Synthetic users vs. research; assumption audits; unhappy paths; ugly wireframes.
UX · graphic design · motion — 4 hours, drop-in friendly
UX and Product Design: Research, Flows, Wireframes, and Prototypes
A five-prompt UX planning kit — assumptions, flows, wireframe text, edge cases, accessibility — and the one rule that keeps AI-assisted UX honest: personas are hypotheses, not people.
01This Week at a Glance
You will make: one task flow + one low-fi wireframe + a critique checklist for an app idea.
Edge Case Hunt logic; accessibility specificity; the prototype chain.
"An app for plant people" — the five prompts, live.
Edge Case Hunt on your idea; trust-damage rankings.
The week's stakes
UX is where AI's biggest lie lives: confident personas, quotes from users who don't exist, statistics from studies that never happened. Fake research is worse than no research — it carries authority without evidence.
The professional split
AI for structure and pessimism (assumptions, flows, edge cases, microcopy). Research for evidence and empathy (what's actually true of real people). AI can imagine your users; only research can know them — label which one you're holding.
02Key Concepts
1 · Synthetic users vs. real research
AI-invented users are hypotheses wearing name tags — useful exactly as hypotheses (they force assumptions into the open), poison the moment anyone treats them as findings.
- Label every persona:
HYPOTHESIS — needs real research. - For each persona claim, write the interview question that would test it.
- Never present synthetic users as research. That's fabrication, and it gets discovered.
2 · The assumption audit: the cheapest UX win
Before any design: "list every assumption hiding in this idea," then sort — safe to assume / must research / dangerous to guess. Ten minutes that reorder a project. This is AI at its best: pessimistic structure, no facts required.
3 · Flows: the happy path is the easy 20%
AI drafts a clean flow in seconds. The value move is the next prompt: "now give me the unhappy paths." Bad photo, no signal, ambiguous result, abandoned mid-task. Models are surprisingly good at enumerating failure — they've read every support forum on earth. Most of real UX is designing the recovery.
4 · Wireframes: ugly on purpose
Grey boxes, real structure, real microcopy. Polish hides thinking — yours and the model's. AI writes the wireframe text (each screen's purpose, elements, exact button labels, empty states); you draw the boxes. Microcopy is a real deliverable: error messages are where products feel human or hostile.
Watch out
AI-generated visual UI looks finished and thinks nothing. Wireframes stay grey this week. App-builders return in Week 11 — with inspection habits.
5 · The Edge Case Hunt: pessimism as a service
After any flow: "8 edge cases ranked by how badly each damages trust." Then accessibility made specific — not "be accessible" but this flow's risks (photo capture with low vision, colour-only status, touch targets in gloves) — verified against WCAG, because models paraphrase guidelines loosely.
6 · From plan to prototype
flow → wireframe text → boxes → clickable path. AI accelerates every document before the canvas; the canvas is still judgment.
03This Week's Words
Persona (hypothesis)An invented user sketch. Question generator, never "findings."
Synthetic usersAI-simulated research participants. The industry fight of the moment.
Assumption auditListing what you're assuming before designing. Sort three ways.
Task flowStep-by-step path to a user goal — including unhappy paths.
Unhappy pathWhat happens when things fail. Where the real design work lives.
WireframeDeliberately ugly structural layout. Grey boxes, real microcopy.
MicrocopyButtons, errors, empty states — where products feel human or hostile.
Edge caseThe unusual-but-real situation that breaks naive designs.
WCAGThe accessibility standard. Verify against the source, not paraphrase.
Information architectureHow content is organized and named so people can find things.
04Demo: "An App for Plant People"
I made this, so can you
A vague idea becomes an honest UX plan in five prompts — with the model's most dangerous output (plausible personas) caught and labelled in front of the room.
Prepared demo artifact
Instructor backup / finished example: ux-plan/flow.md ux-plan/wireframe.html ux-plan/personas.md. Use it if the live tool run fails, or to reveal the finished structure after critique.
- Assumption audit — every assumption in the sentence, sorted; the dangerous-to-guess items starred.
- Personas as hypotheses — two contrasting hypotheses, labelled, each with 5 test questions. "These people do not exist. They're hypotheses wearing name tags."
- Flow + unhappy paths — diagnose-my-plant with three failure states designed in; the Mermaid code pasted into mermaid.live becomes an instant diagram.
- Wireframe text — five screens with exact microcopy; one element overruled live and marked ⚠ ("on a diagnosis app the camera IS the app — moved, mine").
- Edge Case Hunt + accessibility — eight cases ranked by trust damage; one accessibility claim verified against actual WCAG numbers.
05Step-by-Step Workflows
The five-prompt UX planning kit (~45 min for any idea)
- Audit assumptions → sort three ways.
- Hypothesize 2 personas → label → write their test questions.
- Flow the core task → demand unhappy paths.
- Wireframe text per screen → exact microcopy → you box it.
- Attack: edge cases ranked by trust damage + flow-specific accessibility → verify against WCAG.
Wireframe from text (paper or Figma)
- Take one screen's wireframe text.
- Draw the boxes top to bottom — 5 minutes, ugly.
- Write the microcopy into the boxes — real labels, not lorem.
- Mark ⚠ any element you overruled — your judgment trail.
- Connect the happy path; add one unhappy-path branch.
06Prompt Recipes
1 · The assumption auditI want to design "[vague idea]". Before any design: list every assumption hiding in that sentence — about users, problems, context, and devices. Then sort them into three groups: safe to assume / must research / dangerous to guess. Don't design anything yet.
2 · Personas as hypothesesDraft 2 contrasting persona HYPOTHESES for this idea. Label each clearly as "UNVALIDATED HYPOTHESIS". For each: context, the frustration that brings them here, their moment of need, and what they'd never do in an app. Then write the 5 interview questions that would test whether each persona is real.
3 · The flow with unhappy pathsCore task: "[the one job]". Write the task flow step by step. Then add 3 unhappy paths ([likely failures — bad input, no connection, ambiguous result]) and how the user recovers from each without a dead end.
4 · Wireframe text + microcopyDescribe a [5]-screen low-fi wireframe for this flow, in text. Each screen: its single purpose, key elements top to bottom, and the EXACT microcopy for buttons, error messages, and empty states. Grey boxes only — no visual styling, no colours, no branding.
5 · The edge case hunt + accessibilityNow attack this flow: list 8 edge cases or failure states we haven't handled, ranked by how badly each damages user trust. Then write an accessibility checklist specific to THIS flow (not generic WCAG boilerplate): the real risks for [photo capture / colour-coded status / outdoor use / screen readers] in this design.
Bonus · The microcopy de-roboterRewrite this microcopy in a voice that is [calm, plain, slightly warm]: [paste]. Rules: no exclamation marks, no "oops", no blame, every error says what happened and what to do next, reading age ~12.
07Lab: Edge Case Hunt (70 min)
Cold start friendly. No idea? Standing brief: an app that helps housemates fairly split chores without becoming the chore police.
- Kit prompts 1–3 (20 min): audit → labelled personas with test questions → flow with unhappy paths.
- Wireframe (20 min): prompt 4 for the text, then box it yourself — paper or Figma, 5 screens, microcopy written in, ⚠ marks where you overruled.
- The Hunt (15 min): prompt 5 → pick the 3 edge cases that most damage trust → amend your flow/wireframe to handle them. The amendment is the assessed thinking.
- Critique checklist (15 min): 6 lines (Week 2 method) from your brief + edge cases. Test against your own wireframe — at least one line should fail.
Zero-AI path (fully equal)
The question structures run beautifully on paper with a partner playing pessimist. Microcopy written by hand. Same deliverables, equal credit.
08Integrity Check
This week's four
- Is every persona still wearing its
HYPOTHESISlabel — everywhere it travels? - Did any "users want…" sentence sneak in without a source? Kill it or mark it.
- Accessibility claims: checked against actual WCAG, or model paraphrase?
- Portfolio language: "Research direction AI-assisted; all user claims are labelled hypotheses pending research." Honesty reads as method maturity.
09Troubleshooting
| Problem | Fix |
|---|---|
| Personas feel so plausible | That's the danger. Plausibility ≠ truth. Read the label aloud; write the test questions |
| Flow is suspiciously clean | You got the happy path. Demand the unhappy paths — that's where the design is |
| Wireframe text wants to be pretty | "Grey boxes only, no styling" — repeat; fresh chat if it persists |
| Microcopy sounds like a chatbot | Run the de-roboter; ban "oops" by name |
| Edge cases are generic ("slow internet") | Demand specificity to YOUR flow: "edge cases unique to photo-diagnosing plants outdoors" |
| Figma overwhelms | Paper. Genuinely. The thinking is identical and faster today |
10Resources
- ArticleLinkedAI as a UX Assistant — Nielsen Norman GroupThe respected research org's framework for which UX roles AI can play — matches this week's structure/evidence split.
- VideoLinkedAI Isn't Ready for UX Design — NN/gThe skeptical counterpoint from people who test claims for a living. The skeptics are right about the dangerous part — know their argument. (Not embeddable — opens on nngroup.com.)
- ArticleLinkedAI-Generated Synthetic Users vs. Personas — IxDFDirectly on this week's central rule: what synthetic users can and cannot tell you.
- StandardLinkedWCAG Quick Reference — W3CThe actual accessibility standard. Verify model claims here — paraphrase is not compliance.
- ToolLinkedMermaid Live EditorPaste AI-generated flow-diagram code, get instant diagrams. The fastest text-to-flow path there is.
- DocsVerify before teachingFigma AI features documentationFigma's AI tooling changes quarterly — check what exists on your tier the week you teach. The planning chain works regardless.
Optional extension
- Take your 5 persona test questions and interview one real person (a housemate counts). Note every place reality disagreed — that delta is what research is for.
- Build the clickable path: 5 frames in Figma, linked. With the plan done, it's shockingly fast.
- Hard mode: write the research plan that would retire your "dangerous to guess" assumptions: method, participants, questions, time.
Next week: 3D Workflows → — Blender scene scaffolds via Python, and the Topology Reality Check.