Mark what the product vouches for
Tags: trust, citations, uncertainty
Every AI interface makes a claim about how reliable it is, and most make it by accident. The model produces one stream of text; the interface renders all of it in the same font, in the same box, in the same confident voice — so the user is invited to believe the quoted fact and the invented one to exactly the same degree.
Deciding which parts of an answer the interface will vouch for is a design decision. If nobody makes it deliberately, the default styles make it.
The output is not one kind of thing
A single paragraph of model output routinely contains several different epistemic objects:
- Text copied from a document the user supplied.
- A summary of that document, which may be a fair reading or may not.
- A general claim from training.
- A plausible sentence produced because the shape of the answer needed one there.
Those have genuinely different reliability, and the user has no way to tell them apart, because nothing in the presentation distinguishes them. The interface has flattened four kinds of claim into one voice — and that flattening, not the model being wrong, is the problem.
Three things an interface can do with a claim
| What it means | Right when | |
|---|---|---|
| Assert | Present it plainly, as the product's own statement | You'd be willing to be held to it — it's deterministic, or checked, or the cost of being wrong is trivial |
| Attribute | Present it with its source attached and reachable | The claim came from somewhere specific and the user can check in one click |
| Withhold | Don't present it as an answer at all | You have no basis for it, and a plausible sentence would be worse than a gap |
Most AI features use the first for everything. The work is deciding, per surface, which claims move to the second row — and being willing to use the third at all, since an empty state saying the product doesn't know is harder to ship than a paragraph that sounds fine.
Confidence scores are the wrong instrument
The usual answer is a number: 87% confident. Two problems.
Calibration. A model's reported confidence is frequently not a probability of being right in any usable sense, so the number is often decoration with a decimal point.
Actionability. Told a claim is 73% likely, a user has no procedure. Nothing in that number says what to do next, so they either ignore it or treat anything above a vague threshold as true.
What people can act on is where it came from and what happens if it's wrong. A citation that opens the source passage gives someone a concrete next move. Knowing an action is reversible tells them how much to care. A percentage is not something a person can do anything with.
The ladder from sensitive data
There is an existing discipline worth stealing. For sensitive values — medical records, account numbers — the rule is well established: show the minimum necessary by default, make the full value a deliberate act to reveal, and for the most consequential cases record that the reveal happened. The same three moves apply to model claims.
- Minimum necessary. Show the least the user needs to act. Not the whole retrieved chunk, and not the model's reasoning by default. Excess detail reads as thoroughness and functions as noise.
- Deliberate reveal. The evidence is one intentional click away, and it is the real thing — the actual source passage, not a summary of it.
- Record what was claimed. For anything consequential, keep what the system asserted, what it was based on, and when. Partly so a wrong answer can be traced, and partly because a team that can't reconstruct what the product said last Tuesday cannot improve it.
Grounded in
Four distinct epistemic objects identified inside a single stream of output, and the three things an interface can actually do with each one.
Trust also runs in the other direction — what the user is asked to trust the product with. On DocRobot the blocker was never quality; legal would not permit confidential documents near a hosted model at all, so redaction became the first step in the flow rather than a setting. The trust model was the product rather than a feature of it.
Anti-patterns
- One voice for everything. Quoted fact and invented sentence rendered identically, so the interface vouches for both by default.
- A citation that expands into more generated text. Worse than no citation: it looks like verification while adding another layer of the same uncertainty.
- Percentages as a trust surface. Uncalibrated, and unactionable even when honest.
- Reasoning dumped by default. Excess detail reads as thoroughness and functions as noise.
Related patterns
- Design the failure state first — "confidently wrong" is the failure this pattern is the interface for.
- Redaction moves to the export — the other direction of trust, and where the boundary actually sits.
The full argument: What the user is allowed to trust.