Skip to content
Why GEO Why us Pricing
Get started free
HomeProductAnswers & competitors

See what the engine actually said.

The answer explorer is the record of what each engine said — see every verbatim reply, the rivals it named instead of you, and the sources it pulled from.

The answer explorer

Answers and competitors is the evidence layer under every MentionBeat number: the verbatim reply each engine gave to each prompt, kept and searchable, alongside the rivals it named instead of you and the sources it pulled from while answering. A rate tells you that you lost a prompt. This tells you what was said, who won it, and why — which is the difference between a metric and a decision. Every sampled response is stored, not summarised. Filter down to the exact slice you are arguing about and read what the model said.

Filter to the loss

By engine, prompt, ICP, journey stage, market, or outcome — named, not named, recommended, ranked. "Show me every prompt where Gemini recommends someone else" is two clicks, not an export.

Read it verbatim

The full text as returned, with your brand and every rival highlighted, the position each appeared in, and the citations the answer carried where the engine supplied them.

Compare across repeats

The same prompt sampled five times, side by side. Consistent absence is a content problem; inconsistent presence is a retrieval problem. They need different fixes.

The field around you

Competitive position in AI answers is not a ranking table you can look up — it only exists relative to the questions being asked. MentionBeat scores every brand that appears in your answers the same way it scores you.

  • Rival leaderboard — share of voice per competitor, per engine, over time
  • Head-to-head — the prompts a specific rival wins that you do not
  • Uninvited entrants — brands appearing in your category that you never listed
  • Position, not just presence — being named fourth is not being named first
  • The framing — how the answer characterises each option, in its own words

Head-to-head is usually where a quarter's content plan comes from: a short list of specific questions where a specific rival is being preferred, each with the answer text that explains the preference.

Where the engines get their facts

For web-grounded engines, MentionBeat records the domains retrieved while answering your prompts and ranks them by frequency. The list is frequently uncomfortable: a review roundup you have never heard of, a four-year-old forum thread, a competitor's comparison page.

Those pages are effectively writing your category's answer. Knowing which ones they are turns "improve our AI visibility" into a short list of specific, addressable sources.

Correcting a wrong source →

Not named is four different problems

Every other tool reports one number: you were not mentioned. That number is useless, because the fix for each cause is completely different — and one of them has no fix at the page level at all. For every answer, MentionBeat classifies which of these actually happened.

What happenedWhat it meansWhat to do about it
NamedThe engine retrieved you and put you in the answer.Hold the position and watch it.
Retrieved, not namedYou were in the candidate set the engine pulled, and it chose someone else.A selection problem. The page was found and rejected — change what it says, not whether it exists.
Not retrievedRetrieval happened for this question and you were not among the candidates.A retrieval problem. Classic SEO and source work applies.
Cited, not namedYour page was consulted, but your brand was not credited in the answer.An attribution problem — usually a page that answers the question without naming who is answering.
Answered from memoryNo retrieval happened at all. The model answered from training.No amount of page work changes this one. Entity and third-party corroboration is the only lever.
UnknownNo ranked results were captured for that question.Nothing — and we say so rather than reporting it as an absence.

That last row is the one that matters most for trusting the rest. “We could not tell” is never rounded down to “you were not there”, and where an engine does not publish what it retrieved — Perplexity, for instance — the classification is labelled as measured by proxy rather than presented as direct evidence.

From an answer to an action

Each pattern in the explorer maps to a different remedy. Guessing between them is how GEO work gets wasted.

What you seeWhat it usually meansWhere it goes next
Never named, on any engine, for a whole intentNo page of yours answers that question in a liftable formContent Studio — a page family for the intent
Named by parametric engines, missing from grounded onesThe model knows you; the live crawl cannot reach or parse youFix queue & site health
Named, but never as the recommendationComparison and proof material is missing or unciteableContent Studio — comparison pages
Named with a wrong fact attachedA retrievable source states something false about youAccuracy & corrections
One rival appearing everywhere you are notThey own the sources the category is answered fromHead-to-head view, then a targeted plan

Questions about the evidence

Do you store the full answer text?
Yes. Every sampled response is kept with the prompt, engine, timestamp and any citations it carried. Rates are derived from those stored answers, which is what makes a disputed number checkable rather than a matter of trust.
How are competitors identified?
You name the rivals you care about, and MentionBeat also surfaces brands that keep appearing in your answers whether or not you listed them. The second group is often the useful one — categories in AI answers rarely match the competitive set a company believes it has.
What are most-cited domains?
For web-grounded engines, the sources retrieved while answering your prompts, ranked by how often they appear. It tells you which third-party pages are effectively writing your category's answer, and therefore where a correction or a placement is worth pursuing.

Keep going

See what the engines say about you.

Run a suite and read the answers — including the ones that name someone else.