SEO & AI Search

The thing reading a small business website in 2026 is increasingly not a person. It is a retrieval system assembling an answer, deciding whether to name a business, whether to link to it, and whose page to quote instead.

The default SEO audit returns a 200-row export of issues and fixes whatever is red. That process fails completely against an answer engine, because none of those rows measures whether the business actually gets cited.

The Four Diagnostic Failures

Ranking and Citation Have Come Apart: A page can sit at position 1 and still lose every AI answer to a forum thread, invisible to rank tracking.

Measuring Noise: AI-visibility tools sample once and report a tidy percentage, when these systems rarely answer the same question the same way twice.

Unsourced Advice: Almost none of the AI-optimisation advice in circulation separates what a platform documented from what a vendor asserted.

No Priority: An audit that lists forty findings has ranked nothing and told the client to spend on all forty.

The AI Search Audit

A four-tier diagnostic engagement answering one question older audits cannot: can an answer engine retrieve this business, resolve who it is, and actually cite it, and if not, what are the three cheapest things that change that.

Four Tiers, One Question Each

Retrieval: Separates the four retrieval agents that answer engines actually cite from the three training crawlers that only consume bandwidth, catching misconfigurations a flat robots.txt scan would miss.

Citation, Measured Honestly: Refuses to report an AI-visibility rate from fewer than three sampled runs, after two runs of the identical prompt seconds apart named different winners and shared zero cited sources.

On-Page and Technical: Diagnoses whether the page earns the click it already ranks for, and whether a crawler can even read it, in that order.

Three Fixes, Not Forty: Every tier returns exactly three prioritized fixes with evidence and effort attached, plus an explicit list of what is already fine.

The Verdict

On one live engagement, the audit found a business named by the AI in 21 of 30 samples but linked in only 11 of 30, with Reddit cited more than the business's own site on its highest-anxiety questions. All four tiers, measured independently days apart, converged on the identical finding without ever reading each other's conclusions.

The full four-tier engagement, retrieval through AI-citation sampling, ran for $0.26 in paid API spend. The architecture is not specific to SEO: it works for any business whose category is answered by forums, and for any consultancy selling into a field full of confident, unsourced advice.