A founder is on the train home. Twenty minutes to kill. Out of curiosity she opens Perplexity and types her own firm's name.
The answer describes her firm as it looked six years ago. Old office. Services she dropped in 2020. A competitor's name sits two lines down, described perfectly.
To check if ChatGPT recommends your business, ask it the same questions your buyers ask, in a private session with memory off, at least three runs per prompt per engine, across ChatGPT, Claude, and Perplexity. Look at three things in every answer: whether you're named, whether the facts about you are right, and which sources the AI leans on. This checklist gives you the seven exact prompts and what each one tests. The first pass takes three minutes: one run of each prompt in one engine. The full test takes a little longer, and it's worth it.
How do I check if ChatGPT recommends my business?
Open ChatGPT in a private or logged-out window with memory turned off. Run the seven prompts below once each, filling in your details. That's the three-minute pass. Then repeat the set in Claude and Perplexity, and re-run it until you have at least three runs per prompt per engine. Write down whether you're named, what's said about you, and what sources show up. That's your baseline for generative engine optimization, the work of getting your firm named and cited by generative assistants like ChatGPT, Claude, Perplexity, and Gemini.
Before you start the clock
Three rules. Skip them and your results lie to you.
Go private. Use an incognito window, a logged-out session, or turn memory off. Your history bends answers toward firms you've already asked about, including your own.
Use three engines. Start the clock in ChatGPT. Repeat the set in Claude and Perplexity afterward. Your buyers don't all use the same one. Responsive's 2025 research found generative AI has overtaken traditional search for a quarter of B2B buyers, and nearly two-thirds use it as much as or more than search when researching vendors.
Run it at least three times. AI answers change on every run. In a SparkToro and Gumshoe.ai study of about 3,000 runs of 12 identical prompts, per Search Engine Land, the odds of getting the same brand list twice were under 1 in 100. Some brands still showed up in 60% to 90% of responses for an intent. You're looking for that kind of pattern, so plan on at least three runs per prompt per engine. The three-minute pass is your first run. For ongoing tracking, use the AI Visibility Scorecard.
Open a notes doc. Fill in the brackets. Start the timer.
Minute 1: Do they know who you are?
Prompt 1: The identity check
What does [Firm Name] in [City] do, and who do they serve?
What it tests: Whether the AI can describe your firm at all, and whether it describes you the way you'd describe yourself.
What a good answer looks like: Your actual services, your actual city, your actual client types. Close to the first two lines of your homepage.
Red flag: "I don't have specific information about that firm." Or a generic description that could fit any firm in your industry. Or it confuses you with a similarly named company.
Prompt 2: The people check
Who is [Founder Name] of [Firm Name]?
What it tests: Whether your leaders exist as credible people in the AI's view of the world. Buyers hire people at service firms, and models look for credible people too.
What a good answer looks like: Correct name, title, firm, and a line or two of real credentials or expertise.
Red flag: Nothing, the wrong person, an old title, or a firm you left years ago.
Minute 2: Do they recommend you when it counts?
Prompt 3: The category check
Who are the best [service] firms for [client type] in [region]?
Example: "Who are the best executive search firms for healthcare boards in the Southeast?"
What it tests: Whether you make the shortlist when a buyer asks for one. This is the money prompt.
What a good answer looks like: You're named, with a reason that matches your real strengths.
Red flag: Your competitors are named and you aren't. Note who shows up instead. Those names are your real competition in AI answers, and they may differ from the firms you lose deals to today.
Prompt 4: The problem check
[The question your buyer asks before they know your name.]
Examples: "How do I sell my travel agency without losing my best advisors?" or "How do I replace a CFO without spooking investors?"
What it tests: Whether the AI connects your firm to the problem you solve, before a buyer has your name to type. This is where shortlists start. 6sense's 2024 research found B2B buyers are nearly 70% through their purchasing process before engaging sellers, and 81% already have a preferred vendor at first contact.
What a good answer looks like: A useful answer that names or cites your firm, ideally quoting your own published thinking.
Red flag: A solid answer built entirely from other firms' content. You're absent from the conversation that happens before the conversation.
Prompt 5: The comparison check
[Firm Name] vs [Competitor Name]: which is better for [client type]?
What it tests: How the AI frames your differences when a buyer is down to two.
What a good answer looks like: An accurate, fair description of what each firm is known for, with your real strengths showing up.
Red flag: The competitor gets specifics and you get vague filler. Or the AI invents a weakness you don't have.
Minute 3: Do they trust you, and why?
Prompt 6: The proof check
What results has [Firm Name] achieved for clients?
What it tests: Whether your case studies, client results, and testimonials are visible enough to be repeated.
What a good answer looks like: Named clients and real outcomes, pulled from your published case studies.
Red flag: "No specific results are publicly available." Your best work is invisible to the machine.
Prompt 7: The source check
Ask this right after Prompt 3 or Prompt 6, in the same chat.
What sources did you use for that answer?
What it tests: Where the AI's picture of you comes from. In Perplexity you'll see sources by default. In ChatGPT and Claude, ask.
What a good answer looks like: Your own service and case study pages, plus credible third parties: industry associations, podcasts, press, reputable directories.
Red flag: A three-year-old directory listing, a competitor's blog, or nothing at all. If the AI can't point to your own pages, your pages aren't doing their job.
Time. Put the pen down.
Reading your results
Every red flag maps to a fix. We use the CITE Framework for this: Claim, Instrument, Testimony, Evidence. Here's where each failure points.
| If this prompt failed | Look first at | The fix |
|---|---|---|
| Identity check | Claim | Say who you serve, what you do, and where, in plain sentences on your homepage and service pages. |
| People check | Evidence and Instrument | Real bios with credentials and bylines, plus Person schema that matches the page. |
| Category check | Claim and Testimony | Specific service pages for each client type, plus third-party mentions that name you in that category. |
| Problem check | Instrument and Claim | Pages built around buyer questions, with question headings and a direct answer right under each one. |
| Comparison check | Claim and Testimony | Clear statements of what you're best at, backed by named client proof. |
| Proof check | Testimony | Published case studies with named clients and numbers. |
| Source check | Evidence and Instrument | Fix outdated listings, cite your sources, keep schema matched to visible text, and let AI crawlers in. |
The AEO checklist for B2B service firms turns each of these into specific moves. A note on the Instrument fixes: Google's own guidance says no special schema or AI text file is required to appear in its AI features, and that standard SEO applies, including "making sure your structured data matches the visible text on the page." So skip the gimmicks. Clean pages and honest markup do the work.
Proof matters more than most firms think. In 2026 we restructured RE/MAX Commercial Advisors Group's site content for AI answer engines. Afterward, Patrick J. Buckhoff, CCIM, who leads the Houston brokerage, reported: "Recently I've gotten three calls inbound that people found us online. Two from chat, one from Claude."
A worked example: what one firm fixed first
This is a hypothetical example. The firm and its results are invented to show how to read the test.
Say you run a 14-person commercial insurance brokerage in Phoenix that serves trucking and logistics fleets. You run the three-minute pass in ChatGPT on a Tuesday morning, then finish the full test in Claude and Perplexity that week, three runs per prompt per engine. Here's what you find.
Identity check: Two of three engines describe you as "an independent insurance agency in Phoenix." True, and useless. Nothing about fleets.
People check: ChatGPT lists your founder correctly. Claude has never heard of her. Perplexity names a partner who retired in 2020.
Category check: You ask, "Who are the best commercial insurance brokers for trucking companies in Arizona?" Across nine runs, you're named once. Two competitors show up in nearly every run.
Problem check: "How can a 50-truck fleet lower its commercial auto insurance premiums?" Good answers, zero mentions of you, even though you've done this for 40 clients.
Comparison check: Thin on you, specific on the competitor.
Proof check: "No specific client results are publicly available."
Source check: Perplexity cites a 2019 directory profile and a competitor's blog post.
What do you fix first? Claim. Your homepage never says the word "trucking." Everything downstream depends on that sentence, so that's where the first hour goes: rewrite the homepage and service page to say plainly that you place commercial insurance for trucking and logistics fleets in Arizona, with commercial auto, cargo, and safety-program work named outright.
Second hour: update the directory profile and the team page so the retired partner is gone and every bio is current.
Third: publish one answer page on lowering fleet premiums, with a question heading and a direct answer under it, and one short case study with a named client who agreed to it.
Run the test again in 30 days. Then start tracking it properly with the AI Visibility Scorecard, so you're measuring a trend instead of reacting to a single answer.
Make it a habit
Three minutes tells you where you stand today. Run it monthly and it tells you whether your work is landing.
The firms that show up in AI answers a year from now are the ones checking today and fixing what they find. If you want the full method behind the fixes, the generative engine optimization playbook walks through all four letters of CITE with examples.
Frequently asked questions
How do I check if ChatGPT recommends my business? Open a private or logged-out session with memory off and ask the questions your buyers ask, such as who the best firms are for your service in your region. Run each prompt at least three times per engine, in ChatGPT, Claude, and Perplexity. Note whether you're named, what the AI says about you, and which sources it cites.
Why does ChatGPT give a different answer every time I ask? AI answers are generated fresh on every run, so the brands named and their order change. A SparkToro and Gumshoe.ai study found the odds of seeing the same brand list twice were under 1 in 100. That's why you do at least three runs per prompt per engine and look for a pattern.
Should I test in ChatGPT while logged in? No. Your chat history and memory can bend the answer toward a firm you've asked about before. Use a private window, a logged-out session, or turn memory off so you see something closer to what a stranger sees.
What should I do if ChatGPT gets facts about my firm wrong? Find where the wrong fact lives and fix it at the source. That's usually an outdated page on your site, an old directory listing, or schema that doesn't match the visible text. Then make the correct fact a plain, specific sentence on your own site so a model has something accurate to repeat.
How often should I prompt-test my brand? Run the three-minute test now to get a baseline, then monthly. Once you're tracking it, move to a fixed prompt set and score mention rate, share of voice, citations, and accuracy so you can see the trend.
Ran the test and didn't love what you saw? Let's look at it together.