AI Guides › Workbench
By Nigel Guy · 8 min read
The usual way people try this: ask ChatGPT for "a profitable app idea", accept the first plausible one, ask it to build the thing, and post the link the same afternoon. Every step produces something shiny, so it feels like progress. You end up with a generic idea, a backend you never checked, and a public page collecting email addresses with no privacy notice.
The rule: make ChatGPT interview you before it suggests anything, review a private preview before anything goes public, and sort out the UK data-protection basics before you ask for a single sign-up.
This kit is four prompts run in order inside ChatGPT Work, with ChatGPT Sites doing the build and hosting.
| Stage | What it does | Cost at time of writing | Best for | The catch |
|---|---|---|---|---|
| 1. Interview (Work, by voice or typing) | Pulls your hobbies, skills and recurring annoyances out of you, then shortlists ideas | Included in Plus or Pro; Voice on web/mobile uses your daily Voice allowance | Ideas grounded in your own life | "Profitable" is a guess until someone pays |
| 2. Plan (same Work chat) | Turns the chosen idea into a feature list, data model and hosting plan | Draws from your shared Work and Codex allowance | Knowing what the backend stores | Must fit what Sites supports |
| 3. Build and deploy (Sites) | Builds the app, shows a private preview, publishes to a URL | Sites needs Plus, Pro, Business, Enterprise or Edu; not Free or Go. Public-beta usage limits apply | A small app or landing page | Every deployment URL is production; no UK data residency |
| 4. First 100 sign-ups (Work or Chat) | Produces a launch plan sized to your time and audience | As above | A week-by-week plan | UK marketing rules still apply |
On price: OpenAI's public pricing page showed no UK pound figures when checked, so no £ amount is quoted here. Check the price in your account before upgrading. Sites and Work on web both need a paid plan; Free and Go get only limited Work on the desktop app and no Sites.
Paste this into Work, then switch to voice if you prefer to talk.
You are a product researcher interviewing me so we can find one small web app I could realistically build, launch and charge for on my own.
About me: [YOUR_SITUATION — e.g. full-time job, 5 hours a week spare, based in the UK]
Budget I can spend before any revenue: [£_BUDGET]
How to run the interview:
1. Ask one question at a time. Wait for my answer before the next one.
2. Cover three areas in this order: hobbies and things I already know well; ideas I've had and dropped; small problems I hit every week, at work or at home.
3. When an answer is vague, ask for a recent specific example.
4. Stop after about 12 questions, or sooner if I say "enough".
Then:
5. List 3 candidate app ideas drawn only from what I told you. For each: who it's for, the problem in one sentence, what someone would pay for, and the single riskiest assumption.
6. Recommend one, and say why the other two lost.
Rules: don't invent market sizes, revenue figures or competitor names; say what I should go and check instead. Before giving the shortlist, confirm every idea traces back to something I said.
Fill in: your situation and your budget in pounds.
Same chat, straight after the shortlist.
Take the idea we chose: [IDEA_NAME, or "your recommendation"].
Write a build plan I can hand to ChatGPT Sites. It will be hosted on Sites, which supports a D1 relational database for saved records, R2 storage for uploaded files, and Sign in with ChatGPT.
Include, in this order:
1. The smallest version worth launching: 3 to 5 features, and what is deliberately left out.
2. Each screen and what a visitor does on it.
3. The data the app stores: table names, fields, and which fields are personal data.
4. How sign-in works, and whether visitors need an account at all.
5. Anything this plan needs that Sites may not support (background jobs, raw TCP, special frameworks). Flag it rather than designing around it silently.
6. A short list of the personal data collected and why each item is needed.
Format: numbered headings, plain English, no code yet. If something about the idea is still unclear, ask me before writing the plan.
Fill in: the idea name. The personal-data list in point 6 matters in stage 4.
Still in Work. Mentioning @Sites or the word "website" starts the Sites flow.
@Sites Build the app from the plan above as a website.
Constraints:
- Use D1 only for the records listed in the plan; don't store anything extra.
- Add a privacy notice page explaining what is collected, why, how long it's kept, and how to ask for it to be deleted. Leave [BUSINESS_NAME] and [CONTACT_EMAIL] as the contact details.
- Any sign-up form needs an unticked opt-in box for marketing emails, separate from creating an account.
- No payments yet. If the plan needs them, leave a placeholder and tell me.
Show me a private preview first. Don't deploy until I say "deploy". Before you show the preview, list anything from the plan you couldn't build.
Fill in: business name and contact email.
What happens next, according to OpenAI's documentation:
You're a launch planner for a one-person UK business. The app is live at [SITE_URL] and does [ONE_LINE_DESCRIPTION].
My audience: [WHO_IT_IS_FOR]
Where they already gather: [COMMUNITIES_FORUMS_PLATFORMS, or "unsure"]
Time I can give each week: [HOURS]
Plan the route to the first 100 sign-ups:
1. Three channels, ranked by fit, each with the first action I can take this week.
2. A 4-week schedule, one table row per week: action, time needed, what counts as success.
3. Draft copy for the first post or message in each channel.
4. What I should measure each week, using the Sites analytics view (unique visitors, page views) plus sign-up count.
Rules: no buying email lists, no messaging people who haven't opted in, no fake reviews or invented user numbers. Keep every tactic compatible with UK GDPR and PECR. If you don't know where my audience gathers, ask instead of guessing. Before you answer, check that each week is doable in the hours I gave you.
Fill in: URL, description, audience, where they gather, hours per week.