How to check if ChatGPT recommends your business — a free DIY protocol

Straight answer

You don’t need a paid tool to find out whether ChatGPT recommends your business. You need an afternoon and a spreadsheet. The protocol: write 8–12 prompts phrased the way your buyers actually ask, run each one 3–5 times in ChatGPT in a fresh chat, repeat the same set in Perplexity and Google AI Overviews, open the citations behind every competitor mention to see why they got picked, and log all of it as share of voice. Rerun the same set monthly. The rest of this guide is the detail: what to record, which mistakes quietly make the numbers meaningless, and how to read what you find.

Why asking once proves nothing

The most common way to run this check is also the most useless: ask ChatGPT one question, once, and treat the answer as the verdict. It isn’t one. Large language models are non-deterministic: the same prompt produces different answers on different runs, and not only because of sampling settings. Research from Thinking Machines showed that LLM inference endpoints return different outputs for identical prompts even at temperature zero, because the batch size on the server varies with load and changes the arithmetic underneath. Stack retrieval on top, which pages the engine happens to pull for this particular run, and a single answer is an anecdote, not a measurement.

So treat the whole exercise as sampling. Three to five runs per prompt is the minimum that separates “we are never mentioned” from “we are mentioned sometimes,” and those are very different findings that lead to very different work.

Step 1: write buyer-phrased prompts, not brand prompts

Don’t ask “What do you know about [your brand]?” The moment your brand name is in the prompt, the model will oblige and talk about you. That tests whether it can, not whether it would. The question that matters commercially is whether you come up when a buyer asks the way buyers ask: by category, problem, and location, with no brand named.

Build 8–12 prompts across four shapes:

1. Recommendation — “Can you recommend a [service] for a small [industry] business in [city or country]?”
2. Shortlist — “What are the best [category] providers for [niche]?”
3. Comparison — “[Competitor A] vs [Competitor B] — which is better for [use case]?” Run this even though you are not in the prompt; the answer shows what the engine believes about your market, and whether it volunteers alternatives.
4. Problem-phrased — “I need [outcome] without [constraint] — who does that?”

The best source of phrasing is not your imagination. It is your sales calls, your support inbox, and your Search Console queries: the words buyers actually use, including the imprecise ones. A prompt set built from marketing language tests a market that doesn’t exist.

Step 2: run each prompt 3–5 times in ChatGPT

The mechanics matter here, because ChatGPT personalizes and a contaminated run is worse than no run.

1. Fresh chat, every run. Never re-ask inside a thread that has already mentioned your brand — the context is contaminated and every answer after that is steered.
2. Neutralize personalization. ChatGPT’s memory carries facts across conversations. If you have ever discussed your business with it, memory will skew the results toward you. Use a temporary chat, turn memory off, or run logged out.
3. Note whether the answer used web search. ChatGPT sometimes answers from training data and sometimes searches live; recommendations behave differently in each mode. Record which one you got — search runs come with source links, training-data runs don’t.
4. Record the same fields every run: mentioned yes or no, position in the list if mentioned, tone (recommended, neutral, caveated), which competitors were named, and which sources were cited, if any.

One discipline point: when the answer is bad, resist the follow-up. Asking “what about [your brand]?” feels natural and ruins the data. That is steering, and it belongs in a separate curiosity session, not in your log.

Step 3: repeat the set in Perplexity and Google AI Overviews

ChatGPT is one engine, not the market. Google’s AI Overviews alone reached two billion monthly users by mid-2025, and they appear directly inside ordinary Google searches — your buyers hit them without choosing to use an AI at all. Perplexity is smaller, but it is the most transparent of the three: every answer carries numbered citations, which makes it the best diagnostic instrument you have. So run the same prompt set in both:

1. Perplexity — same discipline as ChatGPT: fresh thread, 3–5 runs per prompt, record mentions and copy out every citation URL. The citations are the point; you will use them in the next step.
2. Google AI Overviews — type each prompt as a normal Google search. Record whether an AI Overview appears at all, whether you are mentioned in it, and which links it shows. Overviews don’t trigger on every query, and “no Overview shown for this prompt” is itself a data point — it tells you that query still resolves through classic blue links, where your existing SEO does the work.

Expect the engines to disagree. They ride different search indexes underneath, and we’ve written up how each engine chooses its sources separately, so being visible in one and absent in another is normal, not a paradox. The spread is diagnostic: it tells you which index needs the work.

Step 4: read Perplexity’s citations to see why competitors win

This is the step most people skip, and it is the one that produces the actual to-do list. A competitor mention on its own is just a bruise. The citations behind it are an explanation.

For every run where a competitor gets recommended, open each cited page and classify it:

1. What kind of page is it? A third-party “best X for Y” listicle, an industry directory, a review or comparison site, a Reddit thread, or the competitor’s own service page?
2. Whose domain is it on — the competitor’s own site, or somewhere they earned a mention?
3. Does the cited page answer the prompt’s question directly, near the top, in plain extractable text — or is the mention buried?

Do this across ten or twenty competitor mentions and a pattern appears fast: a concrete shortlist of pages and page types the engines already trust for your category. That list is your gap analysis. If every citation is a directory or listicle you are not in, the work is earning those third-party mentions. If competitors get cited from their own well-structured service pages and you never do, the problem is closer to home — your pages aren’t answering the question in a form a model can lift.

Keep in mind that being retrieved and being cited are different hurdles. Ahrefs’ study of 1.4 million ChatGPT prompts found the engine retrieves far more URLs than it credits, only about half of retrieved URLs got cited, and that citation tracked how well a page’s title and content lined up with the specific sub-questions the engine generated. The pages that win are the ones shaped like answers to the buyer’s actual question.

Step 5: track share of voice in a spreadsheet

One sheet, one row per run. The columns that have earned their place:

• date, engine, prompt, run number
• mentioned (yes or no), position, tone
• competitors named
• sources cited (URLs, for Perplexity and search-mode runs)
• web search used (yes, no, unknown), plus free-form notes

Your headline metric is share of voice: the runs where you are mentioned, divided by total runs, per engine, per month. Compute it for your main competitors too — the same rows contain everything you need.

Then read it with the non-determinism in mind. Zero mentions across 45 runs is a real finding. A competitor at 40% while you sit at 5% is a real finding. A move from 22% to 26% month over month is probably noise. What deserves your attention is the zeros, the large gaps, and trends that persist across three or more monthly runs, not single-month jitter, which the randomness guarantees.

Reading the results

Three patterns cover most outcomes.

Invisible everywhere. Usually not a content problem but a plumbing one: the underlying search indexes can’t crawl you, don’t rank you, or your pages render empty to AI crawlers. The fix starts with technical SEO: being indexed and ranking in Bing and Google is the price of admission to every engine built on top of them.

Visible in one engine, absent in another. Index divergence. ChatGPT leans on Bing, AI Overviews on Google — a site that has only ever done Google-focused SEO is often strong in Overviews and missing in ChatGPT. The gap tells you which index to work on next.

Competitors win through third-party pages. If Step 4 showed the engines citing directories, listicles, and review sites you are absent from, the gap is authority, not code — and closing it is the core of white-hat GEO: earning mentions on the sources the engines already trust, and making your own pages extractable enough to be cited directly.

One honest limit to end on: this protocol measures visibility; it doesn’t move it. Moving it is real work, indexing and rankings on the SEO side, extractable content and earned authority on the GEO side, and it takes months to compound, which is exactly why the baseline is worth establishing now. Be suspicious of anyone who reads your spreadsheet and promises a guaranteed ChatGPT mention. Citations aren’t deterministic; that is the first thing this protocol teaches you firsthand.

Sources

1. Thinking Machines — Defeating Nondeterminism in LLM Inference. Why identical prompts return different outputs even at temperature zero: batch-size variance under load (Horace He et al., 2025).
2. TechCrunch — Google’s AI Overviews have 2B monthly users. Sundar Pichai’s Q2 2025 earnings-call figure (Zeff, 2025).
3. Google — AI Overviews in Google Search. Official help page on what Overviews are and when they appear (Google).
4. Ahrefs — Why ChatGPT cites one page over another (study of 1.4M prompts). Roughly half of retrieved URLs get cited; citation tracks alignment with the engine’s sub-questions (Linehan, Ahrefs).