The Citation Ledger: How We're Building an AEO Benchmark
SourceRank is building the first AEO benchmark sourced entirely from live, tracked measurement runs across real client accounts. This page publishes the methodology before the numbers, so anyone can check the method instead of taking our citation rates on faith.
Most "AI visibility" statistics floating around right now trace back to a single vendor's small sample, a one-time crawl, or a press release with no method attached. We think a benchmark is only worth citing if the method that produced it is public and repeatable. So we are publishing the method first, as a standalone page, and will publish the report itself once the underlying data clears an honest sample-size bar. This page is that method, and it is also the page we will update in place the day the first quarterly report ships.
What is the Citation Ledger?
The Citation Ledger is the name for SourceRank's ongoing measurement program: a fixed, per-account set of real buyer questions, run against every major AI answer engine on a recurring cadence, with every citation and mention logged as a timestamped snapshot. It is not a one-off audit. It is the same prompt set, run again and again, so a change in citation rate can be attributed to something that actually happened, a page shipped, a schema fix, a backlink landed, rather than noise.
The quarterly AEO benchmark report is what happens when we aggregate the Ledger across every client account, strip anything identifying, and publish the pattern.
How does the measurement actually work?
Each client engagement runs a tracked prompt set: 15 or more real buyer questions, phrased the way an actual buyer phrases them, not the way we would phrase them internally. The set is fixed per account so results are comparable across time.
For each prompt, the SourceRank AEO measurement engine queries every tracked answer engine and records a structured snapshot. Right now the engine tracks five platforms: ChatGPT, Gemini, Microsoft Copilot, Google AI Mode, and Google AI Overviews. That list will grow as new answer engines earn enough query share to matter.
Mention rate
Share of tracked prompts where the brand is named at all, per engine and overall.
Citation share
Share of answers where the brand is the cited source, not just mentioned.
Total responses
How many prompt-by-engine combinations were actually queried in that run.
Sentiment
Whether the mention is neutral, favorable, or unfavorable.
Competitor co-mentions
Which named competitors showed up in the same set of answers.
Recommendation flag
Whether the engine's answer actively recommended the brand, not just named it.
That snapshot is what "citation rate by engine" and "citation rate by industry" will be computed from once there are enough of them.
What counts as a citation, specifically?
We separate three things that get conflated in most AI-visibility claims. A mention is the brand's name appearing anywhere in the answer text. A citation is the engine naming the brand as a source for the specific claim it is making, the AI-answer equivalent of a backlink. A recommendation is the engine actively suggesting the brand as an option, not just referencing it in passing.
The benchmark report will state citation rate, not mention rate, as the headline number, because citation is the metric that tracks with the thing clients are actually paying to move: whether the engine treats the brand's own pages as the source of truth.
How will anonymization work?
No client name, domain, or identifying detail will appear in the aggregated benchmark. The report publishes patterns across buckets, never a single account's numbers.
Industry buckets
Every data point is rolled up to an industry bucket, for example "B2B SaaS" or "professional services", before publication.
Minimum bucket size
A bucket only ships in the report once it holds a minimum number of distinct accounts, so no bucket can be reverse-engineered to a single client.
No absolute figures
Traffic, revenue, and exact rankings never appear. Only rates, deltas, and distributions.
Opt-out available
Any client who does not want their measurement data included in aggregate research can opt out, disclosed in the standard engagement terms.
What will the quarterly report actually contain?
Citation rate by engine
How often brands get cited by ChatGPT versus Gemini versus Copilot versus Google's AI surfaces, aggregated across every tracked account.
Citation rate by industry
Whether some verticals are structurally easier or harder to get cited in right now.
Typical gap size
The distribution of the gap between a brand's citation rate and its top competitor's, so a client can see whether their own gap is normal for their industry.
What changed after intervention
For accounts with before-and-after snapshots bracketing a specific change, a schema fix, a comparison page, a backlink campaign, what the citation-rate delta looked like. This needs the most runway, since it requires paired snapshots on either side of a known change.
Why isn't the report live yet?
Because publishing a "benchmark" from a thin sample would make it exactly the kind of unaccountable statistic this page is trying to be an alternative to. This methodology page will be updated in place, not replaced, the day the first quarterly release ships, so the URL and the method stay stable even after the numbers land.
- +The Citation Ledger is SourceRank's name for its recurring, per-account AI-citation measurement program, the source data behind the quarterly AEO benchmark.
- +A citation is not the same as a mention. The benchmark reports citation rate because that is the metric tied to being treated as a source, not just being name-dropped.
- +Five engines are tracked today: ChatGPT, Gemini, Copilot, Google AI Mode, and Google AI Overviews.
- +Anonymization happens by industry bucket with a minimum account count per bucket, never by publishing a single client's numbers.
- +The report ships once the data clears an honest sample-size bar, not on a fixed calendar date regardless of what the data supports.
Mention rate
The share of tracked prompts where a brand is named anywhere in the AI answer, regardless of whether it is cited as a source.
Citation rate
The share of tracked prompts where a brand is named specifically as the source for a claim the engine is making, the AI-answer equivalent of a backlink.
Recommendation rate
The share of tracked prompts where the engine actively suggests the brand as an option, not just references it.
Tracked prompt set
The fixed list of 15 or more real buyer questions run against every engine on a recurring cadence for a given account, kept stable over time so results are comparable.
Snapshot
One full run of the tracked prompt set against every tracked engine for one account on one date, logged with mention, citation, sentiment, and competitor data.
Answer engine
Any AI system that synthesizes a direct answer from web content rather than returning a list of links: ChatGPT, Gemini, Copilot, AI Mode, AI Overviews, and others as they gain query share.
Common questions
Is this the same as a Google ranking?
No. A ranking is a position in a list of links. A citation is whether an AI answer engine treats a brand's content as the source for a synthesized answer. A brand can rank well and still never get cited, or vice versa.
Why publish the methodology before the report?
Because a statistic without a public, checkable method behind it is not a statistic, it is a marketing claim wearing a number. Publishing the method first lets anyone judge whether the eventual numbers are trustworthy before they exist to be argued about.
Will individual client results ever be identifiable?
No. Reporting happens at the industry-bucket level with a minimum account count per bucket, and any client can opt their data out of aggregate research entirely.
How often will the benchmark update after it launches?
Quarterly, matching the cadence of the underlying tracked-prompt-set runs, so each release reflects a full cycle of new snapshots rather than a partial one.
What engines are excluded right now, and why?
Any answer engine that has not yet earned meaningful query share is not worth the measurement overhead. The tracked list is reviewed as engines gain or lose relevance, not fixed permanently.
Can I see my own account's numbers before the industry benchmark ships?
Yes. Every SourceRank client already sees their own tracked-prompt-set results in their account snapshots. The benchmark is the anonymized, aggregated version across all accounts, a separate publication from a client's own dashboard.