LLM and AI search visibility tools
One account, one question, 19 readings - and 2 different model tiers answered without being asked.
A reading that does not record which model produced it is reporting an uncontrolled variable. The tier served to our account drifted mid-run, so we record it per reading and never mix same-tier and cross-tier comparisons into one number.
2
model tiers served to a single account across 19 readings, without being requested
What we measured
- Across 19 readings from one account, 2 different model tiers answered without being requested, so every reading has to record which model produced it.
- We report same-tier and cross-tier comparisons separately and never pool them: 15% mean overlap within a tier, 20% across tiers.
- Of the 17 domains cited across the 2 questions on this page, only 2 appear on both - this market has almost no consolidation.
- On the two questions this page answers, 17 distinct domains are cited between them - a tool market with almost no consolidation.
The variable most tools do not record
We set out to measure how stable an assistant's shortlist is. Before we could, we had to deal with something simpler: on a free account you cannot choose which model answers you, and the tier drifted during our run.
That single fact governs the whole measurement. If a reading in January was served by one tier and a reading in February by another, a difference between them may be the model, not the market. So the tier is recorded on every reading, and the headline number is computed on same-tier pairs only.
For the record, both numbers: same-tier repeats of one identical question overlapped 15% on average; cross-tier repeats overlapped 20%. We publish both and mix neither.
Why this matters more for LLM visibility than for search
A search engine's ranking does not change because your session was routed differently. An assistant's answer can, and the routing is invisible to the person asking.
That makes three things mandatory in any honest reading: record the model, record the account type, and never pool surfaces. Anonymous, logged-in, free, paid and API sessions are different instruments pointed at the same subject, and a number that averages them is not measuring the subject.
It also sets a floor on how many readings a claim needs. One reading on an unrecorded tier supports no claim at all.
The competitive picture on these two questions
The two questions this page answers cite 17 distinct domains between them, with barely any overlap: one rank tracker and one automation directory appear on both, and everything else appears once.
That is what an unconsolidated market looks like. It is also, for a small player, the encouraging shape - the questions are not owned by anybody yet.
The questions this page answers
ai search visibility tool
880/mollm visibility tool
720/mo
2 questions, 1,600 searches a month between them. All 2 answers we examined on them were produced by a web search, which means a page can reach every one.
Who an assistant cites on these questions today
- nightwatch.io 2
- zapier.com 2
- aiclicks.io 1
- amplitude.com 1
- llmclicks.ai 1
- visible.seranking.com 1
- rankability.com 1
- reddit.com 1
- tryprofound.com 1
- withgauge.com 1
17 distinct domains are cited across these questions; the 10 above are the most frequent, with the number of answers each appears in. Counted from the answers captured in our corpus, not from a market census.
What to ask before you buy
- Ask which model and which account type produced each reading. If the tool cannot say, its trend line has an uncontrolled variable in it.
- Ask whether surfaces are ever averaged together. They should not be.
- Ask to see two readings of the same question taken close together. If they differ and the tool reports a single state, you have learned what its number is worth.
Honest limits of what we measured
- The tier drift was observed on a free account. Paid and API access may behave differently and we have not measured them.
- 19 readings across 5 questions in one trade, on one assistant. Directional, not definitive.
- The 17-domain count comes from the answers captured in our corpus for these two questions, not from a market census.
Questions we get asked
Does using a paid account fix this?
It may pin the tier, and we have not tested it. Until someone measures it, the safe practice is the same: record what answered, and never assume.
Why publish the cross-tier number at all?
Because hiding it would be a choice made after seeing the data. We publish both figures with their definitions and let the reader see that we did not pick the flattering one.
Is this the same as AI visibility or AEO tooling?
Same measurement, different vocabulary - and, in our corpus, a different set of cited competitors. That is the only reason these are separate pages.
What is the minimum honest reading?
Several readings of the same question, each stamped with its model tier and account type, with the raw answers kept. Anything less is a screenshot.
Start with one free reading
One business, real questions, the raw answers back - before you decide on anything paid. If you are running this for a book of clients, say so and we will talk terms rather than quote a page price.
Free reading
Takes about a minute.
Related
Pricing is per category, because the work differs by how competitive the answers are. See every category we measure and what a reading costs →
Published 31 August 2026. Every figure on this page is recomputed from raw observation files by a single script, so any of them can be traced back to the readings behind it. We make your site citable. We do not promise placement in any assistant’s answer - we measure whether you are in it, and show you the raw answers either way.