Citation Rate vs Citation Share: How to Read AI Visibility Metrics

Citation rate answers one question: in how many of the answers we checked did this engine cite you?

The formula is simple. Count the answers that cite your domain or brand. Divide by the answers that actually came back. If you tracked 50 prompts on one engine and 12 answers cited you, your citation rate is 24%.

The denominator matters more than people expect.

• Answers returned, not prompts requested. A failed or empty response is missing evidence, not a clean miss. Counting it as zero understates you.
• One engine at a time. Google says AI Overviews and AI Mode "may use different models and techniques" and return different links. A rate pooled across engines hides which one is failing.
• Your prompt set. The rate only describes the questions you chose to ask. Engines rewrite prompts into their own search queries, so your set is a proxy for the real input.

Citation rate is about you alone. Your competitors can double their citations and your rate stays the same. That makes it the cleaner number for tracking your own progress.

Citation share answers a different question: of all the citations handed out across these answers, what fraction went to you?

The formula: your citations divided by total citations in the answer set. If 200 citations appeared across your tracked answers and 20 pointed to you, your share is 10%. It's the metric most tools sell as AI share of voice.

The denominator here includes everyone else. That changes what the number means. Share moves when you change, and it also moves when competitors, aggregators, Reddit threads or news sites change.

Microsoft reports a metric called Citation Share in Bing Webmaster Tools' AI Performance report. Microsoft frames it as "an observational metric", not a ranking system or a competitive scoreboard. It does not expose competitor domains or represent traffic share. Our definition of citation share covers how to read that version, and the Bing AI Performance report playbook walks through the rest of that report.

Read the two together. Rate tells you whether you're showing up. Share tells you how crowded the room is.

Share is zero-sum. Whoever already wins the query sets the size of the denominator.

AI citations are concentrated at the top. A study of AI Search Arena data (arXiv:2507.05301) found the top 20 most frequent news sources accounted for 67.3% of all citations for OpenAI models. It measured a Gini coefficient of 0.83 for OpenAI, 0.77 for Perplexity and 0.69 for Google. That data covers March to May 2025, so treat the exact values as dated. The pattern is the point: a few large sources take most of the pool.

Here's a hypothetical. Say you run a mid-sized accounting software brand.

• Month one: 200 total citations across your tracked answers, 20 of them yours. Share: 10%.
• Month two: engines cite more sources per answer, and a large competitor publishes heavily. Total citations rise to 400. You get 30. Share: 7.5%.

You gained 50% more citations. Your share fell by a quarter. A report showing only share would call that a bad month.

The reverse also happens. Share can rise because answers got shorter and cited fewer sources, while you gained nothing. Without the raw counts, you can't tell the difference.

Both numbers count citations. A citation is not proof of influence.

• Cited is not the same as used. Mechanistic research on how models cite (arXiv:2606.28358) repaired over 90% of missed citations by adjusting internal model components without changing the sources. Citation output is partly separate from which source shaped the answer.
• Citation units differ by engine. Anthropic's web search tool cites text spans with short quoted passages. Perplexity cites whole sources by index. One "citation" on each engine is not the same thing, so summing them into one figure mixes units.
• Retrieval is discretionary. Engines decide per query whether to search the web at all. An answer with no live search cannot cite you, whatever your site does.
• Prompt trackers sample. In Otterly's 2026 experiment, Microsoft's own report counted 42,817 citations for one test site, while Otterly's prompt-based tracker logged 3,934. The time windows may not match. It still shows that any prompt set is a sample.

The same experiment shows why per-engine reporting matters. Of the 3,934 tracked citations, 2,718 (69%) came from Microsoft Copilot, while ChatGPT, Gemini, AI Overviews and AI Mode almost never cited the site. A blended score would have hidden that split.

A useful AI SEO report shows its working. Follow these steps.

1. Report rate and share together, per engine. Never blend engines into one figure without also showing each one.
2. Print the denominators. Write "cited in 12 of 48 answers returned" next to 25%. Write "20 of 200 citations" next to 10% share.
3. Show raw counts over time. A trend in absolute citations beside a trend in share exposes the zero-sum effect.
4. Compare like-sized brands. Brand size is the biggest uncontrolled factor in AI citation data. Benchmarking a challenger against a household name reports brand size, not performance. Our guide to comparing AI visibility against competitors covers choosing a fair peer set.
5. Hold the prompt set constant. Changing prompts changes both denominators. Version your prompt list.
6. Prefer first-party counts where they exist. Bing's report shows the grounding queries Copilot actually ran. As of the August 2026 evidence review, that data had no API, so export it by hand.

If you want to build this without a tool, the spreadsheet method for AI share of voice and the step-by-step share of voice playbook both keep denominators visible.

Pick the metric that matches the decision in front of you.

Use citation rate when

• You're checking whether a fix worked, such as unblocking a crawler or server-rendering key pages.
• You want a number that competitors cannot move.
• You're reporting progress on one engine month over month.

Use citation share when

• You're mapping who dominates a topic before deciding where to get mentioned.
• You want to see which third-party sites engines cite instead of you.
• You're comparing yourself with peers of a similar size.

Use both when

• You're presenting to leadership. Rate shows your trajectory. Share shows the competitive context. Raw counts stop either number from being misread.

For a quick starting point, the free checker on this site shows whether engines name you for up to three buyer questions. It's a snapshot, so treat it as a baseline, not a trend. Nothing in either metric is a promise: you control access to your pages and the facts on them, not what the models choose to cite.