How AI assistants choose which companies to recommend — and how to get recommended

AI assistants recommend companies that meet three conditions at once: they exist in the search index as an authority on the topic, they are mentioned by sources the model trusts (directories, publications, reviews), and their website contains machine-readable data about who they are, what they do and where they operate. If any one of these three is missing, the AI will recommend a competitor — even when you are objectively the better choice. This is the essence of AEO (Answer Engine Optimization) and it differs from classic SEO in that you are not targeting a position in a results list, but a mention inside a generated answer.

Perplexity combines its own crawler with the Bing index and always shows citations — meaning the source is technically verifiable. ChatGPT with search uses Bing plus its training data, so companies mentioned on large publications have an advantage even without recent indexation. Google AI Overview pulls directly from the Google index and depends most on classic SEO signals — so if you are not on page one for a query, you are probably not in the AI answer either. Claude and Gemini combine similar sources. The conclusion: without solid SEO there is no AI visibility — AEO builds on SEO, it does not replace it.

Structured data (schema.org JSON-LD) is the only part of a website AI reads without interpretation. When your site explicitly states the organisation type, business area, services offered, geographic area covered and contact details, the model does not have to guess from prose — it gets facts. Key types for a service company: `LocalBusiness` or `ProfessionalService` with an `@id` anchor, `Service` on each service page, `FAQPage` for questions and answers, and `BlogPosting` with a publication date. In July 2026 Feather Studio implemented all of these types with linked entities — one of the few agencies in Serbia with a complete entity graph.

AI most often cites four formats: direct answers to a question in the first paragraph under a heading (the model looks for a ready-made sentence it can lift), concrete numbers and ranges (prices, timelines, percentages — because they are verifiable and useful), comparisons in "X versus Y" form (because that is how users ask), and structured lists of steps or criteria. What AI almost never cites: marketing copy without data, generic phrases about "innovative solutions", and pages that hide the answer behind three paragraphs of introduction. You write for a machine looking for a fact, and a human reads it looking for clarity — fortunately, that is the same text.

A three-month plan: month 1 — add complete schema.org markup (LocalBusiness with @id, Service per service, FAQPage), rewrite H2 headings on your most important pages into questions and put a direct answer immediately underneath. Month 2 — open profiles on platforms AI cites: Google Business Profile, LinkedIn Company Page, Crunchbase, Clutch.co and Wikidata (a free entity that models read directly). Month 3 — publish content answering concrete questions from your industry with real numbers, and pursue one guest appearance on a relevant publication. Measurement: ask Perplexity and ChatGPT your target queries before and after, and track growth in branded queries in Search Console.