How to Benchmark Your AI Citation Rate Against Competitors on the Same Questions
A citation rate only compares on the same questions, engines and runs. How to benchmark against your category leaders with Machine Relations Index segment data, and why one industry average misleads.
To benchmark your AI citation rate against competitors, measure it on the same buyer questions, the same engines and the same dates as theirs, and compare inside one question shape of your category. A single "industry average citation rate" fails that test. Every instrument counts on its own prompt set and its own denominator, so your number and its number measure different things. The Machine Relations Index publishes the like-for-like version: for each category and question shape, the share of observed answer runs that cited each domain, with the run count and the number of dates beside every rate.
A citation rate is cited runs over observed runs, and the scope decides it
A citation rate means nothing until you know which questions it was counted on. In the Index release of October 13, 2026 (methodology v2.1, window May 10 to October 13, 2026), the Enterprise Software segment for "how to choose" questions holds 147 observed answer runs across 9 dates. Erpresearch.com was cited in 41 of them, a 27.89% citation rate.
Change the question shape and the same domain in the same category reads differently. On "best tools" questions Erpresearch.com was cited in 60 of 149 runs, 40.27%. On "is it worth it" questions, 46 of 146, 31.51%. One domain, one category, three rates between 28% and 40%.
That is why a competitor's headline figure, or a vendor's industry median, cannot be your bar. Unless you know its questions, engines, dates and how a run is counted, you cannot measure yourself the same way, and a comparison across two methods compares the methods.
What the leaders look like on "how to choose" questions
Read straight off the October 13 release, each rate on its own segment's runs:
| Category / How buyers choose | Observed runs | Leading source | Leader's rate | 10th-ranked rate | Sources cited |
|---|---|---|---|---|---|
| Enterprise Software | 147 across 9 dates | erpresearch.com | 27.89% | 8.84% | 243 |
| Fintech | 149 across 9 dates | solidgate.com | 22.82% | 8.72% | 254 |
| Cybersecurity | 218 across 10 dates | microsoft.com | 19.27% | 6.88% | 406 |
| AI Visibility & GEO | 200 across 10 dates | youtube.com | 15.50% | 7.00% | 619 |
Three readings a CMO can act on:
- The top-ten line sits near 7% to 9% of runs in these four segments. A source cited in fewer than one "how to choose" answer in fifteen ranks below tenth in all four.
- The leader runs 2.2 to 3.2 times the tenth-ranked rate. The first place is a separate tier, not a slightly better score.
- Your AI competitors are the leaderboard, not your deal list. Enterprise Software's top ten mixes a market database, a trade publication (TechTarget), a consulting firm (Deloitte) and a vendor (NetSuite). Hundreds of sources share the rest of the citations. Benchmarking only against the companies you meet in deals misses most of who the engines actually cite.
For how far concentration swings between industries, Machine Relations Research's analysis of citation rates by vertical is the deeper read.
How to benchmark your citation rate on Monday
- Pick the segment. Your category crossed with the question shape your buyers ask: how to choose, best tools, comparisons, is it worth it, problem-first research, top lists or news. Benchmark each shape separately.
- Read its leaderboard. Note the leader's rate, the tenth-ranked rate and how many sources were cited. That band is your target, and the domains in it are your benchmark set.
- Measure yourself the same way. Run your own prompt set for that shape on the same engines, then divide the runs that cite your domain by the runs observed. Repeat it on several dates before you trust it. The Index publishes a rate only after at least 10 observed runs across 7 distinct dates; below that a segment shows as collecting, which is not a zero. Hold your own number to the same floor before you act on it.
- Read it per engine, not blended. Engines answer on different days. The October 13 release names Gemini as quiet, with 20 of 129 collected calls returning an answer, and states coverage for every engine. A blended rate can fall because one engine went quiet, not because a buyer saw less of you.
- Ask any benchmark for its denominator. Before you hold your number to someone else's: which prompts, which engines, which dates, how many runs, and what counts as cited. A cohort study such as Attrifast's brand-size benchmark, which Attrifast ran on about 200 of its own instrumented sites, shows a direction for that cohort. It is a bar for your number only if you measured it the same way.
Moving up a leaderboard means becoming a source the engines already cite for that question shape, or being cited by one. That is the work Machine Relations describes: earning trust with the machines your buyers now ask first.
FAQ
What is a good AI citation rate?
There is no universal good rate; there is your segment's. In the Machine Relations Index release of October 13, 2026, the tenth most-cited source on "how to choose" questions was cited in 6.88% to 8.84% of runs across Enterprise Software, Fintech, Cybersecurity and AI Visibility & GEO, and the leader in 15.50% to 27.89%.
How do I calculate my AI citation rate?
Count the answer runs that cite your domain and divide by the answer runs observed, inside one category and one question shape, per engine. Use several dates so a single day's answers do not set the number.
Why do two tools give my brand different citation rates?
They count on different denominators: different prompts, engines, dates and rules for what counts as a citation. Neither number is wrong, and neither can be compared with the other.
How often should I re-benchmark?
Re-read your segment after new run dates land. A segment built on nine dates moves more with each new date than one built on fifty, so treat a young segment's rank as provisional.