AI NativeNumbers summarize a scan. The answers explorer lets you read what the AI actually said. Every answer a scan collected is stored here with the prompt that produced it, the engine that answered, and the outcome state assigned to it. Reading the answers turns a rate into a reason.
What the explorer shows
The explorer opens on a product and shows all answers across every run, with a total count at the top. Brand names are highlighted green and competitor names amber so you can read an answer and immediately see who is named and in what order. Each row shows the prompt, the engine, the outcome badge, and sentiment. Clicking a row expands the full answer text.
Outcome states
The outcome ladder runs from absent through to recommended first:
- Absent: the answer did not name your brand.
- Mentioned: the answer named you but did not put you forward as a choice.
- Listed option: you appeared in a list without being recommended.
- Recommended: the answer put you forward as a choice.
- Recommended first: you were the first or primary recommendation.
A "unknown" state appears when the scoring algorithm could not assign a clear outcome.
Filters and sorting
The filter bar lets you narrow by prompt class (unbranded or branded), outcome, engine, persona, sentiment, and accuracy. Filters carry through pagination: page two of a filtered set applies the same filter, and the sort order is preserved the same way.
Filtering by outcome "absent" and class "unbranded" isolates answers where you did not appear on discovery questions - the set the attack queue is built from. Filtering by accuracy "inaccurate" shows every branded answer where a verified fact was contradicted.
Accuracy detail
When a branded scan checks answers against your verified facts, each contradiction surfaces in the expanded view as a diff: the fact label, what AI stated, and what your verified value actually is. This is the fastest path from a low accuracy rate to the specific claim that needs fixing. You manage verified facts in the Brand Truth Studio.
Citations
Where an engine cites sources, those domains appear as chips in the expanded view. These are the same citations that feed the WHY ledger and the sources analysis, so the answers explorer and the source view are two windows on the same evidence.
Branded vs unbranded answers read differently
On unbranded prompts, the signal is mention rate, share of voice, and recommendation. On branded prompts, mention is very high and share of voice is not meaningful; what matters is sentiment and accuracy. The explorer keeps these classes filterable so you read each for what it actually measures.
Questions
How many answers are shown at once?
Fifty per page. Page number and total appear at the bottom. All active filters carry through every page.
Does the explorer show only the most recent scan?
No. It shows answers across all runs for the product. To read only one scan's answers, use the run-level answers view from the run page.
What does "unknown" outcome mean?
The scoring algorithm could not determine a clear outcome, typically because the answer did not follow a standard recommendation format. Unknown is distinct from absent: absent means the brand was not named, unknown means the outcome was indeterminate.
How do I find all inaccurate branded answers?
Set accuracy to "inaccurate" and class to "branded." Each expanded row shows the specific fact contradiction with the label, AI's claim, and the truth.
Can I export the answers?
Bulk export is available from the run-level reports view. The answers explorer is a reading and filtering interface.
Are answers updated in real time during a running scan?
No. Answers are written when the scan finishes. The run page shows in-progress status, but the explorer does not update incrementally.