Trust and Transparency

Multi-Audience Split

Present the same AI output differently for each person who looks at it, without duplicating the underlying data.

Phase: Design · Build
Applies to: Designers, content designers
Oversight: Varies with role
Key factor: Oversight capability

Overview When to use Questions to ask Example Anti-patterns to pay attention to References Related

Overview

An AI system usually serves more than one audience in the same flow. A customer sees the outcome. A reviewer sees the evidence and controls. An operator sees the pipeline state and configuration. Showing the same view to all three buries what each needs and exposes what they don't need to see. Multi-Audience Split designs the surface differently for each role, drawing on the same data but presenting it at the right level of detail, with the right actions available.

When to use

Use when

The same agent output is used by people with different roles and different decisions to make.
Showing detailed evidence or controls to an end user would be confusing or overwhelming.
Regulation requires that different people have access to different levels of information.

Not needed when

A single role interacts with the output, and there's no meaningful difference in what people need.
The output is a simple status message with no review requirement.

Questions to ask

Ask these at kickoff, when design work starts. Tick them off as your team answers.

Who are the distinct audiences for this output, and what decision does each one need to make?
What does each audience need to see to make that decision, and what would confuse or overwhelm them?
Is there anything one audience shouldn't see, for confidentiality, role separation or simplicity?
Does the same surface adapt by role, or are these genuinely separate views?
When someone has more than one role, which view takes precedence?

Example

An order approval flow has three views of the same order.

One order · Three viewsDo

Customer

Your order is being reviewed. You'll get a confirmation once it's approved.

No scores, no field-level breakdown.

Reviewer

Unit price £38.00 Check: header vs line item

Evidence: invoice p. 2, line 3 · Prices updated today

ApproveReturn for correction

Operator

Queue: 42 orders · 6 below threshold

Pipeline healthy · Config v12 · Confidence trend steady

Same order, three surfaces, none of them cluttered with what the other two audiences need. The reviewer's view uses Confidence Translation, Inline Evidence and Data Freshness Signal.

Anti-patterns to pay attention to

The reviewer and the people affected see the same component or copy.
Affected people see scores, codes or model terms.
Affected people get no explanation and no route to a person.

References

Current guidelines · EU AI Act2 references

For orientation only, not legal advice. Applies to high-risk systems. EU AI Act links open a readable version of the official text published on EUR-Lex; ISO/IEC 42001 is a paid standard.

EU AI Act Art. 13(2) Instructions for use must be concise, complete, correct and clear, and relevant, accessible and comprehensible to deployers. Designing every audience's view to that standard extends the same principle.
EU AI Act Art. 14(4)(c) People overseeing the system must be able to interpret output correctly. A surface cluttered with customer-facing content doesn't support correct interpretation.