How NarraLoom Measures AI Search Visibility

Buyer questions come from real search-demand data. The questions in the scan are checked against the AI engines measured in that scan, and the captured results are stored before the report is rendered. Nothing is invented to fill a gap: if a question set is thin, the panel is smaller, and if an engine cannot be measured, that check is marked unmeasured rather than counted as absence.

Every finding in a NarraLoom report is scoped to that one scan. This page explains where the questions come from, what is tested, what is kept as evidence, and what a single scan cannot establish.

Where the buyer questions come from

Questions start from real search-demand data. NarraLoom uses advertiser-grade keyword and search-demand data to discover candidate buyer questions for the business and its market, then qualifies which of those questions belong in the scan panel.

NarraLoom does not write substitute buyer questions to reach a target count. If a market produces fewer usable questions, the panel is smaller. The number of tested questions varies from one scan to the next, and the report states the panel it used. For how agencies think about panel size, see how many buyer questions to test in an AI visibility audit.

What the scan actually tests

The audit works on the buyer questions in that scan's approved panel — no more and no less. Each question is checked against the AI and search engines measured during that scan.

NarraLoom can address Google AI Overviews, ChatGPT, Perplexity, Claude and Gemini. That does not mean all five are successfully measured every time. An engine may be unavailable, may not be captured, or may simply not be part of a given scan. The report describes what happened in that scan, not a general state of the internet. For why each engine is reported on its own, see measuring visibility separately across AI engines.

Present, not present, and unmeasured are different

Present. The brand was named in the captured answer for that tested question and engine.

Not present. The engine returned a usable measured response, and the brand was not named in that captured answer.

Unmeasured or not captured. No usable response was captured for that question and engine, so no presence or absence conclusion is drawn for it.

An engine failure, an unavailable engine, or an unmeasured check is never counted as brand absence. Those cells stay separate in the report so you can see exactly how much of the panel was measured.

What NarraLoom stores as evidence

A measured check can retain the tested question, the engine it was checked against, the captured answer evidence, any source or citation URL returned with that answer, and the UTC time of the check.

The request and returned response can be stored with integrity hashes, so the evidence can be tied back to what was actually captured. That is what makes a finding defensible in front of a client: the report is not a summary of an impression, it points at stored material.

Outside sources are not automatically competitors

AI answers often name sources other than the audited brand. NarraLoom can display the source and citation evidence an engine actually returned for a tested question.

A directory, a forum, a publisher, a social platform or any other domain surfaced in a scan is not automatically a direct competitor. Those sources are classified before anything is presented as competitive evidence, and the report describes them as outside sources surfaced in this scan.

How the report uses stored evidence

Measurement happens first; rendering happens afterwards. When a report is opened, it reads the stored measurement result for that scan — opening a report is not supposed to re-run any check against an engine.

Counts shown in the report come from the server-side measurement record rather than being recalculated in the page you are looking at. Two people opening the same report see the same numbers, because both are reading the same stored result.

What "not present in this scan" means

It means the brand was not named in the measured responses captured for the tested buyer questions represented by that report.

It does not mean the brand is absent everywhere in AI search, absent for every possible buyer question, absent from every AI engine, or absent permanently. It does not mean the brand was placed below another company, and it does not mean a customer, a visit or an order was gained or forgone. The report says what was found and what was not found in that scan, and stops there.

What this audit cannot tell you

One scan does not establish universal AI visibility, behaviour on questions that were not tested, or behaviour on engines that were not successfully measured. It does not predict future AI answers.

It is also not an official ranking, and it does not measure traffic, leads, orders or revenue outcomes, nor any causal effect on a category. Competitor coverage is limited to what appeared in the captured evidence, so it is never exhaustive. These limitations are stated in the report itself, not only here.

Why this method matters to an agency

The value is defensibility. When a client pushes back on a finding, you can separate what was tested, what was captured, what was not captured, which outside sources appeared, and what the evidence does and does not support.

That separation is the difference between an opinion about AI search and a record you can hand over. You are not asked to vouch for a number you cannot trace.