Attribution & Observability
Advanced analytics for understanding AI accuracy patterns and detecting conflicts.
What Attribution & Observability is
Attribution & Observability contains advanced diagnostic tools for understanding AI accuracy patterns in depth. It includes analysis that goes beyond individual scan results to surface patterns, conflicts, and trends across all your scan history.
Hallucination Heatmap and Conflict Tracker are available on Starter and above. Evidence export is Growth and above.
What's in this section
Hallucination Heatmap
A visual grid showing which combinations of AI model and tracked query produce the most inaccuracies.
Each cell in the grid represents one model × query combination. Darker cells mean more inaccurate verdicts for that combination. Use this to identify your highest-risk query and model combinations at a glance.
How to use it: Focus on the darkest cells. These are the combinations where AI is most consistently getting your brand wrong. Run targeted scans on those specific queries and models to get the full evidence.
Conflict Tracker
Shows cases where AI makes contradictory statements about your brand — either within a single response or across different scans.
For example: AI states your product costs $49/month in one response and $99/month in another. Or AI says your product includes a feature in one paragraph and says it doesn't in another.
How to use it: Conflicts often reveal the most confusing or outdated information AI has about your brand. These are good candidates for Accuracy Fix actions and for updating your website with clearer, more consistent information.
Evidence Export
Export your scan evidence as a PDF document. The PDF includes scan results, verdicts, Integrity Scores, and the evidence behind each verdict.
Use this when you need to share documented proof of AI inaccuracies with a partner, customer, or platform.
Evidence export is available on Growth and above.
How to use Attribution & Observability
This section is most useful after you've accumulated at least 2–3 weeks of scan data. With limited scan history, patterns are harder to identify.
Check it:
- Monthly, to review patterns across your full scan history
- When you suspect a specific AI model is consistently problematic
- Before making major website or content changes, to establish a baseline