Guides · 2026-08-28 · By VSNARY | Emmanuel Orta
Free AI visibility checkers: what the free tier actually measures
Most of them are lead magnets, which is fine — a lead magnet can still be a good instrument. The question is whether the check is hard enough to fail, and a surprising number are not.
Free checkers in this category are almost all lead magnets. That is not a criticism — a lead magnet has to be genuinely useful or it does not generate leads. The useful question is what the check actually does, because the category has a specific failure mode: a scorer that almost everyone passes.
The score that cannot fail
A large free checker was observed grading a site 99% on citation readability, with the verdict that agents could access all content — and the same vendor then selling the fix. Elsewhere, an AI-visibility tool graded a site 100% and still surfaced remediation to buy.
Both numbers were arguably correct. That is the point. Access is close to solved for competently built sites, and free tools are giving that check away. Within a year, most sites that are trying at all will score in the high nineties on it. A metric everyone passes carries no information, and a report built on it has to manufacture urgency somewhere else.
So the first question to ask a free checker is not what did I score but what would have made me fail. If there is no plausible answer, the score is a formality.
What the free tiers typically test
| Check | Common in free tiers | Fails on |
|---|---|---|
| Does a page render for a crawler | Yes | almost nobody |
| Is schema present and well formed | Yes | a moderate share |
| Is there an llms.txt | Sometimes | 18.3% |
| Is a machine file served stale | Rare | 30.9% |
| Is any answer engine refused | Rare | 3.4% |
| Is the payload mostly code | Rare | 13.7% |
| Is the response a challenge page at 200 | Very rare | 1.9% |
The pattern is consistent: the checks that are cheap to implement are the ones almost nobody fails, and the checks that find real defects require fetching the bare URL as several identities and comparing — which most free tools do not do, because it costs requests.
The generic-score problem
A second failure is subtler. Many graders run an identical checklist against every site and penalise absences that do not apply. A documentation site and a plumber's five-page site are scored against the same rubric, so the plumber loses points for lacking things no plumber needs, and the score becomes a measure of how closely the site resembles the rubric's author's assumptions.
Worse is a rubric that scores dimensions it did not measure. One widely circulated framework assigned a reputation score out of twenty to every entity while its own third-party review field read not evaluated here on every one. That is what happens whenever a rubric has more dimensions than the evidence behind it — which is most rubrics, including ones that get sold.
What a free check has to do to be worth running
Four properties, all cheap to verify from the outside.
It fetches the bare URL. Adding a cache-buster asks the origin a fresh question and bypasses the object a crawler would actually receive. This single detail separates tools that can find the most common defect in the dataset from tools that structurally cannot.
It reads the content type, not just the status. A robots.txt returning 200 with HTML is parsed as allow-everything.
It varies identity. One request as a browser and one as a named crawler, compared on status and byte count. Without this, edge refusals are invisible.
It includes a control. A request to a path that cannot exist. If a site answers a nonexistent URL the same way it answers its homepage, you are measuring a firewall, not a website — and any score produced through it describes someone else's product. That case fires on 1.3% of scans, and it is the difference between a report and a fabrication.
What a refusal should do to a score
One behaviour separates a careful free tool from a careless one, and it is worth testing deliberately.
A scanner once graded a named company 15 out of 100, an F, reporting no sitemap and a robots.txt that did not return 200. All of it was false. Every path requested — homepage, pricing, robots, every sitemap variant, and a URL that could not exist — returned a small challenge response from the host's bot wall. The scanner had measured a firewall and published a grade about a company.
The correct handling is to return no score at all: mark the scan unmeasured, state why, and let the grade be null. A refusal is not a bad result — it is the absence of a result. Any tool that converts we could not reach this site into a low number will eventually publish a false accusation about somebody, and you have no way of knowing whether the number it showed you was one.
Test it: point the tool at a domain you know is behind an aggressive bot wall and see whether it returns a low score or an honest refusal.
Reading any result honestly
Whatever tool you use, hold two things. A single scan is one reading from one vantage point on one day: a deterministic block reproduces perfectly, so running it twice from the same machine is not corroboration. And an audit is a snapshot — across 48 sites measured daily for fifteen days, 68.8% carried a defect on the last day while 93.8% carried one at least once.
A free check that tells you what it requested and what came back is useful regardless of who is selling it. A free check that returns only a grade is asking you to trust a number in exchange for your email address.
Every figure above came out of this scanner.
Point it at your own domain and see the same measurements, free.
Questions this post answers
Are free AI visibility checkers any good?
Some are, but the category has a specific failure: a check almost nobody fails. Access is close to solved for competently built sites, so a high score on renderability carries little information. Ask what would have made you fail; if there is no plausible answer, the score is a formality.
Why did a tool grade my site 99% and still sell me a fix?
Because the thing it measured is genuinely fine and the thing it sells is something else. That is not necessarily dishonest, but it does mean the score and the recommendation are about different subjects, and the report has to create urgency somewhere the measurement did not.
What should a free scan check that most do not?
The bare URL rather than a cache-busted one, the content type rather than the status code, at least two different client identities compared on status and size, and a control request to a path that cannot exist.