LLM citations
The sources an AI model names or links when it answers, the AI-era equivalent of a ranking.
In depth
What it really means
An LLM citation is any source a model references when generating an answer: a linked footnote, a named brand inside the sentence, or a quoted passage. In AI search, being cited is what ranking used to be.
Citations are also the only concrete thing you can measure about AI visibility. Everything else is inference. If assistants cite you consistently for your category, you are winning that surface. If they cite three competitors and a review site, you have a specific, fixable gap.
Three types, three different jobs
| Type | What it looks like | What earns it |
|---|---|---|
| Linked citation | A numbered source under a Perplexity answer | Ranking well plus a liftable passage |
| Named mention | “Tools like Zapier, Make and n8n…” | Category association built through third-party mentions |
| Quoted passage | Your sentence appearing near-verbatim | A self-contained, factual, well-written passage |
Named mentions matter most for B2B, because they are what puts you on a shortlist. They come from Reddit, G2, roundups and podcasts more than from your own site.
Citation share = (answers citing you ÷ total answers generated for your prompt set) × 100
Run the same 30 to 50 prompts monthly across each assistant. Track citation share per engine separately, because they diverge sharply. It is common to sit at 30% in Perplexity and 5% in ChatGPT for the same brand, and that gap tells you exactly where to work: Perplexity rewards ranking, ChatGPT rewards being talked about.
Pros & cons
Pros
- The one AI visibility metric that is countable rather than inferred.
- A citation carries implied endorsement that a search listing does not.
- Diagnosing which engine you underperform in points straight at the fix.
Cons
- Noisy. The same prompt can produce different citations across runs, so single readings mean nothing.
- Referral traffic is small even at high citation share, so this metric rarely defends itself in a traffic dashboard.
- Manual to collect unless you buy a tracking tool.
- Being cited inaccurately still counts as a citation, which is why sentiment needs tracking alongside it.
The mistake people make
Expecting citations to show up as traffic. They mostly do not. The visible effect appears elsewhere: branded search volume rising, demo requests where the prospect says they “heard about you,” and sales cycles that start further along. Track citation share as a leading indicator and branded search as the lagging one, and stop looking for it in your referral report.
Best practices
- Publish clear, quotable passages with facts and named sources attached.
- Keep every claim accurate, since contradiction across sources gets a source dropped.
- Earn third-party mentions on the sites assistants retrieve from and were trained on.
- Track citations across assistants monthly against a fixed prompt set.
- Study the queries where competitors get cited and you do not, then close that specific gap.
- Track sentiment alongside share, because a negative citation is not a win.
FAQs
What are LLM citations?
The sources a model names, links or quotes when it answers. In AI search, being cited is the equivalent of ranking.
How do I get cited by AI models?
Write liftable, factual, well-sourced passages, and earn mentions on the third-party sites those models trust. See GEO for the measured tactics.
How do I track AI citations?
Run a fixed prompt set monthly across the major assistants and log which sources appear. Tools exist, but a spreadsheet works for the first six months.
Why do citations matter if they send little traffic?
Because they shape the shortlist. The effect shows up as branded search and better-informed demo requests, not referral sessions.
Do citations replace rankings?
They add a surface. For AI-first researchers, citations are what count, and rankings still feed several of the engines directly.
Keep reading
Related on LymLyt
Beyond LymLyt
Further reading
Want this working on your site?
We build the content behind the term, ranked in search and cited by AI.
Book a 30-min call →