Skip to main content

Brand Safety

The Brand Safety module detects AI hallucinations about your brand and products — cases where AI states untrue facts (wrong price, non-existent feature, invented history, false availability).

What it's for​

  • Compares AI answers against your ground truth facts.
  • Classifies detected mismatches by severity: critical / high / medium / low.
  • Points to the recommended action and remediation status (e.g. takedown / content fix).

Availability​

Brand Safety is available in the Commerce, Commerce Pro, Enterprise and Agency packages (not in Brand). See Limits.

Ground truth​

Ground truth is a set of true statements about your brand and products against which we evaluate AI answers. Categories include pricing, features, company history and certifications. See the glossary.

How to start​

You need defined ground-truth facts — the more complete the set, the more effective the detection.

Step by step​

  1. Open Brand Safety → Ground truth and add facts (statement + category + source URL). Facts can be activated / deactivated.
  2. Run checks (the module also runs in the nightly monitoring cycle).
  3. Review detected hallucinations: the AI answer excerpt, the linked fact, type and severity.
  4. Filter by severity (start with critical) and act — fix content, update the fact, or set a takedown status.

FAQ​

Where does a price hallucination come from? AI relies on web content, which may be stale. If AI states an old price, it's a signal that content across channels needs refreshing (see Content Refresh).

How is this different from trust in the product module? Product trust checks the price and availability of specific SKUs in shopping queries. Brand Safety covers a broader set of facts about the brand and products (features, history, certifications), not just price.

What do I do with a critical hallucination? Prioritize fixing content in the channels AI relies on (see the AI Sources module), and where justified start a takedown process. You track the status in the module.

Do detections count toward the AI budget? Yes — detection uses an AI model, so usage counts toward your budget, protected by the daily safety valve.