How to check if ChatGPT recommends you

A do-it-yourself version of our free check: 5 question shapes, 3 runs each, 5 assistants, and exactly what to write down. Takes about an hour.

By Dasha. Updated November 1 2026. https://heardofyou.io/guides/check-if-chatgpt-recommends-you

Ask each assistant five realistic buyer questions, three times each, signed out. For every answer, write down four things: were you named, who was named instead, which sources it cited, and what it said about you.

Five questions, three runs, five assistants: 75 answers, about 75 minutes. Your result is a mention rate (named in X of 75), not a rank. This is the exact method behind our free check (/check). Do it yourself today, or let us do it.

Before you start
- Sign out, or use a fresh account. Signed-in assistants use memory. Yours already knows your product, so it will flatter you.
- Start a new chat for every run. No follow-ups in the same thread. The earlier answer biases the next.
- Set your buyers' location. ChatGPT follows the location the app reports. Gemini follows the network's country. Recommended businesses change with location (Chen and Koudas, 2026). Selling to the US? Test from a US location or a US VPN.
- Note the date and the model name shown on screen. Models change. Next month's comparison is only fair if you know what you tested.
- Use the web-enabled modes: ChatGPT with search, Perplexity by default, Gemini and Claude as they come. That is what buyers use.

The five question shapes

Real buyer prompts are short (Stella Rising, n=524):
- two-thirds are 15 words or fewer,
- about a quarter use the word "best",
- 28% name a price limit,
- 32% say something personal about the asker.

The five shapes below cover that mix. Write one prompt per shape, in the plain words your customers use. Not your marketing words.

| # | Shape | Template | Example (Ledgerly, an invented invoicing app for freelancers, used only as an example) |
|---|---|---|---|
| 1 | Best for use case | best [category] for [main use case] | best invoicing app for freelancers |
| 2 | Price-capped | best [category] under $[budget] a month | best invoicing app under $10 a month |
| 3 | Persona | I'm a [role] with [constraint], what [category] should I use | I'm a freelance designer who hates accounting, what invoicing app should I use |
| 4 | Alternatives | alternatives to [your biggest competitor] | alternatives to FreshBooks |
| 5 | Problem first | how do I [problem your product solves], with no category word | how do I get clients to pay invoices faster |

Shape 5 matters most for small products. It shows whether the assistant brings you up when nobody named your category. Shape 4 shows whether you appear next to the incumbent at all.

The full 25-prompt version, five per shape, is in our free prompt library (/prompts).

Run each prompt three times on each assistant

Open ChatGPT, Gemini, Claude, Perplexity and Muse. For each prompt: new chat, paste, read, record. Then twice more. That is 5 prompts x 3 runs x 5 assistants = 75 answers.

Why three runs. Two runs of the same prompt gave the same brand list less than 1 time in 100 (SparkToro, 2,961 responses). One answer is noise. Three per assistant is still small, but it tells "never" from "sometimes" from "usually". And you can repeat it in a month.

Why five assistants. Across 536,301 answers, ChatGPT, Claude and Gemini all named the same company only about 3% of the time (BusySeed). If your buyers are startup builders, Claude matters: it is the most-paid-for assistant among them. If they are consumers, Gemini and ChatGPT matter most, and Muse sits inside WhatsApp, Instagram and Facebook.

What to record for every answer

Make a sheet with one row per answer and these columns:

| Column | What goes in it |
|---|---|
| Prompt, assistant, run number, date, model | The basics, so you can repeat it fairly |
| Named? | Yes or no. Count only your product's actual name |
| Recommended or only cited? | "Recommended" means the assistant suggested you. "Cited" means it linked a page that mentions you without suggesting you. AI can quote your page and still recommend a competitor |
| Position words | "top pick", "best overall", "also worth a look", "if you need something simpler". Tone of placement, not a rank |
| Competitors named | Every product it suggested, in order |
| Sources cited | Every link it showed. These are the pages the assistant reads for your category: your target list |
| What it said about you | Price, audience, features. Check for anything wrong or out of date |

The sources column is the most valuable thing you will produce. Across 207 million citations, over 84% came from third-party sources (Muck Rack). The pages cited when you are not named are the pages you need to be on.

How to read your result

Add up the "Named?" column. That is your mention rate: named in X of 75 answers, and X of 15 per assistant.

Example result for Ledgerly (invented, for illustration, made with four assistants before Muse was added): named in 12 of 60 answers. ChatGPT 5 of 15, Gemini 3 of 15, Claude 3 of 15, Perplexity 1 of 15. Named instead: FreshBooks, Wave, Bonsai. Sources cited most: G2, a Zapier "best of" list, two Reddit threads. The to-do list writes itself. A G2 profile with real reviews. A pitch to the Zapier list's editor. Honest posts in those two threads, under the founder's own name.

Then read what the assistant said about you when it did name you. Buyers check. 89.9% of AI shoppers verify AI advice with another source before buying (Product.ai). 94% of B2B tech buyers fact-check it (TrustRadius). A wrong price or audience costs you the buyer who looked you up.

Be honest about the time

Sixty answers, recorded properly, take about an hour. That assumes you type fast and do not get absorbed in the answers. Many founders get to 20 and stop. Fine: 20 answers beat one screenshot. Just write down that you stopped, so next month's comparison is fair.

Keep the prompt set fixed and run it again at 30, 60 and 90 days. Change the prompts and you will not know whether you improved or just asked differently.

Limits

This is a sample, not a census. Fifteen answers per assistant show a pattern. They cannot measure a 5% change. Assistants also personalise, and vary by location and model version, so your numbers and a friend's will differ.

Treat the mention rate as a thermometer, not a scoreboard. Our paid audit uses the same method with 25 prompts instead of 5. More answers, same honesty.

What to do next

Rather not spend the hour? We run this check for you, free: 5 buyer questions, 3 runs each, on ChatGPT, Gemini, Claude, Perplexity and Muse. 75 answers, results by email within 2 working days, and a person checks every report.

Ran it yourself and your number is low? Read why AI doesn't recommend your product (/guides/why-ai-doesnt-recommend-you) to find the cause.

Questions

Why run each question three times?

The same question gives the same brand list less than 1 time in 100. One answer is a coin flip. Three runs per assistant give a rough rate you can compare next month.

Can I just use ChatGPT?

You can, but the assistants agree on only about 3% of recommendations. If your buyers are developers, Claude matters. If they are consumers, Gemini matters. Check all four at least once.

Should I be signed in?

No. Signed-in assistants use your memory and history, and yours knows your product. Use a signed-out window or a fresh account, so you see what a stranger sees.

How long does it take?

About 75 minutes for 75 answers, if you type fast and record as you go. Most founders never find the hour, so we run the same check free, with results in 2 working days.


Sources
- SparkToro, AIs are highly inconsistent when recommending brands (https://sparktoro.com/blog/new-research-ais-are-highly-inconsistent-when-recommending-brands-or-products-marketers-should-take-care-when-tracking-ai-visibility/)
- BusySeed via IPS, three assistants agree on 3 in 100 recommendations (https://ipsnews.net/business/2026/10/23/three-ai-assistants-agree-on-only-about-3-in-100-company-recommendations-study-finds/)
- Stella Rising via Search Engine Land, how real people prompt AI (n=524) (https://searchengineland.com/real-people-actually-prompt-ai-geo-479731)
- Chen and Koudas, location effects in AI recommendations, 2026 (pre-print) (https://arxiv.org/abs/2610.24081)
- Muck Rack, Generative Pulse, 207 million citations, October 2026 (https://www.globenewswire.com/news-release/2026/10/28/3390134/0/en/generative-pulse-earned-media-still-drives-more-than-84-of-ai-citations.html)
- Product.ai survey of AI shoppers (n=550) (https://finance.yahoo.com/technology/ai/articles/product-ai-survey-finds-four-120000400.html)
- TrustRadius, 94% of B2B buyers fact-check AI research, via MarketScale (https://www.marketscale.com/industries/software-and-technology/94-of-b2b-buyers-fact-check-ai-research-and-peer-reviews-still-close-the-deal)