AI Visibility Checker: How to Test Whether AI Answers Mention Your Startup

An AI visibility checker answers one question: when someone asks an AI assistant about your kind of product, does your startup come up? Paid tools answer it by running prompts at scale. You can answer it yourself in an afternoon with a list of questions, a few clean browser sessions and a spreadsheet, and you'll learn more from reading the answers than from a single score.

This guide is that manual check, step by step. It gives you a baseline: a snapshot of where you stand today. To repeat it every month, see our guide to tracking your startup's AI visibility over time. To act on what you find, see how to get your startup mentioned in ChatGPT and Google AI answers. We don't sell a checker, so nothing below depends on buying one. Every product feature and source was checked on 19 October 2026.

What an AI visibility check should tell you

A checker that only says "you were mentioned 4 times" leaves out most of what you need. Build the check around four questions, and you'll know what to fix as well as where you stand.

The first three come from reading answers. The last comes from your own site, and it's the one to check first.

Before you start

Gather five things before you open any assistant. The check goes faster, and later checks will match this one.

1. 12 test prompts. Part 3 gives you a template.
2. Three competitors. Pick the products buyers most often compare you with. Ask your sales calls and support inbox, not your own wish list.
3. A reference description. Write down the facts an answer should get right: what your product is, who it's for, the main features, the pricing model, and your name and web address. You'll score accuracy against this.
4. A spreadsheet. One row per answer. The columns are in Part 4.
5. The assistants to test. ChatGPT, Perplexity and Google's AI features are a sensible starting set. Add Copilot, Gemini or Claude if your buyers use them.

Part 1: Check that AI crawlers can reach you

Answers that search the web can only cite pages their crawlers can fetch and index. Check access first, because a block here explains bad results better than anything you'll find in the answers. Our guide to getting mentioned in AI answers explains each crawler in more detail. For the check itself, work through this table and mark each row pass or fail.

Three details catch people out.

Training and search are separate. OpenAI says its crawler settings are independent of each other: blocking GPTBot, which gathers training data, doesn't block OAI-SearchBot. But "Sites that are opted out of OAI-SearchBot will not be shown in ChatGPT search answers" (OpenAI, Overview of OpenAI crawlers (opens in a new tab)). Likewise, "Google-Extended does not impact a site's inclusion in Google Search nor is it used as a ranking signal in Google Search" (Google, Common crawlers (opens in a new tab)). AI Overviews and AI Mode follow your Googlebot rules.

Google's AI features use the normal index. Google says a page must be indexed and eligible to show a snippet to appear as a link in AI Overviews or AI Mode, and that no special markup is needed (Google Search Central, AI features and your website (opens in a new tab)).

Changes take a day. OpenAI says it can take about 24 hours for its systems to adjust after a robots.txt change, and Perplexity gives a similar window. If you fix a block, wait a day before you test.

If any row fails, fix it and note the date. You can still run the rest of the check, but read the results as "before the fix".

Part 2: Set up clean test sessions

Your own chat history, saved memories and signed-in accounts can all shape an answer. You want to see what a stranger sees, so set up sessions that start from nothing.

• Use a private browser window for each assistant, and don't sign in where the assistant works without an account.
• In ChatGPT, if you sign in, use a Temporary Chat. OpenAI's help pages say Temporary Chats don't create memories, but they describe personalisation in different ways in different places (OpenAI, Temporary Chat FAQ (opens in a new tab)). Turn personalisation off for the chat if the option is there, and write down which setup you used.
• Start a new chat for every prompt. Earlier messages in a conversation shape later answers. If you ask about your product first, the assistant will happily bring it up again.
• Note whether the assistant searched the web. ChatGPT and other assistants show their sources when they search. An answer with sources and one without are different tests, so record which you got.
• Test from one location. Answers can vary by country and language. If you sell in several markets, test each one separately.
• For Google, note which feature you saw. AI Overviews only appear for some searches, while AI Mode is built to answer in that format. Record "no AI Overview" as a result; it's useful.

Then run each prompt twice, in two separate new chats. In a study by Rand Fishkin of SparkToro and Patrick O'Donnell of Gumshoe.ai, volunteers ran 12 brand-recommendation prompts 60 to 100 times each, and found less than a 1-in-100 chance of getting the same list of brands twice (Search Engine Journal, 30 January 2026 (opens in a new tab)). Two runs won't remove that noise, but they stop one unusual answer from deciding your result.

Part 3: Run the prompts

Use 12 prompts across five kinds of question. Each kind tests something different, so keep the mix even if you change the wording.

Fill in the brackets with the words your buyers use, not your marketing copy. Someone looking for an invoicing tool types "what do freelancers use to send invoices", not "best-in-class invoicing platform". Good sources for real wording are sales calls, support emails, launch comments and the queries in your Search Console report.

Twelve prompts, two runs each, in three assistants gives you 72 answers. Paste each answer into your sheet, or save a screenshot named with the prompt ID, assistant and run. You'll want the full text later when you check accuracy.

Part 4: Score each answer

Score every answer on a 0 to 3 scale. The scale rewards being named, being described correctly and being recommended for the asker's situation. It doesn't reward position in a list, because list order changes too often between runs to mean much.

For brand prompts (P11 and P12), a 0 means the assistant didn't recognise your product, or confused it with something else.

Check accuracy against your reference description

For each answer that names you, compare it with the reference description you wrote earlier. Mark any of these that are wrong:

• Category: what kind of product it says you are.
• Audience: who it says you're for.
• Features: a missing core feature, or one you don't have.
• Pricing model: free, freemium, one-time or subscription, and any price it quotes.
• Status: whether it treats you as active, or as shut down or renamed.
• Name and web address: spelling, and the link it gives.

An answer can name you and still hurt you. An assistant that says you're "enterprise only" when you have a free plan is sending buyers away.

Record what else the answer contains

Two more columns make the check far more useful.

• Competitors named. List every brand in the answer, not only your three. New names here are often products you didn't know you were competing with.
• Sources cited. When the answer shows links, record the domains. Perplexity, Google's AI features and ChatGPT with search usually show them.

Your sheet ends up with these columns:

Part 5: Work out your results

Five numbers summarise the check. Work them out for all answers, then split them by assistant and by kind of prompt.

• Mention rate: answers that name you, divided by all answers.
• Visibility score: your total points, divided by three times the number of answers. A score of 100% would mean every answer recommended you.
• Accuracy rate: answers that name you with no errors, divided by answers that name you.
• Share of voice: times you're named, divided by the total times any brand is named. Work it out for each competitor too.
• Top sources: the ten domains cited most often across all answers.

A worked example

These numbers are made up to show the arithmetic. They aren't a benchmark.

Say you ran 12 prompts twice in three assistants, for 72 answers. You were named in 18 of them, so your mention rate is 18 ÷ 72 = 25%. Your scores add up to 33 points out of a possible 216 (72 × 3), so your visibility score is about 15%. Four of the 18 answers that named you got your pricing wrong, so your accuracy rate is 14 ÷ 18, or about 78%.

Across all 72 answers, brands were named 240 times. You were named 18 times, so your share of voice is 7.5%. Competitor A was named 60 times, a 25% share. Split by kind, you might find all 18 mentions came from brand and comparison prompts, which name you in the question, and none from category or problem prompts. That pattern means assistants know your name but don't connect it with your category, and it's the one to fix first.

Part 6: Write a one-page report

A short report turns the spreadsheet into decisions, and gives you something to compare against next time. Keep it to one page.

Reading the results

Most first checks fall into one of a few patterns. Each points to a different fix.

An access check failed. Fix it before anything else. Until crawlers can reach and index your pages, better content won't show up in search-based answers.

You're named only when the prompt names you. Assistants know your product exists but don't connect it with your category or the problem you solve. That usually means too few independent pages describe you in those terms. Getting listed and discussed where your category is covered helps, from startup directories to communities and comparison articles.

You're named, but described wrongly. Look at the sources the wrong answers cite. Often it's an old directory listing, a launch page from before a pricing change, or your own outdated page. Fix your site first, then the listings.

The same few sources keep appearing. That's your target list. If a review site or a "best tools" article is cited again and again for your category, being on that page matters more than another blog post on your own site.

You're named only when the assistant searched the web. Recent pages about you are working, but the model's own knowledge likely doesn't include you yet. Keep building the pages that search finds.

Our guide to getting mentioned in AI answers covers each of these fixes in detail.

Free and paid checkers

Tools that check AI visibility do this same work at a larger scale: more prompts, more runs, more assistants and markets. They're worth considering once a manual check is too slow for what you need. Before you rely on one, ask how many times it runs each prompt, whether it uses the same apps your buyers use, whether it reports mention rate rather than a "rank", and whether you can see the full answers. Our guide to tracking AI visibility has a fuller list of questions to ask a vendor.

Where Zutto Launch fits

Zutto Launch is our product, so weigh this section with that in mind. We don't make an AI visibility checker, and no launch platform can promise you'll be named in AI answers. Zutto Radar, which is coming soon, does a different job: it will follow public launches in your niche and the competitors you choose, not AI answers. What a launch gives you today is a public page that describes your product in plain text, and one more place where people can discuss it.

• A free launch with a product page that stays up after launch day.
• A monthly leaderboard decided by community votes, so you have a month to find your audience rather than a few hours.
• The Weekly Top 3 get a promo kit: a short AI launch video, reels on our YouTube Shorts, Instagram and X accounts, and a section in our newsletter. Votes decide the Top 3; paid placement never buys a spot.
• Monthly winners can earn partner-sponsored rewards.
• Spotlight ($19, one time) adds a 2,500-word story on our blog, a featured listing in your category and a permanent do-follow link with no badge required.
• Premium Launch ($50, one time) adds an AI-generated 30-second launch video, 3 more do-follow backlinks and two posts on our blog comparing you with your alternatives.

Optional paid placement (Spotlight and Premium Launch) is clearly labelled and never changes leaderboard ranking or prize eligibility. See pricing for what each option includes, and Zutto Radar for what's planned. Zutto opens to waitlist members first.

Your next step

Block out an afternoon this week. Run the access checks first, then fill in the 12 prompt templates and run them in three assistants. When the report is done, put a date in your calendar for the next check, and use our guide to tracking AI visibility to turn this one-off check into a monthly routine.