What Citation Share means
Define a stable set of relevant questions, run them on a schedule across each platform, and record linked sources separately from unlinked brand mentions. State whether the reported proportion uses observed answers or counted citations as its denominator.
The prompt set has to stay fixed for the number to mean anything. Changing questions between reports makes month-over-month comparison meaningless, which is a common way agencies flatter their own results.
Why it matters for AI search
Without it, AI SEO is unfalsifiable. Screenshots of favourable answers prove nothing — models return different answers to the same question asked twice, vary by account and region, and change with updates.
Citation share, run consistently with variance measured, converts that noise into something you can act on. It's less precise than rank tracking and honest reporting says so.
See it in action
Same prompt, ten runs. Variance is the thing most reporting hides.
A citation appearing once in ten isn't the same as nine in ten.
Reporting a single run as a result is the most common measurement error in AI SEO. Multiple observations show variation; they do not automatically establish statistical confidence.
How to get it right
Measuring it properly
- Build the prompt set from real query data and sales conversations, then freeze it
- Run each prompt multiple times per cycle so variance is measured rather than ignored
- Track competitors across the same set — your share only means something in context
- Record how you're described, not just whether you appear; being cited with wrong facts is worse than absence
- Report ranges and confidence rather than false precision
Citation coverage and competitive share of voice
State the denominator before comparing visibility. Citation coverage can mean the proportion of a fixed set of observed answers containing a link to your site. Mention coverage counts answers naming the brand, whether linked or not. These are different measurements.
For example, if 10 of 50 recorded answers link to your site, linked-answer coverage is 20%. If those answers contain 40 counted brand citations across you and a fixed competitor set, and 10 are yours, your citation share within that competitor set is 25%. Label both calculations so a reader knows which one is being reported.
Competitive share of voice adds context: your absolute mentions can rise while competitors gain faster. Keep the questions, competitor set, platforms, locale and sampling method stable; an answer can name more than one brand, so answer-coverage percentages need not add to 100%.
Report answers naming nobody, repeat observations to show variance and keep mentions, linked citations, website visits and revenue separate. Share of voice is not proof that a particular answer caused a sale.
Common questions
How many prompts should we track?
Typically 40 to 60 for a focused category. Enough to cover the real question clusters, few enough to run consistently at multiple repetitions.
Very broad categories may need more; the constraint is whether you can run it reliably every cycle.
Can citation share be gamed?
Easily, if the prompt set changes between reports or if only one run per prompt is recorded. Both make the number look better than reality.
That's why we freeze the set, publish the methodology, and hand over the raw data.
Does citation share correlate with revenue?
A change in citations alone does not establish a revenue effect. AI-assisted research may later appear as direct or branded traffic, and attribution is incomplete.
We triangulate with branded search lift and self-reported source data. It's decision-useful, not precise, and anyone claiming precision is overstating it.
Related concepts
These come up alongside Citation Share constantly.