🔍 New: AI Search Trail — see exactly where AI found (or lost) your brand. See how it works →
/ resources · reference

How citation scoring works

What the score means, how we calculate it, and what you should actually care about versus what you can safely ignore.

§01What “cited” means

A citation is counted when an AI model’s response to a query explicitly names your brand or domain in a positive or neutral recommendation context. We don’t count:

  • Mentions in a negative comparison (“unlike X, avoid Y”)
  • Generic category mentions without a specific brand name
  • Responses that name your brand only as context for recommending a competitor

Each query is run three times per model to account for stochasticity. A citation is counted as “consistent” if it appears in at least 2 of 3 runs. This filters out one-off hallucinations and reflects what a real user would reliably see.

§02The score calculation

Your overall citation score is a weighted average across all queries and models:

# citation score formula
score = (citations / total_query_model_pairs) × 100

# example: 12 queries × 4 models = 48 pairs
# if you were cited in 18 of 48 pairs:
score = (18 / 48) × 100 = 37.5%

The score is intentionally simple — it’s a percentage. The complex part is the query set. We run your domain against buyer-intent queries in your category, not vanity queries like “what is [your brand name].” Getting cited on branded queries is easy and doesn’t tell you much.

▸ important nuance

A 40% score on hard buyer-intent queries beats an 80% score on easy branded queries. The queries matter as much as the score.

§03Per-model breakdown

Beyond the overall score, we show you your citation rate broken down by model. This matters because the models have meaningfully different citation behaviors (see Understanding the AI engines we check).

A brand that scores 80% on Perplexity but 15% on Claude should interpret that very differently than a brand that scores 40% across engines. Perplexity cites broadly; Claude cites selectively. Being cited by Claude at all puts you in a small group.

We also show which specific queries you were and weren’t cited on. The uncited queries are your highest-leverage improvement targets — they represent buyer-intent moments where you’re invisible.

§04What to optimize for

Don’t optimize for your overall score in the abstract. Optimize for specific uncited queries first.

If you’re not cited on “best [category] tool for [use case]” — that’s a content gap. The fix is usually building the page that most directly answers that query: a comparison page, a use-case landing page, or a structured explanation of your positioning in the category.

The overall score is a lagging indicator. The query-level breakdown is where you find the levers.

§05Score benchmarks

citation score rangewhat it signals
80–100%Citation monopoly — you are the default answer in your category
50–80%Strong presence — consistently mentioned, competitive with leaders
25–50%Visible but inconsistent — cited on some queries, absent on others
10–25%Weak presence — showing up only on easy branded queries
0–10%Effectively invisible — the AI conversation is happening without you

Median score across the B2B SaaS brands we’ve checked: 31%. Most brands are visible but inconsistent. The gap between them and category leaders is almost always content clarity, not brand authority.

See how your brand actually scores.

Run a free ChatGPT sample on your domain. Email + one-time code; no credit card for the sample.

Run a free check →