Attribution rate

What is attribution rate?

Attribution rate (in the AI search / GEO sense, distinct from traditional marketing attribution) is a practitioner-coined metric that measures how often an AI search engine credits a specific source when answering queries about a topic. No vendor or academic literature defines a single canonical formula; the denominator and counting rules below are crystallized from GEO measurement practice, not from a published standard.

The formula:

Attribution Rate = (responses that cite the target source) ÷ (total evaluated responses) × 100%

Three denominator choices the practitioner must lock in advance and keep stable:

• Per engine vs aggregated: compute and report per engine first; an aggregated number across ChatGPT / Perplexity / Claude / Copilot / Google hides large structural gaps (see the Status section for the ~7-8% vs ~29% asymmetry).
• All probed responses vs only responses containing any citation: pick one and document it. Responses that cite at least one source is the stricter denominator and more comparable across engines that differ in willingness to cite at all.
• Verbatim citation vs source mention: tighter definitions require the engine to actually link the source; looser definitions count any name-mention.

A high attribution rate suggests the source has become a citation-default; the engine reaches for it reliably when the topic comes up. There is no industry-standard threshold yet; the underlying concept was formalized in the 2023 Princeton GEO paper¹ alongside related measurement methods.

Status in 2026

One of the most commonly used KPIs for GEO programs. No standardized formula yet. Practitioners differ on whether to count any mention (broadest), inline citation with a link (narrower), or top-three cited sources only (strictest). Profound², Otterly, and similar tools likely implement slightly different definitions internally; cross-tool comparison should not be assumed apples-to-apples without checking each tool's methodology page.

Worth tracking attribution rate per-engine, not aggregated: per Ahrefs' August 2025 long-tail study (15K queries), the cross-engine average overlap with Google's top 10 is ~12%, but that headline averages 5 measurements that bundle Perplexity's ~29% with the much lower ~7-8% for ChatGPT, Gemini, and Copilot³. Ahrefs lists query fan-out as one plausible explanation alongside personalization and Perplexity's independent index. AI Overview sits at 38-76% depending on Ahrefs' measurement window: their July 2025 study found 76%, and a February-March 2026 update on roughly 2× the data revised this to 38% (the drop reflects both a broader candidate pool and Ahrefs' own improved citation-parsing methodology). One attribution rate number across all 5 engines hides this structural asymmetry.

How to apply

Attribution rate is the headline KPI for any GEO program. The setup is the same regardless of which engines you target:

• Sample with discipline, not opportunistically: pick a fixed prompt set (start at 10 queries representing real audience intent; scale to 30+ for stable per-engine estimates), probe at fixed cadence (weekly is the typical floor) on a fixed engine list (ChatGPT, Perplexity, Claude, Copilot, and Google with Gemini chat, AI Mode, and AI Overview each probed separately and rolled up under Google for the rate calculation), and lock the region / login state where possible (Google AI Overview behavior differs across countries; incognito reduces personalization confounds). For each probe, record cited URL, domain, position in the source list, the answer's verbatim wording, and whether the citation is linked. 10 queries × 5 engines × 4 weeks ≈ 200 observations is typically enough to spot a real trend separate from week-to-week noise.
• Track per-engine separately: aggregate attribution rate hides the actionable signal. Your number may be 35% on Perplexity (well-cited) and 0% on ChatGPT (locked out). That gap is the next thing to fix.
• Anchor the denominator: attribution rate is (queries where your domain is cited) / (total queries probed). Keep the denominator stable across weeks; adding or dropping queries breaks comparability. If you must rotate, do it in monthly cohorts.

What to skip: composite "AI visibility scores" that bundle attribution rate, citation match rate, and brand mention rate into one number. They average away the per-engine gaps you actually need to see.

How it relates to other concepts

• Companion to citation match rate, which counts only linked citations (a stricter denominator).
• Distinct from citation share, which is your fraction of all citations for the same prompt set rather than the absolute frequency.
• Distinct from brand mentions in AI answers, which counts unlinked brand-name references and typically exceeds attribution rate for well-known brands.
• Output frame of cite-ability: cite-ability is the input property of the content, attribution rate is the measured outcome.
• Per-engine instance: AI Overview citation is the Google-specific attribution event; Microsoft Copilot citations is the Microsoft-side equivalent and the only one with a publisher-facing native dashboard as of 2026 (Bing Webmaster Tools "AI Performance").
• One of the most commonly used success metrics for Generative Engine Optimization programs.
• Denominator alignment: in the Citation vs Mention vs Link taxonomy, attribution rate counts the citation cells (linked + unlinked), not the mention cells. The strict definition counts only linked citations; the looser definition counts any inline source attribution. Whichever you pick, document it so the rate is comparable across measurement windows.