Type your brand into a free checker, get a score, screenshot it, post it in Slack. Everyone nods. Nobody knows what to do next.
That is the problem with a one-off LLM visibility check. It is a photo of one moment, taken with one prompt, on one engine. Treat it as a starting point and it is useful. Treat it as a verdict and it will mislead you.
What an AI visibility checker actually does
A checker sends a handful of prompts to one or more AI engines and reads the answers. It looks for your brand name, sometimes your domain, and reports back with a score or a yes and no. Most free versions share the same limits.
A small fixed prompt set
One run, no history
One or two engines
No competitor context
None of that makes them useless. It makes them a smoke alarm. They tell you something may be wrong. They do not tell you where the fire is.
Why one check lies to you
AI answers are not stable. Ask ChatGPT the same question three times and you can get three different shortlists. Your brand can appear in the first, vanish in the second and return in the third.
That variance is noise, and a single run cannot separate noise from signal. A good score today may be luck. A bad score may be a weak sample.
Prompt wording matters just as much. "Best CRM for a ten-person agency" and "top CRM tools" pull from different parts of a model's memory and different sources on the web. A checker that tests only one phrasing has tested almost nothing.
How to run a check you can trust
You can get a far better read from any checker, free or paid, by changing how you use it.
Build your own prompt set
Write 15 to 20 prompts the way a buyer would type them. Cover three types.
Category prompts, such as "best tools for X"
Problem prompts, such as "how do I fix Y"
Comparison prompts that name you and a rival
Keep the list in a spreadsheet. Same prompts, every time. That is the only way a number can change for a reason.
Test more than one engine
ChatGPT, Perplexity, Gemini and Copilot draw on different sources. A brand can lead in Perplexity, where citations are visible, and barely register in ChatGPT. One engine gives you a partial answer.
Repeat each prompt
Run every prompt at least three times. Record how often you appear, not just whether you did. Two appearances out of three is a different story from one out of three, and both are different from zero.
Log position and sources
Being mentioned is the floor. Note where you sit in the list and which URLs the engine cites. If a competitor's comparison page keeps getting quoted instead of yours, you have found something to fix.
How to read the result
Resist the single score. Read three numbers instead. Mention rate is the share of prompt runs where your brand appears. Position is how high you rank when you do. Citation share is how often the engine links to your own pages rather than someone else's description of you.
A brand can have a high mention rate and almost no citations. That means the model knows your name but learned it from third parties. You are visible, and you do not control the story.
The signs you have outgrown a free checker
Stay with a free tool if you are checking once a quarter and want a rough read. Move on when any of these sound familiar.
You re-run the same prompts by hand every week
Leadership asks if visibility is improving and you cannot say
A competitor keeps showing up in answers you want
You need to prove a content change moved the number
Each one points to the same gap. You need history, competitor comparison and consistent prompts, which no single check provides.
That is where LLM brand visibility tracking takes over. It runs your prompt set on a schedule across engines, stores every answer, and shows the trend, the competitors named beside you and the sources behind each answer.
Run this test today
Pick ten prompts. Run each one three times in ChatGPT and Perplexity. Count the appearances and write down the sources. It takes an hour and gives you a baseline more honest than any score.
Then see what the same data looks like without the manual work. LLM Search Console tracks your prompts across the major engines and shows who gets named when you do not.
Subscribe to the Substack newsletter and get one tested AI search tactic every Monday, short enough to use before your standup.