AI optimization (AIO): make every AI system describe your company accurately, and recommend it when it fits

How the work is done

Ask the assistants first

The audit begins with questions, not crawls. We ask ChatGPT, Gemini, Claude, Copilot, Perplexity, and Google’s AI features what your company does, what it sells, what it costs, where it is, who runs it, and how it compares with the alternatives, several times each, and record every answer. Then we trace each wrong or hedged answer back to the source that produced it: your own page that says it in prose but not in markup, a directory listing from three years ago, a press release about a product you discontinued, or a competitor with a similar name. The result is a list of facts, a list of errors, and for each error the place it came from. That list is the roadmap.

One fact set, signed by you

Before anything is published we write the canonical fact set: every name, description, category, price, location, person, date, and identifier you want every system to hold, in the exact wording you want it held. You review and sign it. It sounds bureaucratic and it is the step most companies have never done, which is why their facts have drifted. From here on, every piece of structured data, every profile, and every mention is checked against this document, and every change to it is dated.

Machine-readable on your site

Models can read prose, but they trust markup more, because it removes inference. We put the canonical facts into Organization, Product, Offer, and Person structured data, link every profile you control with sameAs, state the same facts in plain prose on the about page and a corrections page, and publish an llms.txt describing the site for AI crawlers. This is technical SEO work with an unusual audience: not a ranking algorithm, but a model deciding whether it knows enough to answer without hedging.

The same facts where the models read

An assistant that finds your fact stated identically on your site, your business profile, two industry directories, a review site, and a trade publication states it with confidence. One that finds three versions hedges. So the next stage is alignment: the same facts, worded the same way, on the sources models draw on for your category, with disambiguation wherever you share a name with something else. Where independent coverage is thin, our link building team earns it through digital PR and editorial outreach that states the facts we want held, whether or not a link comes with it. We do not edit Wikipedia for clients, post reviews, or create profiles under other names; each is a policy violation on the platform concerned and each would put the rest of the work at risk. If your company already meets Wikipedia’s notability bar, we will say so and point you to the proper route.

Corrections and the two clocks

Some answers change within weeks of the source changing, because the assistant retrieved them. Some are baked into the model and change only with the next model version. The monthly report separates the two, so you know which corrections have landed, which are waiting on a source, and which are waiting on a release, and we file corrections through each platform’s feedback route where one exists and log the outcome. Alongside accuracy we track share of recommendation: for a fixed set of prompts where your product is a fair answer, how often each assistant names you rather than a competitor. If the honest answer is that the assistant is right to name the competitor, the report says that too.

AIO is the entity layer of our AI search programme, next to answer engine optimization for conversational engines, generative engine optimization for Google’s AI features, and LLM optimization for long-horizon model knowledge. The differences are laid out in SEO vs AEO vs GEO vs AIO vs LLMO vs SXO. To see what the assistants say about you today, start with the audit, or see pricing.