Before you change anything about how your business shows up in AI search, you need a baseline. What is ChatGPT, Claude, and Perplexity already saying about you, right now, today. Most founders have never actually checked, which means most founders are making decisions about a problem they have not measured. This is a practical walkthrough for closing that gap in about twenty minutes.

Why this check matters more than it used to

ChatGPT reported more than 900 million weekly active users in February 2026 (TechCrunch), and Google says its AI Overviews feature crossed 2.5 billion monthly users (CNBC). A meaningful and growing share of the people who might hire your business are asking an AI system first. If you do not know what that system is telling them, you are operating with a real blind spot in how prospects are forming their first impression of you.

Set up the test correctly first

A few basics matter before you run a single prompt. Ahrefs' guidance on monitoring brand mentions in ChatGPT recommends testing in a logged-out or incognito browser session, since AI tools may personalize answers based on your saved memory, past chats, or custom instructions, which will skew results away from what an actual prospect would see (Ahrefs). Run each prompt more than once, ideally 10 to 20 times over a few sessions, since answers can vary between runs. And test with web search turned both on and off where the tool allows it: with search off, the model draws only on its training data, and with search on, real-time sources shape the answer, so the two settings can produce meaningfully different results (Ahrefs).

The prompts to run

1. The direct category prompt

Ask the question a prospect would actually type. "What are the best [your category] for [your specific use case]?" or "Who should I use for [your service] in [your city or niche]?" This tells you whether you show up at all in the most natural version of the question.

2. The comparison prompt

"Compare [your business] to [a known competitor]." This reveals what the AI system already believes distinguishes you, and whether that belief is accurate. It is also one of the fastest ways to spot outdated or wrong information about your business, since the model has to commit to specifics to make a comparison.

3. The brand-to-entity prompt

"List ten things you associate with [your business name]." This surfaces the concepts, services, and reputation currently attached to your name in the model's understanding, which may not match how you would describe yourself.

4. The entity-to-brand prompt

"List ten businesses you associate with [your category or keyword]." If your name does not appear, this tells you who the model considers your competitive set to be, and by extension, who is currently winning the visibility you want.

5. The source-check follow-up

Whenever a tool does mention your business, or a competitor, follow up directly: "What's the source for that?" or "Where did you find that information?" This is the single most useful prompt in this whole exercise, because it tells you exactly which external sources are currently shaping how AI systems describe you, which is where you should focus first if the description is thin, outdated, or wrong.

What to look for in the answers

Four things matter more than whether you were mentioned at all:

  • Accuracy. Is what is said about you actually true and current?
  • Specificity. Does the answer include real detail, services, outcomes, named clients, or only vague, generic language?
  • Sourcing. When you ask for a source, does the model point to something real and independent, or does it hedge?
  • Company. Who else gets named alongside you, or instead of you, and what does their content look like that yours might not have?

Research on how ChatGPT, Claude, and Perplexity choose sources found that different systems draw on genuinely different kinds of sources, with Perplexity in particular leaning heavily on community platforms for some categories (Discovered Labs). That means running the same prompt across all three tools, not just one, since a gap in one system does not necessarily mean a gap everywhere.

If you're not being mentioned, or the mention is thin

This is the expected outcome for most founder-led businesses running this check for the first time, so it is worth not overreacting to it. It usually means one of two things: your proof is trapped somewhere an AI system cannot easily read it, in emails, on sales calls, in private conversations, or it exists but is not specific and attributed enough to function as evidence. The stories on our testimonials page show what that specific, attributed proof looks like once it is out in the open. We cover exactly why specific, attributed customer stories are what closes this gap in Why Customer Stories Are the Trust Signal AI Search Actually Rewards, and the full set of signals AI systems weigh before recommending any business in How AI Actually Recommends Businesses.

Make this a recurring check, not a one-time audit

AI-generated answers change as new sources get published and indexed, so a single check is a snapshot, not a permanent status. Run the same set of prompts every four to six weeks, and keep a simple record of what changes. That record becomes genuinely useful once you start publishing new proof, since it lets you connect specific stories you have published to specific shifts in how AI systems describe you, rather than guessing at what worked.

If you want a structured way to close the gaps this check surfaces, that is the specific problem Share One was built to solve. You can see real examples of what well-attributed, specific proof looks like at our case studies page.

FAQ

How long does this check take?

About twenty minutes for a first pass across three tools and five prompt types. A recurring check, once you have a routine, takes closer to ten.

Do I need paid access to ChatGPT, Claude, or Perplexity to run this?

No. Free tiers of all three tools are sufficient for this kind of manual check. Paid tiers matter more if you move to automated, ongoing tracking tools later.

What if my business shows up but with outdated information?

Treat this the same as not showing up at all. Follow up with the source-check prompt to find out where the outdated information is coming from, and prioritize publishing current, specific, attributed content that can eventually replace it in the model's understanding.

Should I be worried if a competitor shows up and I don't?

It is useful information, not a reason to panic. Look specifically at what kind of content is attached to their mention, whether it is reviews, press, case studies, or something else, since that tells you where to focus your own effort.

Can I automate this instead of checking manually?

There are emerging tools built specifically for tracking brand visibility across AI tools over time. For a first pass, or for a business just starting to pay attention to this, manual checking with a consistent prompt list works fine and costs nothing.

What should I do with the results once I have them?

Use them to prioritize. If the gap is a lack of any mention at all, focus on getting specific, attributed customer stories published somewhere independent of your own site. If the gap is inaccurate or outdated information, focus on getting accurate, current information published and consistent across the sources the model already trusts.