Written and maintained by @Concept211 17 live domains · 3,693 recommendations shipped

You get cited by AI when a crawler can reach the page, the answer sits in the served HTML, and one passage answers the question on its own. Everything else follows from those three. The work is mostly technical and mostly one-time per page, and none of it needs a bigger site or more links. The seven steps below are the order we work them in.

Does letting the crawlers in actually work?

Not always, and this is the step most sites get wrong without knowing. Between 9 August and 7 September 2026, across the 16 domains we manage, the share of requests we answered with a 403 ran from 0% for some AI crawlers up to 43.3% for others, on zones configured by the same team. Allowing a crawler in robots.txt and actually serving it a 200 are two different things, and only one is visible from inside your own repo. The full per-crawler table is in what AI crawlers actually fetch.

What are the seven steps?

  1. Allow the AI crawlers, then verify from outside. Add GPTBot, ClaudeBot, PerplexityBot, OAI-SearchBot and Google-Extended to robots.txt, then fetch your own pages as each one and confirm a 200 rather than a challenge. A crawler that is refused at the edge cannot cite you no matter what the rest of the page does.
  2. Serve the answer in the HTML. The text an engine would quote has to be in the response body, not painted in after JavaScript runs. Most AI fetchers do not execute your scripts, so a client-rendered answer reads to them as an empty page.
  3. Phrase each heading as the question the section answers. Convert about a third to a half of your H2s and H3s into the wording a reader would actually type or say. An engine matches a section to a spoken query through its heading, so a heading that is a label gives it nothing to match.
  4. Make each passage answer its question alone. Open the section with a two to three sentence answer that carries its own subject, its own numbers and its own timeframe. Engines lift passages rather than pages, and a passage that depends on the paragraph above it breaks when it is lifted out.
  5. Say who you are in markup. Emit Organization and Article nodes with a sameAs graph pointing at profiles you control. Entity clarity is how a model decides a claim belongs to an identifiable publisher rather than to nobody.
  6. Give engines permission to quote you. Set a permissive max-snippet, keep published and modified dates honest, and publish an llms.txt. A restrictive snippet limit caps quotations below the length a citation needs.
  7. Be the primary source for one thing. Publish something only you can publish: your own measurements, your own tests, a number nobody else holds. A page that summarizes what already ranks gives an engine no reason to name it instead of what it summarized.
The short answer
AI citation is decided in this order: can the crawler reach the page, is the answer in the served HTML, and does one passage answer the question on its own. Schema, snippet limits and first-hand data raise your odds after those three are true. They cannot substitute for them.

What is not on this list?

Domain age above roughly six months, and backlink count above a few dozen referring domains. Both matter for being considered at all and neither predicts citation once you clear the floor, which is why a small site with one original page competes here in a way it cannot on classic rankings.

Where this leaves you

Work them in order and stop at the first one that fails, because the later steps cannot compensate for an earlier one. Step one is the cheapest to check and the most often broken.

Read the full version: how to earn citations in AI search covers each signal in depth, with how to measure whether it moved.

Source for the refusal figures: first-party Cloudflare bot logs across 16 ConceptSEO-managed domains, 9 August to 7 September 2026, reported in full in what AI crawlers actually fetch.