Skip to Content
How AI assistants cite your site

How AI assistants find and cite your site

When an AI assistant answers a question and names your site as a source, that citation is the end of a short journey: the assistant had to find your page, read it, trust it enough to use, and then attribute it. Understanding those four steps tells you exactly what to work on to be cited more often.

This article explains the mechanics. For the bigger picture of why it matters, start with What is GEO?. To act on it, see Improving your Agent Readiness score.

The four steps to a citation

1. Find

An assistant reaches your content in one of two ways:

  • From its training or index — it already knows your site from when it was built or last refreshed. You influence this slowly, by publishing clear content over time.
  • By fetching live — many assistants now browse the web while they answer, sending an AI agent (a crawler or fetcher) to read pages in real time. You can see these agents in the AI Visibility tab, under AI agents.

If an agent cannot reach your page — it is blocked, it times out, or it needs JavaScript the fetcher does not run — you cannot be cited from it. Being reachable is step zero.

2. Read

Once an agent has your page, it has to make sense of it. Assistants strongly prefer content that is:

  • Structured — real headings, lists, and tables, not text baked into images.
  • Server-rendered — visible in the raw HTML, not assembled by JavaScript after load. Many fetchers do not run scripts, so a JavaScript-only page can look empty to them.
  • Self-contained — a clear answer near the top, so the assistant does not have to guess what the page is about.

3. Trust

Assistants weigh whether your content is safe and credible to repeat. Things that help:

  • A clear identity — who published this, and what is the site about.
  • Machine-readable signals — structured data and a published guide to your content (an llms.txt file) make your site legible to an assistant. The Agent Readiness score checks for these.
  • Consistency — content that agrees with what the assistant sees elsewhere is safer to cite than a lone outlier claim.

4. Attribute

Finally the assistant decides whether to name you. Some always link their sources, some only sometimes, and some strip the reference entirely. This last step is the one you control least — which is why the first three matter so much.

What you can and can’t measure

Citations are real but slippery, so Voris is careful about what it claims:

  • The AI Citations tab watches the answer engines for the specific questions you track and records when you are cited — with a confidence band and a directional / measurement label, so a figure is never dressed up as more precise than it is.
  • When a citation leads to an actual visit, that person shows up in the AI traffic channel. But assistants often strip the referrer, so some AI-driven visits arrive with no trace and fall into Direct — see the AI Visibility tab troubleshooting notes for why.

In short: you can measure citations and the visits they drive, but treat the numbers as a reliable direction, not a perfect count.

What to do about it

Every step above comes down to the same handful of fixes — be reachable, be structured, be machine-readable, be trustworthy. The Agent Readiness score turns that into a concrete checklist you can work through.

What is GEO?Improving your Agent Readiness score