AI visibility checker: what a free check can and cannot tell you

· by boneLau

Short answer: a free AI visibility checker asks one or two AI assistants one or two questions and tells you whether your brand appeared. That is a genuinely useful starting signal, and it is not the same thing as measuring your visibility. The difference is not the price — it is the denominator, the sample size, and whether anyone checked that the citation was actually yours.

I build one of these tools, so assume the usual bias. Everything below is checkable: you can run the same test in twenty minutes without paying anyone, and I say exactly what a free check can and cannot see.


What a free checker actually does

Almost every free checker in this category — ours included — works the same way:

  1. You enter a brand name, sometimes a domain.
  2. It sends one question containing your brand name to one or two models. Ours asks "What is [brand] and is it recommended?" through Perplexity, falling back to Claude if Perplexity fails.
  3. It reports whether the brand appeared, and usually lists any citation URLs.

That is the whole mechanism. It runs in seconds because it is one call, and it is free because one call costs a fraction of a cent.


The five things a free check cannot tell you

1. It measures recall, not discovery

The question contains your brand name. So the model is being asked "do you know this brand" — not "would you recommend something for this problem".

Those are different numbers, and the gap is large. On our own brand, the same week produced 17/18 mentions when the prompt already named BrandScope, and 1/26 when it did not. A free check would have shown the 17/18 and looked excellent.

Discovery is the number that predicts whether new buyers find you. A checker that only asks branded questions cannot see it at all.

2. One answer is not a sample

Ask the same question twice and models routinely name a different set of products. Published stability work on this is unkind: one study needed 33–94 repeats per prompt before a brand ranking settled.

A free check is n=1. It is a data point, not a rate. Treat it as "did anything happen at all", never as "this is my number".

3. It cannot tell you whether the citation is yours

When a checker returns a citation URL, the useful question is whose page is that? Three cases look identical in a URL list:

  • your own domain — you are the only thing feeding the answer
  • a third party writing about you — someone else is doing your explaining
  • a different company with the same name — this is not your win at all

We have hit the third case on our own brand: an answer that "mentioned BrandScope" was citing a same-name competitor's pricing page. A checker without ownership classification counts that as a mention.

4. It does not judge accuracy

Being named and being described correctly are different outcomes. A model that calls a monitoring platform an "SEO tool" has mentioned you and damaged you in the same sentence.

Judging accuracy needs your own positioning as the reference, which means a per-answer verdict — not a sentiment score.

5. It has no time dimension

One check is a snapshot. Visibility changes weekly, model versions ship, and retrieval shifts. Anything you cannot repeat on the same question set next week cannot show you a trend.


So what is a free check actually good for?

Three legitimate uses:

  • A zero-cost first signal. If the check says the model has never heard of you, that is real and worth knowing. It costs nothing to find out.
  • A sanity check before buying. Run a free check, then ask any vendor what their free number was and what denominator produced it. If they cannot answer, that tells you something.
  • A before/after anchor for one specific change. Not proof — an anchor. "Before the page existed it said X; six weeks later it says Y" is a useful story even though it proves nothing on its own.

When a free check stops being enough

You need repeated measurement, not a bigger free check, when any of these become true:

You need to knowBecause a check cannot
Whether new buyers can find youit never asks an unbranded question
Whether a change you shipped did anythingone snapshot has no noise band
Whether the citation is actually yoursno ownership classification
Whether AI describes you correctlyno per-answer accuracy verdict
Which engine is the problemone or two engines is not a comparison

Run the manual version first

You do not need a tool to see the gap. Twenty minutes:

  1. Write five questions a buyer would ask without naming your brand ("best [category] for [use case]", "how do I track what AI says about my brand").
  2. Ask them in ChatGPT and Perplexity, in fresh chats.
  3. Record: mentioned or not, at what position, which URL was cited.
  4. Do it again next week, same questions.

The first run usually produces a lower number than people expect. The full setup is here, and the difference between the two kinds of question is explained here.


Last updated 2026-11-03. Our own figures come from weekly runs on brandscope.dev (run 29708e8e, 2026-10-26, five core surfaces). Our free checker asks one branded question through Perplexity with a Claude fallback — stated here so you can weigh what it showed you.

Want your own baseline? Run one brand through the live question set — about 30 seconds, no account needed. Check your brand →

← All posts