How to Check If ChatGPT, Gemini and Perplexity Mention Your Brand (Free Method)

To check for free whether ChatGPT, Gemini and Perplexity mention your brand, write 20 questions your buyers actually ask, run each in all three engines from clean sessions, and log whether you're mentioned, recommended or cited and which sources appear. It takes about an hour. The result is a baseline you can repeat monthly and a short list of the sources shaping answers in your category.

What you need: free accounts for each engine, a private browser window, and the sheet template below.

On this page

• Step 1: What questions should you test?
• Step 2: How do you set up each engine fairly?
• Step 3: How do you log results?
• Step 4: How do you score it?
• From MentionX data: what to expect
• Step 5: What should you do with the results?
• What does a filled-in row look like?
• Worked example: Ledgerline's first free check
• How do you read the answers, not just score them?
• How do you share results with your team?
• What about free online checkers?
• How do you keep the check fair?
• How often should you repeat the check?
• Why do results vary so much?
• What mistakes ruin a manual check?
• How do you turn the check into a habit?
• Common questions
• Where to go next
• Sources

Engine behaviour checked against Google and OpenAI documentation on 30 October 2026. "Our view" marks MentionX opinion. Ledgerline is fictional.

Step 1: What questions should you test?

Don't ask "What is [YourBrand]?" or "Do you know [YourBrand]?". Buyers rarely ask that; they ask about problems and options. Write 20 questions in four groups:

Get the wording from sales calls, support tickets and forum threads. If a phrase sounds like marketing copy, rewrite it the way a customer would say it.

Step 2: How do you set up each engine fairly?

Same location for all runs. Don't sign in with an account that has used your brand a lot; the engine may personalise.

Step 3: How do you log results?

Create a sheet with these columns:

Run each prompt twice per engine if you can. That's 120 answers for 20 prompts × 3 engines × 2 runs. If time is short, run once and repeat the 10 most important prompts.

Tips while logging:

• Paste the answer text into a notes column; you'll want it later.
• "Mentioned" means your brand name appears. "Recommended" means it's suggested as a fit for the question's situation.
• Record all cited domains, not only yours.

Step 4: How do you score it?

Calculate, per engine and overall:

• Mention rate = answers mentioning you ÷ total answers
• Recommendation rate = answers recommending you ÷ total answers
• Citation rate = answers citing your domain ÷ total answers
• Top competitor mention rate, for context

Then count the most-cited domains across all answers. The top 10 are your target list.

There's no "good" universal score; compare against competitors on the same prompts and your own next month.

Which formulas do the scoring?

In Google Sheets, with "Yes"/"No" in the Mentioned column (E) and engine names in column C (Google Sheets functions):

Mention rate, all:        =COUNTIF(E:E,"Yes")/COUNTA(E2:E)
Mention rate, ChatGPT:    =COUNTIFS(C:C,"ChatGPT",E:E,"Yes")/COUNTIF(C:C,"ChatGPT")

Use the same pattern for recommendation and citation rates. For the domain list, split the "Other sources" column into one domain per row and use a pivot table to count them.

From MentionX data: what to expect

Method. Read-only aggregate queries over completed MentionX runs from 22 August to 29 October 2026 (UTC): 2,154 runs across 18 anonymised projects, mostly B2B software, called through engine APIs.

Position matters too. In 102 unbranded answers that named the brand in a list, it was first in 18%, second or third in 14%, fourth or fifth in 13%, and sixth or lower in 56%. Being mentioned often means being one of many.

Limits. A convenience sample, mostly B2B software, through APIs, which can differ from the free consumer apps you'll use. Perplexity's sample is small. These figures are context, not benchmarks for your brand.

Our view: a manual check is worth doing even if you later buy a tool. Reading 120 answers yourself teaches you more about how engines describe your category than any chart.

Step 5: What should you do with the results?

Pick three actions based on what you found:

What does a filled-in row look like?

Illustrative example, not real data:

That single row already gives you two actions: find where the outdated SMS claim comes from, and look at why the review sites cited don't feature you.

Worked example: Ledgerline's first free check

Ledgerline is a fictional accounts-payable tool. Figures are illustrative. Its marketing lead runs 20 prompts in ChatGPT, Gemini and Perplexity, twice each: 120 answers in about 70 minutes.

Three findings shape the next month:

1. Problem prompts are almost empty: 1 mention in 36 answers. Buyers asking "how do we stop duplicate invoice payments" never hear of Ledgerline.
2. A wrong fact repeats. Four answers say Ledgerline has no Xero integration, which launched last year. All four cite the same 2023 review article.
3. One domain dominates the comparison answers. A software review site appears in 11 answers, and Ledgerline's profile there has 6 reviews against a competitor's 140.

The three actions: a problem-led guide on duplicate payments, an email to the review article's editor, and a review request to recent customers.

How do you read the answers, not just score them?

The yes/no columns give you rates. The text tells you why. After logging, spend 15 minutes reading answers in batches by prompt group and note:

• How you're described. Which category, audience and strengths does the engine attach to you? Is it the positioning you want?
• Who you're grouped with. The competitors named next to you show which set the engine thinks you belong to. Sometimes it's not the set you'd choose.
• The reasons given for recommendations. "Best for small teams" or "good integrations" show which claims engines have picked up from sources.
• Hedges and caveats. Phrases such as "some users report" or "may be expensive" often trace back to a review or forum thread you can find in the cited sources.
• What's missing. Your newest features, or the market you most want to win, may never appear.

Put the three most common descriptions in your summary. They often get more attention from leadership than the rates do.

How do you share results with your team?

Keep it short: one table with the four rates per engine, the top 10 cited domains, and three actions. Link the raw sheet. Avoid screenshots of single answers unless they illustrate a pattern you've seen across many runs.

What about free online checkers?

Several SEO vendors offer free AI visibility checkers, including HubSpot's AI Search Grader, Ahrefs' AI visibility checker and Semrush's AI search visibility checker. They're quick, and they're a reasonable first look. Check each one's page for what it covers today, because limits and engines change.

Keep three differences in mind:

Our view: use a free checker for a quick read, and the manual method for anything you'll make decisions on. The prompts you choose matter more than the tool that runs them.

How do you keep the check fair?

Small differences in setup change results, so control what you can:

• Same time window. Run all prompts within a day or two. Engines and their sources change over weeks.
• Same person, same device type. A colleague on a phone in another country may see different answers.
• Signed out or temporary chats where possible, so personal history doesn't steer the answer.
• Exact wording. Copy prompts from your sheet; don't retype them.
• Note the mode. Record whether the engine searched the web. In MentionX data, the share of answers with no citations ranged from 4.1% in Perplexity to 50.7% in Gemini, so "searched or not" is a big source of difference.

How often should you repeat the check?

Monthly, with the same prompts and setup. Keep a "prompt version" note so changes are visible. When you're tracking more than about 50 prompts or several markets, manual checking gets slow; see manual vs automated tracking.

Why do results vary so much?

Generated answers aren't fixed. They depend on:

• Whether the engine searched the web for that run
• Model version and mode
• Location and language
• Account memory and history
• Which pages were retrieved at that moment

Google, for example, notes that AI Overviews and AI Mode may use different models and techniques, so responses and links vary (Google). That's why rates over many answers matter more than any single result.

What mistakes ruin a manual check?

• Testing only brand-name prompts.
• Testing from your work account, which knows your brand.
• Logging only your own citations, not everyone's.
• Running each prompt once and treating the result as fact.
• Sharing a single screenshot with leadership as "proof".

How do you turn the check into a habit?

Put a recurring 45-minute block in the calendar on the same weekday each month. Use the same prompt file, the same engines and the same logging sheet, and add a "date run" and "who ran it" column so the history stays readable. If two people share the work, split it by engine rather than by prompt; that way each person keeps one engine's settings consistent. Once three months of rows exist, you'll be able to tell a real shift from ordinary run-to-run variation, which is the point where the free method starts paying back the time it costs.

Common questions

Can I check more than three engines? Yes. Add Google AI Mode or Copilot if your buyers use them. Keep the same 20 prompts so the engines can be compared.

What counts as "recommended"? The answer suggests your brand as a good fit for the situation in the question, not just lists it. Write your rule down before you start, and apply it the same way every month.

What if I'm never mentioned? That's a useful baseline, not a failure. Check that your brand name isn't being confused with something else, then look at the sources cited for category prompts. Those are the pages to get onto first.

Should I count mentions in a list of ten? Yes, but record the position. Sixth place in a long list is a much weaker result than first.

Where to go next

Turn this into a proper audit with a source map and fix list: How to run an AI visibility audit. Add AI referral tracking in analytics: Track AI traffic in GA4.

Sources

• Google Search Central, AI features and your website
• OpenAI, crawlers; Perplexity, crawlers
• OpenAI, Searching the web with ChatGPT
• Google, Google Sheets function list
• HubSpot, AI Search Grader; Ahrefs, AI visibility checker; Semrush, AI search visibility checker