How to measure citation share

Step 1 — Build the prompt set

Mix four prompt types on a stable 200-prompt set:

• Category discovery — "what's the best X for Y?", "top X tools", "X compared".
• Problem-first — "how do I solve Y?" without naming a category.
• Comparison — "X vs Y", "alternatives to X".
• Feature-specific — "which X supports Z?", "X with Z built in".

Weight the mix by how your buyers actually ask. Category discovery drives most raw impressions; comparison and feature prompts drive most late-funnel citations.

Step 2 — Sample every engine, every week

Run each prompt once per engine per week. Capture the verbatim answer text plus every cited source URL. Do not paraphrase, do not deduplicate at the prompt level — you need the raw record for auditing and confidence math later.

Step 3 — Parse the citations

Resolve every cited URL to a registered domain (root + eTLD). Handle subdomains, redirects and shortened URLs before matching. An answer "cites" your domain if any resolved source URL matches — count answer-level coverage, not raw link count.

Step 4 — Compute and label

CitationShare = answers_citing(you) / answers_analysed, computed per week, per engine, and pooled. Attach a confidence tier to every number based on the count of analysed answers — directional (<200), moderate (200–500), strong (500+).

See citation share (concept) for the definition and worked example, and methodology for the sampling and confidence details.

Frequently asked questions