Start by dropping the idea of optimizing for "AI" as one thing. The engines read different sources, so the work is per-engine. Publish pages that answer the exact question a customer asks, get corroborated by third parties like reviews and a license number, keep your listings consistent, then measure each engine separately. There's no single switch.
Why can't I just optimize for AI once and be done?
Because "AI" isn't one destination. The engines cite strikingly different sources, so the same page can win on one and lose on another. Profound found ChatGPT and Perplexity overlap on only 11% of the domains they cite. Google's AI Overviews and Microsoft's Copilot overlap just 6%. Even Google's own AI Overviews and AI Mode cite the same URLs only 13.7% of the time. If two surfaces from the same company barely agree, there is no universal "AI SEO" checkbox. There is only per-engine work.
11%
Domain overlap between ChatGPT and Perplexity (Profound)
6%
URL overlap between Google AI Overviews and Copilot (Profound)
13.7%
URL overlap between Google's own AI Overviews and AI Mode (Profound)
What actually gets a contractor cited?
- Publish the page that answers the real question. If people ask what a slab-leak repair costs in Tampa, write the priced, city-specific answer in plain text an engine can quote.
- Get corroborated by third parties. Complete review profiles, a visible license number, and at least one mention beyond your own site give the engine evidence you exist and do the work.
- Keep listings consistent. The same name, address, and phone everywhere means the engine reads one coherent business, not three conflicting ones.
- Measure per engine. Ask each one the same real questions and track who it names, because a win on Gemini tells you little about ChatGPT.
Does schema markup get me cited in AI answers?
There's no evidence that it does. Plenty of blogs tell you to bolt on schema.org markup to win AI citations, but the one empirical test of this, from Search Atlas, found no correlation between schema coverage and how often a site got cited by LLMs. There's no peer-reviewed research showing schema causes citations. Schema can still help indirectly: cleaner markup can improve your classic search indexing, and that index is what several engines retrieve from. Just don't treat it as a citation lever, because nobody has shown it is one.
So what about llms.txt, everyone's talking about it?
Skip it as a visibility play. The pitch is that a special llms.txt file tells AI engines how to read your site, but no AI system currently uses it. Google's John Mueller compared it to the old keywords meta tag, a thing you could fill in that nothing actually read, and Gary Illyes said Google doesn't support it and has no plans to. If you want to add one because it's harmless, fine, but don't count it as work that moves your AI visibility, because on the record it doesn't.
How do I know any of it is working?
You measure, and you measure per engine over time. A single check on one engine is close to meaningless when ChatGPT and Perplexity share only 11% of their cited domains (Profound). The honest loop is: pick the real questions your customers ask, ask them on every engine, record who gets named and which sources get cited, publish the page that closes the biggest gap, then ask again. Anything that can't show you a before and an after is guessing.
