The method, fully documented.

Most "AI SEO" is repackaged content marketing. Ours is field work: log what the engines actually say, trace why, change the inputs, and re-run the same prompts to prove the delta. Every step below produces a document you keep.

1. Build the prompt set (days 1–2)

   We draft 40+ prompts the way your buyers actually phrase them — category shortlists ("best X for Y"), competitor displacement ("alternatives to Z"), use-case queries, pricing comparisons, and reputation checks ("is [brand] legit?", "[brand] reviews"). You review and add the prompts your sales team hears on calls.

   prompt log (yours to keep)

2. Run the engines (days 2–4)

   Each prompt runs across ChatGPT, Perplexity, Gemini, Claude and Google AI Overviews. We capture the full answer, the brands named, the order they appear in, the sources each engine leaned on, and a screenshot — because answers drift and you deserve evidence, not summaries.

   visibility ledger, run 01 — spreadsheet + screenshots

3. Score share of answer (day 4)

   Your share of answer is the percentage of prompts where you're cited, weighted by position. We score it per engine and per prompt cluster, alongside the three competitors who currently own the answers. This is the baseline every later claim is measured against.

   share-of-answer scorecard

4. Trace the citation sources (days 4–6)

   Engines don't invent shortlists; they assemble them from pages they trust. We extract every source URL behind your category's answers and sort them: review platforms, comparison listicles, community threads, docs, and news. The result is a map of exactly which doors you need to walk through.

   source map — ranked entry points

5. Diagnose the blocks (days 6–8)

   Why are you absent? Usually some mix of: the engines can't resolve your entity (name collisions, thin schema, no Wikipedia/Wikidata trail), your site has nothing extractable to quote (no comparison pages, no plain-language claims, no llms.txt), or the sources engines trust have never heard of you (no review profiles, no earned mentions).

   diagnosis memo, in plain language

6. File the fix worksheet (days 8–10)

   Everything becomes a ranked worksheet — effort, impact, and owner for each line. Typical items: entity & schema corrections, a citable comparison page engines can quote, llms.txt and crawler access, two review-platform profiles, and the three earned mentions most likely to enter the citation pool. Audit clients can run it themselves; Sprint clients have us run it.

   fix worksheet — ranked, owned, dated

7. Re-run and report the delta (Sprint, monthly)

   Same prompts, same engines. We report movement as a before/after delta on the original scorecard, with fresh screenshots. If a number moves, you can verify it yourself in any chat window — that's the standard we hold every claim to.

   delta report + updated ledger

A note on honesty: answer engines are probabilistic and their answers move. What we guarantee is the work, the evidence, and the founding-client terms on the rate card — not a specific sentence from a specific model. Anyone promising that is selling weather control.