To track your ChatGPT mentions, write a fixed list of 8 to 12 buyer prompts a real customer would type, run them in a clean ChatGPT session on the same day each week, and log three things per prompt: did your brand appear, where in the answer it sat, and was there a link to you. One answer is noise — ChatGPT is non-deterministic, personalized, and grounded on live browsing. The signal is the week-over-week trend across the same prompts. Run it as a loop and you can finally see whether your AI visibility work is moving.
Why is a single ChatGPT check worthless?
One screenshot of ChatGPT naming you tells you almost nothing. ChatGPT answers are non-deterministic, so the same prompt can return different brands on different runs. They are personalized by your account's memory and history, so the version that names your brand may only do that because you have been chatting about it. And when browsing is on, the answer is grounded on live web results that shift week to week. Any of those three alone makes a one-off check unreliable.
This is the same trap most "how to get cited" advice falls into: it tells you to optimize but never tells you how to verify it worked. Measurement that you cannot repeat under controlled conditions is just hope with a screenshot. The fix is not a fancier tool — it is turning the check into a fixed, scheduled experiment where the only thing allowed to change is the answer itself, not your prompt, not your account, not the day of the week.
What prompts should I actually test?
Test the questions your buyers really ask, not your brand name. Searching "is Acromatico good" proves nothing, because you are leading the model to you. Instead, build prompts around the buying decision, phrased the way a customer speaks: "best brand studio for AI visibility," "who can help a small business get recommended by AI," "alternatives to a traditional SEO agency for 2026." If ChatGPT names you for those unprompted, that is a real mention.
Lock in 8 to 12 of these and never change the wording, because a changed prompt resets your baseline. Cover three intent layers: a broad category prompt, a problem-shaped prompt, and a head-to-head or "alternatives to X" prompt. This is the same buyer-question discipline behind getting cited by ChatGPT in the first place — you are simply turning those target questions into a permanent test harness instead of a wishlist.
How do I run the test so it's fair?
Run every prompt in a clean session to strip out personalization. Use a temporary or logged-out chat so ChatGPT's memory of your past conversations cannot inflate the result, and turn web browsing on so you are testing the live, grounded answer your buyers actually get. Run each prompt once, in its own fresh chat, and resist the urge to re-roll until you see your name — re-rolling is how you lie to yourself.
Do the whole set on the same day and time each week. Consistency in the test is what makes the trend trustworthy; if you run it on a Monday this week and a Friday next week with a different account, you have measured your own inconsistency, not ChatGPT's. Treat it like a darkroom exposure: same chemistry, same timing, every time, so the only variable left is the image coming up.
What exactly do I log for each prompt?
Log three fields per prompt, plus the date. First, appearance: was your brand named, yes or no. Second, span: where did it appear — in the first recommendation, somewhere in the middle of a list, or only in a footnote-style source. Third, link: did the answer cite or link to your own domain, a third-party page about you, or nothing. A brand named first with a link to its own site is winning that prompt; a brand buried in position six with no link is barely on the board.
A plain spreadsheet is enough: one row per prompt, one column per week, and a simple score in each cell. Many people score it 0 for absent, 1 for mentioned, 2 for mentioned-and-linked, 3 for named-first-and-linked. The exact scale does not matter; consistency does. This is the operational core of an AI visibility audit — the audit is the first week's snapshot, and this weekly log is what turns that snapshot into a moving picture.
How do I read the week-over-week trend?
Read the slope, not any single cell. A prompt that moves from a 0 every week to an occasional 1 is real progress, even if you do not own the answer yet — it means ChatGPT's grounding is starting to surface you. A prompt stuck at 0 for two months is a clear signal to go build or earn more about that specific buyer question. And a prompt that drops from 2 back to 0 is an early warning that a competitor's coverage is now outranking yours on that topic.
Patience is part of the protocol. ChatGPT leans heavily on third-party sources and consistent facts, both of which take time to accumulate and be re-crawled, so expect meaningful movement over 6 to 12 months, not weeks. The weekly cadence is not about expecting weekly wins — it is about catching the slope early and knowing which lever to pull next, long before the change would show up in any traffic report.
How does this connect to actually getting mentioned more?
The log is only half the loop; the other half is acting on what it tells you. When a prompt stays cold, the fix is almost always upstream: ChatGPT is reading what trusted third parties say about you, so the move is to earn a mention on a source it already trusts, tighten the facts on your own pages so they agree everywhere, and make those answers extractable. The weekly score tells you which buyer question to aim that work at, so you are never optimizing in the dark.
This is exactly the loop we run as one visibility engine across more than 10 brands: measure the same prompts every week, find the coldest high-intent question, fix the source and the facts behind it, then watch the score move. If you want the full do-it-yourself version of the engine, the DIY AI visibility audit walks the build-and-measure side, and our AI visibility audit gives you a clean baseline to start your first week's log from.
What this protocol will not do
This protocol will not give you a guaranteed number or a controllable rank. There is no ranked list inside ChatGPT to be number one in, and because answers are non-deterministic, even a strong week can be followed by a quiet one. Anyone selling "guaranteed ChatGPT placement" is selling something — the honest version is a trend you steer, not a position you buy. What the weekly log honestly buys you is sight: instead of guessing whether your AI visibility work is paying off, you can point at a spreadsheet and show the slope. That is the whole difference between hoping and running a system.
Questions people ask
Build a fixed list of 8 to 12 buyer prompts a real customer would type, run each one in a clean ChatGPT session with web browsing on, and log whether your brand appears, where in the answer it sits, and whether it links to you. Repeat the exact same prompts on the same day every week so the only thing changing is the answer, not your test. The week-over-week trend, not any single run, is the signal.
ChatGPT answers are non-deterministic, personalized by memory and history, and grounded on live browsing results that shift, so the same prompt can name different brands on different days. That is exactly why you measure on a schedule with a fixed prompt set and read the trend across weeks instead of trusting one answer. Use a logged-out or temporary chat to strip personalization out of the test.
Expect meaningful movement over 6 to 12 months, not weeks, because ChatGPT leans on third-party sources and consistent facts that take time to accumulate and be re-crawled. A weekly log is what lets you see the slope early: a prompt moving from never naming you to occasionally naming you is real progress long before you own the answer outright.
Want this done for you?
Want to know if ChatGPT is naming you — and watch it improve week over week? Start with an AI visibility audit.
Get a free AI Visibility Audit →