Your passenger is .
Decide who this must work for, and name what that might cost
You and your partner asked and answered questions through the Draw It Out activity that surfaced learner needs. Now we're turning our attention to our new passenger and adding a new dimension to design for. Your passenger may not be able to use all the elements of your gameful piece or prototype. Decide now that it must work for your passenger. State what that might cost.
This must work for your passenger, even if the cost is .
Name a real cost. Time, a feature you'd lose, a mechanic you'd have to change. If it costs nothing, look again.
Send a player through a piece of your game!
Paper: your description of a gameful piece, a turn, or a summary design brief. Nothing to work with? Use a design brief we call Muster Point. You'll find it below. If you plan to use it, please ask a coach or Carriann to come talk with you.
Pick one element of your prototype. One screen, one activity, one station. Copy the prompt below, paste it into your AI, and describe that element where it asks.
This is the whole walkthrough prompt with your passenger's line already in it. Nothing to add but your element.
Next, read what comes back from your AI tool. Then notice what it marked as a guess and what it stated as fact.
Muster Point, the design brief
Pick one of its five stations as your element. Copy this whole sheet to give your AI the context.
Muster Point
A ten-minute building-safety game for new hires.
Design Brief
Business case
New hires must complete building-safety orientation in their first week. Completion is at 61 percent; the safety office wants 95, and wants people to actually know where the exits are.
Learner
Jess, 27. Started Monday. Desk job at headquarters, laptop, second-floor open office. Plays word games on her phone at lunch. Wants to get orientation done and get back to real work.
Objective
By the end, Jess can name her nearest exit and muster point from anywhere on her floor and can pack a go-bag from memory.
Motivators
Curiosity (a puzzle, not a lecture), Order (one station at a time, progress always visible), Power (a rank to climb and a board to top).
Progression
Jess earns sea miles at every station. 200 miles makes her a Deckhand, 400 a Mate, 500 a Fire Warden. The badge shows on her profile and the department board. Three right answers in a row doubles the miles for the next one.
Bao's note
We need to move the completion rate from sixty-one percent to ninety-five percent, and a lot of people just don't want to do this boring training, just like they don't want to go to muster on the first day of a fun cruise. I made it something you'd want to finish, something more fun, themed like a ship because a building basically is one: decks, hatches, a muster point. I wanted it fast and a little tense, like a real drill. I want it to be fast and fun. And a little competitive, so people have a reason to finish.
Learning flow: five stations
- All hands. The safety officer's 90-second welcome plays as soon as the page opens; a lantern pulses in the corner until it ends. Three buttons underneath, Psst..., Peek, and Bonus round open the evacuation policy, the muster-point map, and the go-bag list.
- Chart the deck. A floor plan of the second floor. "You're at the galley. Which exit is nearest?" Jess clicks an exit on the image. A bell rings for a right answer and she banks 50 miles. The image is labeled: "Second-floor plan showing offices, hallways, stairwells, and four exits."
- Beat the bell. Five quick questions with a 45-second countdown, 20 miles each. When the clock runs out, the screen advances on its own to simulate the time pressure of having to think under pressure when you hear the alarm. Feedback flashes for three seconds; the streak bonus lights up when she hits three in a row.
- Pack the raft. Twelve items on the deck, a life raft on the right. Jess drags the six that belong into the raft's sea bag. The button says "Ace it" and the hint text under the raft says "Hang in there, rookie." A perfect pack unlocks the Warden badge early.
- Crew standings. Jess's miles post to the department board, ranked, with everyone's name and rank. Weekly standings go to the team channel so managers can nudge the stragglers. The top three fly a flag by their names.
Reference
Psst... opens the whole evacuation policy on one screen, top to bottom, so nobody has to hunt for anything. About a ten-minute read.
Mechanics
Narrative frame (join the crew), points (sea miles), rank ladder, badge, streak bonus, countdown timer, drag-and-drop puzzle, immediate feedback (bell and flash), leaderboard.
AI changes roles and revises only this element
Same conversation, same element. Your AI just walked that element as your passenger and named one change. Here it stops being your passenger and becomes your assistant. Now ask it to make the element work for your passenger, and to touch nothing else.
The change list it hands back is the point of this step: it shows every decision the AI made on its own, and that list is what your partner checks in the next steps. Copy this and paste it as your next message. Ask for nothing else.
Keep the change list it gives you on your screen. In two steps, your partner checks it.
Watch it first
Carriann and a coach audit one revision at the front. Watch for the three moves. You make them yourself in the next step.
- Count the changes. One asked for, how many came back?
- Check the guess, then check the fix. Was the guess right? Did the fix do more than the standard asks?
- Hand the decision back. You find; your partner decides.
This quality step is called verify and validate. Never yield your judgment to a tool. Apply your knowledge and wisdom to make this work well for the people you're designing for.
Critical Checks
Go to paper.
Put the screen down for this one. Pick up the Critical Checks card.
- Swap with your partner.
- Read what came back from their AI tool: the walk and the change list.
- Run the four Critical Checks on the card.
- Write your finding on their card and hand it back.
You find; your partner decides.
Prefer to work digital? See it here.
An auditor checks whether the work does what it claims. Not to admire it, not to fix it.
- Where did this come from?
- What did AI change that I didn't ask for?
- What does the learner still have to do themselves?
- Who would this fail without them knowing it failed them, and why?
Your finding, in this form
You didn't ask it to change this, but it did: .
It would still fail for because .
Write it on your partner's card and hand it back. They leave holding the finding about their own work.
Second round
Take your partner's finding back to your AI. Same conversation. Copy this, paste it, and type the finding where it asks.
A colleague found this when they checked your revision: [write or paste the finding]. Show me the evidence in the revision that they're right. Then tell me one more thing they could have caught and didn't. If any part of their finding is wrong, say which part and show me why. Don't defend the revision; judge the finding.
Did it agree with your partner in the first sentence? Don't accept an unsupported agreement. Require it to show you why. The evidence is as important as the verdict.
Bonus round: three lines that make any prompt honest
You've earned this. Every prompt you run can be made better, more inclusive, and more accessible with these three lines.
Add these lines to any prompt in any tool from now on to create more accessible and inclusive designs anytime you work.
- Name the standard and level: "Check this against WCAG 2.2 Level AA."
- Require the citation: "For each barrier, give the success criterion number and quote what it requires, then say whether your fix does more than the criterion asks."
- Require the admission: "List what you cannot check from a description and would need the built element or a real user to confirm."
Your tool will not volunteer the third line on its own. It has to be part of your workflow. Put all three where your tool keeps standing instructions, so they ride along with every prompt: a project or custom instructions in Claude or ChatGPT, a Gem in Gemini, an agent in Copilot. In a tool with no memory, keep them as a saved snippet and paste them in every time.
How to build a passenger simulation that's unlikely to lie
Give the AI one grounding line per passenger that says what the technology actually does (a screen reader reads the page in a line, one item at a time; an unnamed image is announced as "image"; when a timed screen ends, the content it was holding is gone). The cards carry these lines. Without them, the AI imagines.
Tell it to separate what it knows from what it is guessing, and to mark every feeling and decision as a guess. The output arrives pre-sorted: mechanical findings you can trust, behavioral ones you check.
An AI playing a passenger is a draft of a person. It is never a substitute for testing with real people, and it is weakest on lived experience (safety, confidence, what someone would never say out loud). Use it to find the mechanical barriers fast, then go ask a person.
One instruction flipped the finding in this session's script: state the standard, with its source. Asked to name Web Content Accessibility Guidelines (WCAG) 2.2.1 and quote it, it found the criterion asks for a timer that can be turned off, adjusted, or extended. It does not say remove the timer. When the tool sounds confident, name the standard and require the citation, then look it up.
Take it home
Welcome to the whole kit. Everything from this morning is in one place: all ten passengers, the four prompts, the Draw It Out and Critical Checks cards, the three lines, Muster Point, and what to read next.
Two habits to keep. Drawing out the constraint: deciding who this must work for, and what that costs. Checking what came back: finding where the AI made a decision you didn't make.