Skip to content
AVE StudioFor marketing agenciesENHU

Cookbook

How to get cited by Perplexity: a retrieval-side checklist

Citability is a question of format rather than authority — and the worst outcome is being quoted without being named.

Updated ·

Sourced surfaces, of which Perplexity is the best known, work on different logic from closed-book answers. What matters first is not what the model retained from training but whether, at answer time, it finds a page it can lift a claim from and attribute. That is good news, because it responds faster than parametric memory: your fix can take effect after the next crawl rather than after the next model generation. It is also bad news, because citability imposes formal requirements that well-written marketing copy routinely fails. One thing worth knowing upfront: per the vendor's documentation the indexing bot does not collect for foundation-model training, while the user-initiated fetcher generally ignores robots.txt — so access and citation are two separate questions here.

The procedure

  1. Confirm you are crawlable at all

    Check that your robots.txt permits the indexing bot and that your content appears in server-rendered HTML. If the substantive text only assembles after client-side execution, a fetcher can easily see an empty page. This step looks trivial, which is precisely why it gets skipped — in practice the most common silent failure is not content quality but text that is not there when it is asked for.

  2. Write liftable claims

    A citable sentence stands on its own: it has a subject, an assertion, and where possible a number or a date. "We offer a market-leading solution" cannot be lifted, because in isolation it asserts nothing. "We measure how many of ten runs name the brand, and report it with an interval" survives extraction. The practical test: read the sentence with no surrounding context and ask whether it still says something.

  3. Source what you assert

    Sourced surfaces favour content that itself cites. When you state a figure or an external fact, say where it came from. This buys you two things: it makes the claim checkable, and it turns your page into a node rather than a dead end. For your own measurements, describe the method too — what you measured, how many times, when — because a number with no method carries no more weight than an adjective.

  4. Make freshness visible

    State when the content was last updated in a form a human can read, and keep it in sync with the machine-readable date. Do not game it: advancing a date without changing the content buys nothing in the short run and costs credibility over time. When something really is out of date, the honest move is to update it or say it has expired, not to rewrite the timestamp.

  5. Keep the identity coherent

    A citation benefits you only if the surface can attach the page to your brand. That requires the organization name, the canonical URL and your official profiles to say the same thing everywhere. If the page's markup points at a non-existent identifier, or the name appears in a different form page to page, the content stays citable — but it stops reinforcing you.

  6. Measure ghost citations separately

    The least-run step here, and the source of our most expensive lesson. Check not only whether you are cited but whether your name appears in the TEXT of the answer. The two do not travel together. In our own measurement, a category question returned our research cited in five runs out of five while our company name appeared not once — our own data was supporting the field's shortlist, without us on it. The usual cause is content that never states in its visible text who produced it.

Frequently asked questions

What determines which page gets cited?
It cannot be known from outside with certainty, and anyone offering a formula is claiming more than they can know. What is observable: pages with short, self-contained, sourced claims get picked up more often than long general prose. That is a pattern rather than a rule — look at what the surface actually cites in your category.
Do I need llms.txt or special markup for this?
There is no evidence that you do. The documented requirement is crawlability; everything else — llms.txt, extra markup — is not, as far as anyone can currently show, a precondition. Markup has its own value on the entity side, but citability is decided mainly by the shape of your text.
Should I block the bot if I don't want to be trained on?
Separate the two. Per the vendor's documentation the indexing bot does not collect for foundation-model training but for surfacing content — so blocking it does not prevent training, it removes your citability. If training is your objection, the training user-agents are what to address, separately.
How quickly do changes show up?
Retrieval-side fixes are measurable in days to weeks, because after a re-crawl the next answer works from the new state. That is the substantive difference from parametric memory, where you wait for a new model generation. Nobody can promise an exact turnaround, though.
My content is cited but my name isn't. What now?
Name yourself in the VISIBLE text. An author or publisher field in the markup is not enough — answer generation does not necessarily work from it. Put a sentence in the prose saying who ran the measurement or analysis and in what role. One sentence is plenty, but it has to be where the surface reads.

Sources

  1. Perplexity — Crawlers (roles of PerplexityBot and Perplexity-User). https://docs.perplexity.ai/docs/resources/perplexity-crawlers
  2. OpenAI — Bots (separating retrieval from training user-agents). https://developers.openai.com/api/docs/bots
  3. Anthropic — crawler documentation (the search user-agent's role). https://support.claude.com/en/articles/8896518-does-anthropic-crawl-data-from-the-web-and-how-can-site-owners-block-the-crawler