01Methodology
How we measure, and where the measurement stops
Searlixa measures AI visibility by sampling a fixed set of buyer prompts across each AI platform on a repeating schedule, then reporting citation presence, share of voice and source overlap per platform. This page describes exactly how that works, including what it cannot tell you.
Definition
AI visibility is the rate at which a brand or its pages are cited, named or recommended inside AI-generated answers, measured across a defined set of prompts, platforms and dates. It is a sampled statistic, not a rank position.
Why sampling, not ranking
Traditional rank tracking works because a query returns a broadly stable ordered list. AI answers do not behave that way. The same prompt can return different sources on consecutive runs, and the sources differ sharply between platforms.
Published research in 2026 found that the two most similar AI platforms shared only 24.1% of the pages they cited. Every other pair overlapped less. A single reading from a single platform is therefore evidence of very little. We treat each observation as one sample and report the distribution.
The four stages
Build the prompt set
We take the language your buyers actually use, from sales calls, support tickets, search queries and competitor comparisons. Typically 40 to 120 prompts across awareness, evaluation and decision stages. The set is written down and frozen, because a moving prompt set makes trend data meaningless.
Sample the platforms
Each prompt is run across the platforms in scope, on a fixed cadence, from consistent locale settings. We record the full response, the cited sources, whether your brand appears, how it is described, and which competitors appear alongside you.
Code the results
Each response is coded for brand presence, citation of your own domain, sentiment of the mention, and competitive context. Coding rules are documented and applied consistently, so a different analyst would reach the same numbers.
Report with status
Every figure carries an evidence label, a sample size and a date range. Traditional search and AI visibility are reported separately. We do not blend them into a single score, because the two are measured in fundamentally different ways.
What we report
| Metric | Definition | Known limitation |
|---|---|---|
| Citation presence | Share of sampled prompts where your domain is cited | Sensitive to prompt wording; only valid against the frozen set |
| Brand mention rate | Share where your brand is named, cited or not | Unlinked mentions cannot be verified against a source |
| Share of voice | Your presence relative to a named competitor set | Only as good as the competitor list you agree at the start |
| Source overlap | Which pages are cited, and by which platforms | Platforms attribute inconsistently; some cite nothing at all |
| Answer sentiment | How the model characterises you when it names you | Coded by humans; borderline cases are logged, not smoothed |
Evidence labels
Every number we publish carries one of five statuses. They appear on this site, in client reports and in tool output.
- Verified Direct client data or a primary source, with permission to publish.
- Observed Measured by Searlixa. Method, sample size, date range and platforms are stated.
- Estimated Modelled or inferred. The assumptions are stated in plain language.
- Demo Illustrative only. Not real data.
- Updated Materially revised since first publication, with a note on what changed.
What this method cannot tell you
We would rather state these up front than have you discover them in month three.
- It is a sample, not a census. A 60-prompt set describes those 60 prompts. It does not describe every conversation your buyers have with an AI assistant.
- Attribution is incomplete. A large share of AI answers name no source at all. Mentions inside those answers are real but unverifiable from the outside.
- Platform behaviour changes without notice. A shift in retrieval can move your numbers without anything changing on your site. We annotate the timeline when we detect it, and we say so rather than claiming credit.
- Causation is hard. We can show that citation rose after a change. We can rarely prove the change caused it. Where we cannot, we label the claim Observed rather than Verified.
- Referral data understates the channel. Analytics only sees clicks. An answer that names you without a link still influences a buyer, and no dashboard records it.
Reproducibility
Clients receive the prompt set, the raw sampling records and the coding rules. If you want to re-run the measurement yourself or hand it to another firm, you have everything needed to do so. That is deliberate. A measurement you cannot audit is not a measurement.
02Next
Read the rest of the working system
Research publication standard · A worked example, end to end · The audit engagement · Authors and reviewers