How to get cited by ChatGPT and Perplexity

By Seedcite. Published . 5 min read.

No one can guarantee a citation, and no verified formula exists. The reasonable approach is to write clear, sourced, answer-first pages, keep the content as crawlable text, and measure results with a fixed set of prompts. A top-ten Google ranking does not guarantee an AI citation.

Illustration of a plant with lines leading to three source rows, two marked as cited and one open.
Illustration. Three answer sources: two cited, one still open.

One large central plant joined by thin stems to a ring of smaller plants.

Can you actually control whether ChatGPT or Perplexity cites you?

Not directly, and nobody can promise you a citation. These engines choose sources at answer time, and the choice varies between runs. What you can control is how easy your page is to find, read and quote, and whether you measure the outcome honestly.

That framing matters because many guides present citation tactics as settled. For the factors people argue about most, such as structure, schema, freshness and mentions on other sites, we found no verified effect sizes. Everything below is a set of reasonable bets, labeled as such.

What you can control is how easy your page is to find, read and quote, and whether you measure the outcome honestly.

Does ranking on Google get you cited?

Sometimes, but it is not a guarantee. A top-ten ranking does not ensure an AI citation. Reported overlap between cited pages and top-ranking pages differs widely by engine and by study: about 12% for conversational assistants from Ahrefs and about 17% from BrightEdge, with Perplexity at 28.6%, and one report of AI Overview citations matching the top ten 76.1% of the time in July 2025 but only about 37.9% by March 2026 (almcorp.com). All of these figures are secondary and unverified.

The safe reading is that classic SEO is still a good foundation, and that you should check citations separately instead of assuming one follows from the other.

What can you change on your own pages?

The research behind GEO suggests that how a page presents information can change how often it is used in an answer. The Princeton-led study reports gains of up to 40% on its benchmark, but we could not verify its per-tactic figures, and it is a benchmark rather than proof about current assistants (arXiv 2311.09735). With that caveat, these are bets worth testing.

  • Lead each section with a one or two sentence answer a system could lift out intact.
  • Use question-style headings that match real prompts.
  • Include specific, sourced facts, and name the source in the sentence.
  • Keep the content in plain visible text rather than in images or scripts.
  • Make it obvious who you are and what you do: a clear name, a clear offer, a clear location if you serve one.
  • Add an FAQ section with visible text, and only then matching FAQ structured data.

What does a citation-ready paragraph look like?

It answers the question in the first sentence, names its source, and then adds context a reader can use. Compare two openings for a page about pricing a service.

A weak opening: "Pricing is a complex topic and every business is different, so let us explore the many factors involved." It contains nothing a system could quote.

A stronger opening: "Most small agencies price retainers by the scope of monthly deliverables, not by hours, and review the figure each quarter." It is a complete claim, it can stand alone in an answer, and you can attach a source or your own data to it. The second version is only worth publishing if it is true for you, so write what you really know instead of imitating the shape.

Notice what the stronger version does not do. It does not stuff keywords, and it does not invent a statistic to look authoritative. Fabricated numbers are the fastest way to lose a reader's trust.

Do structure, schema and freshness help?

We do not have verified evidence either way. Google states that its AI features need no special markup, no llms.txt and no special schema, and that structured data should match visible text (Google Search Central, AI features). Reported data suggests most llms.txt files receive no AI-crawler requests at all (ppc.land, unverified).

That does not make clean structure pointless. It makes it a hygiene step with an unproven payoff. Do it because it helps readers and search crawlers, and judge any extra citation benefit only by your own measurements.

How do you test whether it worked?

Use a fixed prompt panel, the same approach described in our guide to generative engine optimization. The design is guidance rather than a sourced standard, because we found no accepted method.

  1. Record a baseline first: run 50 to 200 fixed prompts on each engine several times before you change a page.
  2. Log the cited URLs and any brand mentions from every run.
  3. Change one thing at a time, such as adding a direct answer under one heading on a handful of pages.
  4. Re-run the same prompts on a schedule, and compare share of voice, meaning your citations as a fraction of all citations in your competitor set.
  5. Report run counts and variance with the result. If the change is smaller than the run-to-run noise, call it unproven.

What should you skip?

Anything that tries to fake the signals. Fake reviews, undisclosed bots and fake engagement are banned patterns on the platforms that matter, and Reddit requires approval for API access and expects automated accounts to be labeled (Reddit Responsible Builder Policy). A tactic that depends on deceiving a platform will not hold up, and it can damage the brand you are trying to build.

Also skip the idea that a single trick, whether a file, a schema type or a keyword phrase, is the answer. The honest position is that useful, clearly written, well-sourced pages plus careful measurement are what we can recommend today.

Sources

  1. Checked 2026-10-20Google Search Central: AI features and your website https://developers.google.com/search/docs/appearance/ai-features
  2. Abstract onlyAggarwal et al., GEO: Generative Engine Optimization (arXiv, KDD 2024) https://arxiv.org/abs/2311.09735
  3. Unverified, secondaryalmcorp.com on AI Overview citations vs. top-ranking pages https://almcorp.com/blog/google-ai-overview-citations-drop-top-ranking-pages-2026/
  4. Unverified, secondaryppc.land on llms.txt adoption https://ppc.land/llms-txt-adoption-rises-8-8x-but-97-of-files-get-zero-ai-requests/
  5. Per Seedcite research notes, 2026-10-20Reddit Help: Responsible Builder Policy https://support.reddithelp.com/hc/en-us/articles/42728983564564-Responsible-Builder-Policy

Keep reading