Retrieval & ranking
Selection Rate
How often you're actually cited once you've made it into the retrieved candidate set.
Selection rate is how often an engine actually cites your page given that it was retrieved in the first place. It isolates the second gate of AI visibility: being fetched into the candidate set is one thing, being chosen from that set and attributed in the answer is another. A page can be retrieved constantly and selected rarely if its passages are weak.
Separating retrieval from selection is useful diagnostically. If you are rarely retrieved, the problem is corpus, crawlability or relevance. If you are retrieved but rarely selected, the problem is the passage itself — it is not quotable, specific or trustworthy enough to beat the other candidates.
Raising selection rate is the AEO craft: clearer answers, denser facts, cleaner extractable units, stronger entity signals on the page.
Key takeaways
- Selection rate is the share of eligible or retrieved sources that an engine actually cites in its final answer.
- You can be retrieved and still not cited, and selection rate is the metric that exposes that gap.
- Improving selection rate is about being the clearest, most quotable passage among the sources an engine already found.
Retrieved is not the same as cited
Selection rate measures how often an engine, having retrieved a set of candidate sources, chooses to actually cite them in the answer it writes. An engine might pull a dozen passages during retrieval and quote only three. The ratio of cited to retrieved is the selection rate, and it is an AEO metric worth tracking because it isolates a stage most people ignore.
This matters because the retrieval battle and the citation battle are different. Getting into the top retrieved set is about semantic similarity and clean chunking. Getting selected from that set is about being the passage the model finds most useful to quote: specific, self-contained, unambiguous, and easy to attribute.
If you are getting retrieved but rarely cited, your problem is not visibility in the corpus. It is that your passage, once read, is not the one the model wants to lean on. That is a fixable content problem.
Raising your selection rate
Make each passage stand on its own so it can be lifted without surrounding context. State the answer directly, attach a specific number or named source, and keep the claim unambiguous. Vague or hedged passages get retrieved and then passed over for a cleaner competitor.
This is the Refine stage of the ANSWER Framework in action, and it is why we run a recommendation check to see not just whether you appear, but whether engines choose you when you do. Pair strong citation readiness with extractability and your selection rate climbs.
Retrieval vs. selection
| Stage | What it measures | How you improve it |
|---|---|---|
| Retrieval | Did your passage make the candidate set | Clean chunks, strong semantic match |
| Selection | Did the engine cite it in the answer | Specific, self-contained, quotable passages |
Example
A client was getting retrieved for "average implementation time for HR software" but almost never cited. Their passage said implementation "varies depending on your needs," which is true and useless to quote.
We rewrote it to "Most mid-market HR implementations take 6 to 10 weeks," with the basis stated. Same retrieval, but now the model had something concrete to cite, and the selection rate for that query jumped.
Frequently asked questions
How is selection rate different from share of voice?
Share of voice is about how often you appear across all answers. Selection rate is narrower: among the sources an engine actually retrieved, how often it chose to cite you.
Why would an engine retrieve but not cite a source?
Because the retrieved passage, once read, was redundant, vague, or harder to quote than another candidate. Retrieval judges relevance, selection judges usefulness for the answer.
How do I improve a low selection rate?
Tighten your passages: direct answers, specific numbers, named sources, and no dependence on surrounding text. Give the model the cleanest thing to quote and it will choose you more often.
Sources
Related terms
Chunk Retrieval
Pulling the few passages most relevant to a query out of a vector index.
Citation (AI)
The engine pointing to one of your pages as the evidence behind part of its answer.
Citation Readiness
A page's overall fitness to be quoted and confidently attributed by an AI answer.
Extractability
Whether a passage survives being pulled out of context and dropped into an answer.
Retrieval-Augmented Generation
Retrieve the relevant documents first, then let a model write an answer grounded in them.
Want to know where you stand in AI answers?
We’ll show you which engines cite you, and which don’t.
Get a free AI citation audit