Concept

Citation share

Citation share is the share of AI-generated answers, across a fixed prompt set, whose cited sources link to your domain. It's the closest measurable proxy for how often AI engines treat you as evidence.

TL;DR
Citation share = (answers citing your domain) / (total answers analysed), across a fixed prompt set and engine list. Bounded 0–100%, comparable week over week, and directly tied to which pages need to earn AI citations next.
Bounded
0–100%
of analysed answers on your prompt set
Category leader
8–15%
on a competitive balanced set
Real signal
≥3%
you're a considered source
Invisible
<1%
not cited enough to influence answers

The definition

For a prompt set P and engine list E, citation share for domain d is:

CitationShare(d) = answers_citing(d) / answers_analysed

An answer "cites" a domain when at least one of its cited source URLs resolves to that domain (root or any subdomain). Multiple citations of the same domain in one answer still count as one citing answer — we count answer-level coverage, not raw link count.

Worked example

You track 75 prompts across ChatGPT, Perplexity, Gemini and Claude. Each prompt is sampled five times per engine during the monthly scan — that's 1,500 analysed answers. Of those, 90 include a link to yourdomain.com in their sources.

CitationShare = 47 / 800 = 5.9%

If a top competitor sits at 12.4% on the same set, they're being cited roughly 2x more often. Every point of citation share you close is worth roughly 8 additional answers per scan that now name you as evidence.

Citation share vs share of voice

Two different questions:

  • Share of voice — "does the answer text mention my brand name?"
  • Citation share — "does the answer link back to my domain?"

Both matter and they don't move together. Reviews and comparison sites drive mentions; primary documentation, research pages and named product docs drive citations. See citation share vs share of voice for a full breakdown.

Mentions tell you who's talking about you. Citation share tells you who the AI trusts as a source.
The evidence metric

How to grow it

Four levers, in order
  1. Step 01
    Fix the pages that already almost rank

    Find prompts where a competitor cites a domain you rank near on Google. Rewrite those pages for the exact question the prompt asks.

  2. Step 02
    Publish primary evidence

    Structured comparisons, benchmarks, methodology pages and named-customer numbers get cited more than marketing copy.

  3. Step 03
    Earn Wikipedia and analyst coverage

    Wikipedia, Gartner/Forrester/G2 pages, and major trade press are cited disproportionately — especially by Claude and Perplexity.

  4. Step 04
    Re-measure weekly

    Track which prompts newly cite you, which don't, and which pages are being cited. Ship the next round from the gaps.

See the full measurement playbook: how to measure citation share. Pair it with the methodology reference for sample sizes and confidence tiers.

Frequently asked questions

What is citation share?+
Citation share is the percentage of AI-generated answers, across a fixed prompt set and engine list, whose linked sources include your domain. It measures how often AI engines cite you as evidence — not just mention your name.
What is citation share vs share of voice?+
Share of voice counts brand name mentions in the answer text. Citation share counts source links to your domain in the answer's citations. A brand can be mentioned without being cited, and cited without being named. Both matter.
What's a good citation share metric to target?+
Category leaders sit at 8–15% citation share across a balanced prompt set. Anything above 3% on a competitive category is a real signal. Below 1% means you're effectively invisible as a source.
How is citation share ai measured?+
For each prompt, we sample the answer from every tracked engine and parse the cited URLs. Citation share = (answers citing your domain) / (total answers analysed). It's bounded 0–100% and comparable week over week.
How often should I measure citation share?+
Weekly for active tracking, daily for launch and crisis periods. Confidence tiers depend on the count of analysed answers — 500+ answers gives a strong read.

Keep reading