The method

How the audit actually works

There is no proprietary score and no black box. The whole thing is designed so you can reproduce it yourself, which is also the point: if you can check our work, you do not have to take our word for anything.

1. 01

   Building the question set

   Everything starts from what your buyers actually type, which is rarely what a keyword tool would suggest. We take your ICP, your positioning and, where you have them, real sales-call objections, and turn them into 25 bottom-funnel prompts.

   Four rough shapes: category picks (best X for Y), head-to-head comparisons (A vs B), alternative-seeking (X alternatives), and practical qualifiers (implementation speed, integrations, certification timelines). These are the questions asked by someone close to a decision, not someone learning the category.

2. 02

   Running them and capturing sources

   Each prompt is run individually against the tools your buyers use: ChatGPT first, then Perplexity, Gemini and Google AI Overviews. Individually matters, because batching prompts changes how models answer them.

   For every answer we record the full response text, every brand named, the order they were named in, and, where the tool exposes them, the sources cited. The citation list is the important part: it is the map of which pages the model trusts on this topic.

3. 03

   Scoring the gap

   Visibility becomes a per-question score against one named competitor, chosen because they are the company already taking the answers you want. Not a vague index, a count: how many of the questions named them, how many named you.

   Then the diagnostic part: for each question you lost, which specific pages earned the citation, and what those pages have structurally that yours do not. That is what makes the roadmap concrete rather than a list of best practices.

4. 04

   Fixing it

   Work goes where the citations come from. In practice that is usually some mix of comparison and alternatives pages that actually answer the comparison, direct structured answers to the specific questions being asked, presence on the third-party surfaces the models lean on for category questions, and the technical basics of being cleanly crawlable and parseable.

   It is not content volume. Publishing more of the wrong thing does not move citations.

5. 05

   Re-measuring

   The same question set gets re-run on a fixed schedule. Same prompts, same tools, same scoring, so the number is comparable over time rather than a fresh opinion each month.

   If it is not moving, the report says so. A measurement you can only ever pass is not a measurement.