What an engagement actually looks like

Week 1 — Prompt set and competitor list

We start from your language, not ours. Sales call recordings, support tickets, lost-deal notes and existing search data produce a candidate list. We cut it to the prompts a real buyer would type, then freeze it.

A mid-market B2B set typically lands between 40 and 120 prompts, split across problem awareness, solution evaluation and vendor comparison. You approve the set and the competitor list before any sampling begins, because both determine every number that follows.

You receive: the frozen prompt set, the competitor list, and the coding rules we will apply.

Weeks 1–2 — Baseline sampling

Each prompt runs across the platforms in scope, from consistent locale settings, repeated to account for non-determinism. We record the full response text, cited sources, brand presence, how you are characterised, and which competitors appear.

You receive: the raw sampling records. Yours to keep, whatever happens next.

Week 2 — The baseline report

Results are reported per platform. We do not average them into one number, because platform overlap is low enough that a blended figure hides the actionable detail.

Week 3 — Prioritised gaps

Gaps are ranked by commercial value, not by how quickly they can be fixed. The list includes work we recommend against, with reasons. In most engagements that section is longer than clients expect.

You receive: a ranked action list with owner, effort, expected effect and the confidence level attached to that expectation.

Months 1–6 — Engineering and re-sampling

The work itself is mostly search work done properly: coverage of the concepts a query family expects, entity clarity, technical access for the crawlers you have chosen to allow, and pages that answer the question directly in the opening lines.

Re-sampling runs on a fixed cadence against the same frozen prompt set. Traditional search and AI visibility are reported separately throughout.

What we tell you when it is not working

Platform retrieval changes without notice, and a number can move for reasons that have nothing to do with our work. When we cannot attribute a change, we say so and label the claim Observed rather than Verified. If a programme is not producing, we would rather tell you in month three than defend it until month twelve.