AI Answers
What exactly did the AI say?
Example finding
Open the answers behind a change
A moving metric links to the answers behind it, with the competitor named in each visible at a glance.
Quick answer: AI Answers
AI Answers stores every observed answer in full with its prompt, engine, model, market and time, plus the parsed result, so any metric can be traced to the exact text behind it.
What you see
Every observed answer, stored in full with its prompt, engine, model, market, time and parsed result, so any number can be checked against the source text.
The raw answer
The full text returned by the engine, unedited.
The parsed result
Mention, recommendation, competitors and citations extracted from that answer.
Run metadata
Prompt, engine, model, market, language and timestamp.
Filters
Slice by prompt, engine, market and date to find the answers behind a metric.
How it works
- 1
Store
The provider response is saved as received. Raw observations are never altered.
- 2
Parse
Structured fields are extracted, each with an extraction confidence.
- 3
Link
Every metric and finding links back to the answers it came from.
How it is measured
Definitions come first, so numbers stay comparable from one audit to the next.
- Observation
- One prompt run on one engine, with its raw and parsed forms.
- Extraction confidence
- How sure the parser is about each extracted field.
- Coverage
- The share of activated prompts with a successful observation per engine.
Good to know. An answer is one sample. The same prompt can return different wording on another run, which is why the platform reports rates over many runs.
Common questions
Are raw answers ever edited?
No. Raw observations are immutable. Metrics and explanations are calculated above them.
Can I see failed runs?
Yes. A provider failure is recorded explicitly rather than hidden.
Can I export answers?
Basic report and export features are coming in V2.
Who uses it
- Analyst
- The raw material behind every number.
- Brand manager
- The actual wording used about the brand.
- QA reviewer
- A way to spot check the parser.
Common mistakes
Treating one answer as the truth
It is a single sample.
Ignoring extraction confidence
Low confidence fields deserve a manual look.
Overlooking failed runs
They are recorded, not hidden.
A short checklist
- Filter to the slice you care about
- Read several answers, not one
- Compare parsed fields with the raw text
- Note recurring competitors
- Flag odd extractions
See AI Answers on your own brand.
Start with a first audit. We show where you appear, who is named instead, and what to fix first.