AI search optimization: be the answer, not the tenth blue link.

AI search is what happens when someone types a question into an assistant and the assistant goes and looks. ChatGPT does it through Bing. Google AI Mode does it through Google. Perplexity does it through its own crawl and shows the citations. In each case a model reads a handful of sources and writes one answer.

Classic SEO optimises for a page of ten results and a human who compares them. AI search optimization optimises for one paragraph and a machine that has already compared them. The practical difference is that hedged, everything-we-do marketing copy is useless here. A model needs a stated fact on a page about that fact.

The volume shift is real and it is measured. BrightLocal found 45 percent of US consumers now use ChatGPT or another AI tool to find local business recommendations, up from 6 percent a year earlier (BrightLocal, 2026). Ahrefs reported AI search visitors converting 23 times better than organic search visitors on its own site (Ahrefs, Jun 2025). Small traffic, high intent.

LLM SEO is what people call this when they are thinking about the model rather than the search box. It is the same work. A language model cannot recommend what it cannot read, cannot verify, or cannot describe without contradicting itself.

There is one idea worth separating out, because vendors blur it. Getting into a model's training data is not something you can buy or schedule, and nobody who claims otherwise can show you proof. What you can influence is what the model finds when it looks something up live, which is where every local answer comes from anyway. That is the part we work on, and the part we can screenshot.

Each tool grounds its answer somewhere, and knowing where is most of the strategy.

• ChatGPT searches with Bing, so Bing Places and clean Bing indexing decide whether you exist in the source material at all.
• Google AI Mode grounds in Google's index and Google Business Profile, so profile completeness, categories, hours and review volume feed directly into what it says.
• Perplexity attaches citations to its answers, so pages that state a fact plainly get cited and vague pages get passed over.

Measurement is the part most vendors skip, because it is the part that can prove them wrong. There is no rank tracker for an answer, so we do the obvious thing instead. Ten buyer questions, agreed in writing and frozen for 90 days. Run every 30 days in a fresh logged-out ChatGPT session, in Perplexity and in Google AI Mode. Dated screenshots and a count. You can repeat any prompt yourself the same day and get the same picture.

Local businesses whose customers ask a question before they choose, and where a single new customer is worth hundreds or thousands of dollars. The higher the value of one customer, the faster this pays for itself.

• Med spas and aesthetics clinics: our deepest market, with first-party data published city by city.
• Every industry we work in: the buyer question we target for each trade.
• Not a fit: national ecommerce, and markets where a hospital system or national chain already owns the answer. We check before we quote.