Academy
5. What to Do with an Orange or Red Badge
Apply practical strategies—asking the AI to cite sources, rephrasing the request, or grounding with a document—when a badge signals a warning or a serious issue.
An orange badge means at least one of TrustLayer's five controls raised a warning — for example, a claim in the answer wasn't found in the attached sources, or a constraint you set in your prompt wasn't fully respected. A red badge is more serious: it flags issues like the AI claiming it performed an action (sent an email, saved a file) that never actually happened, or generated code containing dangerous patterns. In both cases, the badge is an invitation to pause and verify before you use or share the answer. Never forward a red-badge answer without checking it first.
When you see orange or red, your first move should be to click the badge. It opens a breakdown showing each control's result. For source anchoring specifically, you'll see exactly which claims were matched to a source and which were not. This tells you where the weak spots are so you can focus your verification effort instead of re-reading the entire answer.
Ask the AI to cite its sources. A simple follow-up like "Which document or passage supports that claim?" forces the model to surface its references. If it can't point to one, that's a strong signal the claim needs independent verification.
Rephrase your request. An orange prompt-fidelity warning often means the AI drifted from your constraints or missed part of your question. Break a complex prompt into smaller, clearer steps and re-submit — the new answer will be checked again by TrustLayer.
The most powerful fix is grounding. Answers based on your own documents or knowledge bases can be source-anchored — TrustLayer can verify each claim against real content. Ungrounded answers (pure generation with no reference material) cannot benefit from this check, so the score stays lower. Uploading a relevant document or connecting a knowledge base before asking your question gives TrustLayer something solid to compare against, which typically raises the trust score.
Go to your Knowledge Bases, pick or create one with a document you trust. Then open a chat, attach that knowledge base, and ask a question the document can answer. Compare the badge color and score to the same question asked without the knowledge base.
Remember: even a green badge doesn't guarantee truth — it means the answer is consistent with visible sources, your request, and the tools used. A fact that is wrong in the real world but matches your document can still pass. Your judgment on important decisions remains essential.
When a badge turns orange or red: (1) click it to see which control flagged the issue, (2) ask the AI to cite its sources, (3) rephrase your prompt to be clearer or more specific, and (4) ground the conversation with a trusted document or knowledge base to enable source anchoring. These steps don't just help you verify — they actively improve the quality and trustworthiness of the AI's next answer. And always remember: the badge is a reading aid, not a certificate of truth.