Citations

For claims that matter — safety, health, dosage — the assistant shows the source behind them, and flags anything it can't back up.

When the assistant makes a claim that matters, it can show the shopper where that claim came from — whether the source is a document you added under Knowledge or live data the assistant looked up while answering (a product's details, a policy, an order). Citations are reserved for the claims that matter — so the marks mean something instead of decorating every sentence.

What the shopper sees

A cited claim appears inline in the reply with a dotted underline. Hovering (or focusing) it opens a small tooltip showing the exact sentence from the source it drew on, with the cited part in bold, under a green Verified claim header.

There's nothing to click through to and no page to leave — the proof is right there in the tooltip.

Verified vs. unverified

• Verified — the claim matched a real passage in the source the assistant drew on. Dotted underline, green tooltip with the source sentence.

• Unverified — the assistant stated a safety-critical claim that didn't match any source. It's highlighted in red and labelled Unverified claim, telling the shopper the claim couldn't be backed up. This is a guardrail: rather than let an unbacked critical claim pass as fact, the assistant flags it in plain sight.

Which claims get cited

Citations are deliberately limited to claims where being wrong has a cost — dosages, health, allergens, safety, and authority claims like "certified" or "clinically proven." Everyday details — including prices — stay as plain text even when they come from a source, so the marks don't lose their meaning.

What you can improve

There's no citation setting to turn on. Citation quality follows the sources the assistant has to work with: when a source states the answer plainly, the assistant can verify against it cleanly. Your Knowledge sources are the part of that you control directly — a well-written document is the difference between a Verified and an Unverified mark on a critical claim. See Adding sources and Testing answers.

An Unverified flag on something a shopper should be able to trust usually means the answer lives in a source that's missing, vague, or buried — a signal to tighten that document rather than a bug in the assistant.