The fastest free way to find out whether AI has heard of your company is to open a fresh session with no account history and ask it what your company is, using only the name. If the answer is blank, hedged, or about somebody else, you have an identity problem rather than a content problem, and that distinction changes everything you would do next.
This is a narrower test than a full visibility audit and it answers a more fundamental question. Not whether you win category questions, but whether the system knows you exist as a distinct thing at all. Everything else is downstream of that.
The test, exactly
Four rules, and each one exists because skipping it produces a false result.
| Rule | Why it matters |
|---|---|
| Fresh session, no logged-in account | An account that has searched your company before will surface it for you and not for a stranger |
| Memory and personalisation off | Otherwise you are measuring your own history rather than the engine |
| Company name only, no context | Adding your category tells the engine the answer you are testing for |
| Repeat across four engines | Each keeps its own corpus, so one engine is one data point, not a result |
Ask it plainly. What is [company name]. Nothing else. Then record the answer verbatim rather than summarising it, because the specific wording is what tells you which failure you have.
Five things you might get back, and what each means
| Response | What it means | What to do |
|---|---|---|
| Nothing, or a refusal to guess | No usable record. You are not an entity yet | Identity work, starting with the about page |
| A different company entirely | Your name resolves to somebody with more evidence | Disambiguation, and it will take months |
| Vague and category-generic | It has fragments and cannot assemble a confident picture | State the six identity facts in one place |
| Accurate but outdated | It knows you, from an older version of you | Correction, which is publishing rather than fixing |
| Accurate and current | Identity is fine. Your problem is elsewhere | Move to category questions, not this test |
The row that surprises people
Accurate but outdated is the most common result for companies past their first repositioning, and it is the one teams misread as success. An engine confidently describing a product you discontinued is worse in a buying conversation than an engine saying nothing. Our free AI visibility check records the exact wording across engines so you can see which version of your company is circulating.
Why a blank answer is not a content problem
The instinct on seeing nothing is to publish more. It rarely helps, because the failure is upstream of content quality.
A system needs to recognise your name as naming a thing, find candidate records for it, and pick the right one. A blank answer usually means the second step failed, which is to say no reference record exists anywhere it can consult. Publishing another page on your own domain adds no candidate record, because your own site is the source it was already unable to use confidently.
What adds a candidate is presence somewhere structured and independent. Business registries, funding databases, review platforms, and the structured knowledge bases that knowledge graphs are built from. That is a different task from writing, and it is usually a shorter one.
The follow-up question that isolates it further
If the first answer was blank or wrong, ask one more thing in the same fresh session. Ask who the main competitors are in your category, without naming your company.
If the engine names four competitors fluently and has never heard of you, the category is well documented and you specifically are not. That is good news, oddly, because it means the corpus covers your space and the gap is yours to close rather than a structural absence.
If the engine struggles with the category too, you are in a thinly covered space. Harder in the short term, and considerably more winnable, because being the first well-documented company in a category is a position that compounds.
What this test cannot tell you
Worth being clear about the limits, because a five-minute test oversold becomes a bad decision.
- It does not tell you whether you appear for buying questions. That needs unaided category prompts, which is a different test.
- It does not tell you why. A blank result is consistent with being blocked at the crawler, rendering empty, or simply being undocumented.
- It is one run. Answers vary, so treat a single blank as a signal to look further rather than a finding.
- It says nothing about whether you would be recommended. Being known and being chosen are separate outcomes.
Used as a triage step it is excellent. Used as a report it is an anecdote with a percentage attached.
What to do about each of the five results
Knowing which result you got is only useful if the next step differs, so here is what each one actually implies.
Nothing at all. Do not write more content. Get a reference record to exist somewhere structured and independent, which usually means business registries, a funding database, review platforms and any partner who documents integrations with you. Then publish the six identity facts in one place on your own site so the records agree with each other.
A different company. This is the slowest fix and the one worth deciding about deliberately, because sometimes the right answer is commercial rather than technical. If a much larger organisation owns your name, disambiguation may take years, and adding a consistent qualifier to how you describe yourself everywhere is cheaper than winning the name outright.
Vague and generic. The system has fragments and cannot assemble them confidently. This is the most fixable result. Put the six facts in plain prose on one stable page, make the structured data agree with it, and remove the competing versions scattered across old profiles.
Accurate but outdated. Publishing new material is necessary and not sufficient, because the old version still exists. Find and update or remove the pages carrying the stale claim, including third-party profiles you control, then re-test monthly. Stale facts decay slowly and they do decay.
Accurate and current. Stop here and go and test category questions instead. Identity is not your constraint and further work on it is wasted.
Run it on a competitor too
The single most informative version of this test takes another two minutes. Run the identical question against your closest competitor in the same fresh session.
If they come back described richly and you come back blank, the corpus covers your category and the gap is specifically yours. That is a solvable problem with a known shape, and it tells you roughly how much documentation is required to be legible in your space.
If they come back thin as well, the whole category is undocumented. That is slower going and considerably more valuable, because being the first company in a category that machines can describe confidently is a position that compounds and that nobody else is currently competing for.
Do this for three competitors and you have a rough map of how well documented your category is, which is the context that makes your own result interpretable rather than just discouraging.
How often to repeat it
Monthly is enough for this specific test, and more often than that mostly measures variance rather than progress. Record the date and the verbatim answer each time in the same place.
The value is in the sequence rather than any single reading. A company moving from blank to vague to accurate over four months is making real progress that no single check would show, and that sequence is also the only evidence you will get that identity work is doing anything, since none of it appears in a traffic report.
One caveat on interpreting a bad result, because it is easy to over-read. A blank answer about a young company is normal rather than alarming. Machines describe what has been documented, and a company that is two years old with limited coverage will read as thin because it genuinely is thin, not because anything is broken. The test is diagnostic, not a verdict on the business.
What changes the reading is time. The same blank answer at a company with eight years of history, customers and press coverage means something is actively wrong, usually at the crawler or with a name collision. Age is the variable that turns this result from expected into urgent.
If you take one thing from this page, make it the fresh session. More self-assessments are wrong because somebody tested from their own logged-in account than for any other reason, and the failure mode is the worst kind, since it tells you everything is fine. An engine that has watched you search your own company for two years will happily show it to you and to nobody else. Two minutes in a private window is the difference between a measurement and a reassurance.
If this comes back badly, the full diagnostic walkthrough works through all five causes in the order worth checking, and the one-prompt category test covers the other half of the picture.
Related reading
- What is an entity. The three steps a name has to survive before an engine will treat it as a thing.
- Indexed, retrieved, cited. Which of the three stages a blank answer actually points at.
Five minutes now saves a quarter of guessing.
Run it in a fresh session, write down the exact wording, and repeat it monthly. That sequence is the only evidence you will get that identity work is doing anything, because none of it shows up in a traffic report.
If the answer came back blank or wrong and you want to know which of the causes is responsible, send us the domain for a free AI visibility check.
Shaban runs answer engine optimisation for B2B SaaS companies at Uncited Brands. He works the unglamorous end of the problem, mostly crawler access, retrieval diagnostics and measurement baselines, and publishes the tests that do not go his way alongside the ones that do.