What operators do to get cited by ChatGPT, Perplexity, and AI Overviews (2026)

We asked operators a blunt question: what actually got you cited by ChatGPT, Perplexity, and Google's AI Overviews, and what turned out to be snake oil? Seven who are showing up in AI answers told us. Their answers converge on one uncomfortable idea. There is no GEO trick. You get cited when you are the clearest, most credible, most machine-legible source on a specific question, and you get ignored the moment you try to shortcut that.

What actually moved the needle

Be the primary source, and make it machine-legible. The most-repeated move was the least clever one: publish something worth citing, then make it trivial for a crawler to parse. Ritwick Dey of Panto AI put original datasets, reproducible benchmarks, and clean schema at the center, because "AI overviews prefer citeable, primary material over marketing pages." Timothé Merle of Hikoo rebuilt pages around the literal question a user asks, answering it in the first sentence, because "AI engines extract passages in isolation, so the cleanest direct answer wins, not the highest-ranking page." George Hartley of Nitrosend took it one step further and built for the agent first, every capability exposed as an API endpoint and an MCP tool, so the company shows up where people now do the asking. As he put it, the operators showing up in AI answers "made themselves legible to agents, not just readable by humans."

Be a source the models already trust. Ranking your own page is not the same as being cited. David Arato of Lexicon Legal Content framed it the most plainly: AI citation "follows the same logic as editorial citation," and you get cited "when you're a credible, specific, attributable source." That means earning references from the third-party places a model already reads, and publishing content with genuine information gain, not the same thing every other page says.

Remove the friction. Josh Qian of LINQ Kitchen found that the lift came less from adding signals than from removing conflicting ones. Cleaning up contradictory service descriptions, inconsistent location data, and missing visuals raised the confidence the engines had in summarizing him. Consistency, not volume.

What was snake oil

The contributors were just as specific about what did nothing. Bolting on FAQ schema and dropping in an llms.txt file expecting magic: no measurable lift. Keyword density, which is irrelevant once the model is reasoning over meaning instead of matching strings. Mass-produced AI-written posts, which get flagged as AI and read as filler, so nothing quotes them. Submitting to low-authority AI directories and scraper registries, swapping link badges, and chasing raw brand-mention volume. Josh Qian named the trap directly: chasing authority metrics "by swapping guest posts for faster posts without updating any real service messaging or measurable call outcomes."

The through-line

Two operators landed the same point from opposite ends. Kajol Shah of Budventure Technologies: "AI citation is not just about being indexed. It is about becoming the clearest, most structured source on a specific problem." Sonam Patel of Swappsy: "The real competitive advantage isn't a tactic. It's having something worth citing."

That is the whole finding. The work that gets you cited by an AI answer engine is the same work that earned an editorial citation before the engines existed. Be a primary source, say something specific, make it easy to read for both a person and a machine, and remove the noise that makes a model unsure about you.

How this report was made

This is itself a worked example of the thesis. We posted a research brief, operators responded on the record with named, specific points of view, we scored each response for fit and originality, and we cited the strongest by name with a link back to their company. Several of the people quoted above named exactly that motion, contributing a real, named expert quote to a place AI engines read, as one of the few plays that reliably worked. The generic blends away. The specific gets quoted.