-up hints unless they are part of the question you planned.

3. 3 Record the whole answer

   Write down every brand and product named, in order, and whether yours is among them. Copy any links or sources the answer cites.

4. 4 Note what it says about you

   If your product is named, check the details: the use, the size, the price, who it is for. A wrong description is a finding too.

5. 5 Repeat the run

   Ask the same question in a new session at least three times on each assistant. Then move to the second phrasing and do the same.

Ten questions, two phrasings, three runs and three assistants make 180 answers. That sounds like a lot, but logging goes quickly once your sheet is ready.

What is an AI visibility audit?

An AI visibility audit is a structured record of what assistants say when shoppers ask about your product type: who is named, how often, how your products are described and which sources are cited. Done by hand, it is a spreadsheet with one row per answer.

Use one row per answer, with these columns:

Column	What to write
Date	The day you ran it, because answers change over time
Assistant	ChatGPT, Gemini or Perplexity
Question and phrasing	The exact wording, and which of the two versions
Run	1, 2 or 3
Brands named	Every brand, in the order they appear
You named?	Yes or no
How you were described	Any detail that is wrong or missing
Sources cited	The pages the answer links to

When the sheet is full, four counts tell you most of what a free check can:

1. Your share. Answers that name you, divided by all answers. Work it out per assistant and per question too, not only overall.
2. Who appears instead. The brands named most often in answers that leave you out. Those are your real competitors in AI answers, and they may not be the ones you track in search.
3. What is wrong. Any answer that names you with the wrong use, size, price or audience. Your product page may not state that detail clearly.
4. Which sources come up. The sites the assistants cite most for your questions. They show you where assistants read about your category.

The descriptions matter too, not only the count. An assistant that names you with the wrong details may cost you the shopper anyway.

Does ChatGPT give the same answer every time?

No. The same question rarely returns the same list of brands twice. SparkToro had 600 volunteers run 12 prompts 2,961 times and found that ChatGPT and Google’s AI returned an identical brand list less than 1 time in 100. That is why a manual check needs many runs before its share means much.

SparkToro’s research tested ChatGPT, Claude and Google’s AI Overviews and AI Mode. The same list in the same order was rarer still, fewer than 1 run in 1,000. Its authors concluded that how often a brand appears across many runs is a reasonable measure, and its position in a single answer is not.

The arithmetic makes the same point. If you are named in 1 of 10 answers, your share is 10%, but the likely range around that share is wide. The Wilson method, a standard way to put a range around a share, shows how the range narrows as answers add up:

The same 10% share, read from more answers

Source: Ranges computed by SourceVision with the Wilson method (NIST/SEMATECH e-Handbook, section 7.2.4.1)

Show data

The same 10% share, read from more answers

Answers that named you	Value (%)	Lower	Upper
1 of 10	10	1.8	40.4
2 of 20	10	2.8	30.1
5 of 50	10	4.3	21.4
10 of 100	10	5.5	17.4
20 of 200	10	6.6	14.9

Two practical rules follow:

• Do not compare small checks. Going from 1 of 10 to 3 of 10 next month looks like a tripling. Both results sit inside the same wide range, so it may be noise.
• Low shares need the most runs. One of our users, the Shopify skincare brand Jericho, was recommended in 1 of 225 answers (0.44%) when we measured it in September 2026, across ChatGPT, Gemini and Perplexity. At that rate, a 10-answer check would come back empty about 96 times in 100. A small manual check would have said “never”. The truth was “rarely”, which is a different starting point.

This is where a free check reaches its limit. A manual pass shows you the variation and the competitors, and it can turn up a wrong description worth fixing today. What it cannot do well is repeat enough questions, phrasings and runs, on every assistant, every month, to tell a real change from noise.

That repetition is the work we built SourceVision to do. We ask ChatGPT, Gemini and Perplexity your shoppers’ questions many times over and report how often you appear, with the range across runs. Our methodology explains how we measure and what the numbers cannot tell you. Then we draft what each product page is missing from facts you confirm, and nothing is written to your store without your approval. See how it works.

SourceVision is in an invite-only beta.

Request an invite

Whichever way you measure, the next step is the same: fix what the check found, then ask the same questions again. Our guide on ChatGPT recommendations covers the fixes, from crawler access and product feeds to the details your product pages should state.