The number is a count, not a guess: for every buyer question your brand is eligible to answer, across ChatGPT, Claude and Perplexity, how many of those answers named a competitor instead of you. That's the exact number Reflexa's free check surfaces first — not a sentiment score, not a single scary screenshot, but a tally of specific answers where the recommendation went elsewhere. Here's how the count is built, a worked example from public data, and what it honestly can't tell you.
"Buyers AI sent elsewhere" is the number of eligible buyer-question answers in which a named competitor got the recommendation and your brand didn't — nothing more, nothing less. It isn't a visit, a click or a lost deal; it's an answer. Take a fixed set of real buying questions in your category ("what's the best [category] for [buyer]?", "is [competitor] better than [you]?"), ask each one across the engines people actually use, and count how many of the answers where you had a real shot at being named instead named someone else. That's the same eligible-answer logic behind AI brand intelligence generally — applied to one specific, countable outcome.
Bottom line: the count answers one question precisely — "in how many chances did AI pick a rival over us" — and nothing beyond it.
Where AI share of voice counts the answers where you got named, "sent elsewhere" counts the ones where a rival did instead — the same eligible-answer set, read from the other side. The two numbers should roughly complement each other for a two-horse race, but in a crowded category they don't have to: an answer can name three competitors and still leave you out, so "sent elsewhere" can move independently of your own share of voice if the field of named rivals widens or narrows. Reading both together is more useful than either alone — one says how often you win the mention, the other says how often, and to whom, you lose it.
Why it matters: a flat share of voice can still hide a shrinking or growing "sent elsewhere" count — check both, not just the one that looks better.
A trustworthy "sent elsewhere" number needs the same three fixed inputs as any other AI-answer metric: a defined question set, every engine you care about, and more than one run per question. Skip any of the three and the count stops meaning anything — a single run on a single engine just tells you what that one answer happened to say today, the same caveat that applies to tracking named competitors more generally. Log, for each answer, whether a competitor was named and you weren't, then total it across engines and runs.
Do this: fix the question set and the engine list before you start counting — changing either mid-way makes the trend meaningless.
Reflexa's own AI Visibility Index publishes this exact kind of count in the open, and it shows how large "sent elsewhere" can get even for a known brand. In the September 2026 project-management edition — 100 fixed buyer questions, asked three times each to ChatGPT, Claude and Perplexity — Jira appeared in the top three recommendations in only 59 of its 882 eligible answers. Read the other way: in 823 of those eligible answers, the top-three recommendation went to a different tool entirely — Asana, monday, ClickUp or another named rival, not Jira.
Proof: Reflexa AI Visibility Index — September 2026, project management edition.
Rule of thumb: a well-known brand can still have a large "sent elsewhere" count — recognition and top-3 recommendation rate are two different numbers.
"Sent elsewhere" counts answers, not lost customers — it can't tell you how many of those buyers actually acted on the recommendation, or would have bought from you anyway. An AI answer is one input into a purchase decision, not the whole funnel: some buyers cross-check the recommendation, some already had a shortlist, some never intended to act on that question at all. The honest way to read the number is as a directional signal — "AI is pointing buyers at a rival more often than it's pointing them at us, for these questions" — not as a revenue figure with a dollar sign in front of it.
Good news: you don't need a revenue model to act on it — a rising count is reason enough to look at what the winning answers cite that yours don't.
You can start the same way as any other AI-answer count: write down 15–20 real buyer questions in your category, ask them across ChatGPT, Claude and Perplexity, and mark each answer where a competitor is named and you aren't. Repeat monthly — see how often to re-check — and watch whether the count is growing, shrinking, or concentrated around specific competitors and specific questions, which is usually the more actionable read than the total on its own.
Shortcut: that's exactly what Reflexa's free check automates — it runs the fixed question set for you and shows the count, with the answer text and the competitor named in each one attached.
No single count turns a probabilistic system into a fixed scoreboard. The same answer, asked again, can name a different competitor or none at all — which is exactly why the number needs repeated runs across a fixed question set to mean anything, and why one snapshot should never be read as a verdict on where every buyer went. Treat it as a tracked trend, not a one-time score.
Sources: Reflexa AI Visibility Index — methodology and worked example · Share of voice in AI answers · What is Brand Intelligence?
The free check reads it across ChatGPT, Claude and Perplexity — 3 minutes, evidence included.