Experience designer

The experience designer decides what people see, read, hear and do. With AI (artificial intelligence) in a product, that includes a new job: designing calibrated trust. Users should trust the output as much as the evidence allows, no more and no less. The designer makes the AI's involvement, its sources, its uncertainty and the way out visible, and proves with real users that it works.

This seat also uses AI in its own work: synthesis, copy variants, prototypes. The same data rules apply. Research notes about real people are classified data. Counts and thresholds on this page are illustrative.

1. At a glance

Line: 1st line · Family: Shape · Variants: UX (user experience) design, service design, content design, conversational design, accessibility

2. Your journey: first steps to fully adopted

The designer moves from AI-drafted copy on internal screens to shared, tested trust patterns; each stage adds controls and evidence, never the right for AI to decide what a user reads.

Start here: your first steps

1. Confirm the research data rules with privacy before any tool sees participant notes.
2. Generate copy variants for one internal, non-confidential screen, then choose and edit one yourself.
3. Mask research notes, then cluster them with the approved assistant and check each theme against the raw notes.
4. Load the content style guide and glossary into a reviewed team prompt ("due date", never "payment date").

Fully adopted: how the work runs

What changes for the organization

In any organization

• Who covers it: in a small or unregulated organization, the product owner or a front-end developer does this design work; the answer card checklist still applies.
• What scales: research operations and a design system board can shrink to a shared document and a peer review. What never drops: no participant or customer data in unapproved tools, a human decides what customers read, and a visible fallback when the AI is off.

3. Your decisions

Decision relationships on this site: owner or approver, independent validator or challenger, consulted control partner, evidence recipient, notified.

4. Two kinds of adoption

5. Your ladder

Stays human at Stage 5: choosing the research sample, judging whether real users understood the uncertainty cue, and deciding what a customer reads. Synthetic users do not replace real ones.

6. Priority scenarios

7. The due-date ask from your seat

The designer has a material artifact: how the due date and cutoff read on four channels. No model is in the runtime path, so the design job is clarity, not AI trust.

Policy assistant · designing calibrated trust

The companion policy assistant is where the AI interaction design is material. Staff must check the cited section before they relay a consequential answer, so the design must make that check easy and expected.

8. Hand-offs and separation of duties

9. Evidence and records

10. Skills and learning path

11. Measures and incentives

12. Cost, capacity and workforce

13. Anti-patterns and stop conditions

14. Action kit and 30/60/90

## Experience designer · AI answer card checklist · v1
Disclosure: says it is AI and what the user must do next.
Source: document, version, section; one action to open the passage.
Uncertainty: clarifying question or abstention, never a confidence number.
Abstention: plain message plus the escalation route.
Correction: one-step "wrong or unclear" control that files a report.
Fallback: the non-AI path when the assistant is off; tested with it off.
Accessibility: works with a screen reader and keyboard only; criteria written.
Research: includes stressful context, edge cases and assistive technology.
Versioned: spec linked to the prompt and model version it was tested with.

Questions to ask your enablement lead:

• Which tools are approved for research synthesis, and what masking is required first?
• Where do the shared trust patterns live, and who approves a new one?
• Who tells me when the model or retrieval corpus behind my design changes?