Something worth noticing happened this month: the "get cited by AI instead of clicked by Google" argument stopped being a niche take. It's now the headline advice from the most-followed people in SEO, pulling tens of thousands of likes per video. The ten blue links are being read to fewer and fewer people; the answer box does the talking. On the diagnosis, they're right — and it matches everything we measure.
Then comes the prescription, and it's almost always the same one: find every question people ask in your niche, and generate content answering all of them. Push the button, publish at scale. Conveniently, the button being demoed usually belongs to a tool the presenter owns. That's not an insight — it's an ad wearing an insight costume.
Why mass-generated answers don't get cited
Generative engines don't cite you because you published something with the right keywords in it. They compose answers from sources they already trust — and when we pull apart real AI answers, the pattern is consistent:
- Most of what gets cited is third-party. Review sites, directories, community threads, editorial listicles. The recommended business's own blog is a minority citation at best. Fifty generated posts on your own domain don't change what the machine trusts about you — they change your word count.
- The engines are actively learning to discount slop. Scaled, undifferentiated question-answer content is the most abundant text class on the modern web, which makes it the cheapest signal to ignore. Google's own spam policies now name "scaled content abuse" explicitly. Betting your visibility on the content category every engine is being tuned to filter is a strange bet.
- Recommendation is a trust decision, not a retrieval decision. A model names a business when the facts about it corroborate: consistent identity, readable site, independent mentions. A thousand self-published answers provide zero corroboration — they're all your word.
The advice is "be the answer." The tactic is "manufacture noise shaped like answers." Those are opposites.
What actually moves citations
The unglamorous, measurable version:
- Get mentioned where the engines already look. Find the third-party sources AI actually cites in your niche — they're visible in real answers — and earn your way onto them: reviews, directory profiles, genuine community participation.
- Be machine-readable. Unblocked AI crawlers, structured data, key facts that exist without JavaScript, one clean llms.txt. Boring, fast, and a real share of businesses fail it silently.
- Publish a few pages worth citing, not many pages worth ignoring. An answer-shaped page with real numbers, real dates, and a defensible claim can get cited. Its thirty generic siblings can't — and drag down the neighbourhood.
- Measure blind, before and after. Ask the engines your buyers' questions without naming yourself, record the answers word for word, make the changes, ask again. If a tactic can't show you movement in the actual answers, you didn't buy visibility — you bought an invoice for content.
A simple filter for the advice you're seeing
When the next "AI search changed everything" video crosses your feed, apply one test: does the advice end with a way to verify it worked? If the last step is "generate more," it's a pitch. If the last step is "look at what the machines now say about you," it might be a strategy.
That test is the whole reason Incisory exists. Run the free audit — we ask ChatGPT, Gemini and Perplexity the questions your buyers ask, blind, and hand you every answer word for word. Start from what's true, not from what someone's content tool needs you to believe.