Describe what you need clearly
Whether it turns out right depends mostly on how clearly you describe it. This chapter gives you a fill-in-the-blank formula that works.
The universal template for describing a request
Fill in the blanks in the sentence below and you'll already be clearer than most people:
I want to build a what for who to solve what problem. It needs to be able to one, two, three, but not four, five, six.
For example:
| Blank | Example |
|---|---|
| Who | 5 colleagues at our company who handle foreign trade |
| What | A customer follow-up tracker |
| What problem | Customer info is scattered across each person's WeChat and Excel, so no one can see it all |
| Needs to be able to | Add customers, log every interaction, sort by next follow-up time |
| But not | Skip complex permission systems for now, just use it internally |
Stuck for words? Try "Enhance prompt"
There's an "Enhance prompt" button next to the input box. Click it and the AI rewrites your sentence more clearly before acting on it. Especially useful in two cases:
- You can only say "make a budgeting app" and don't know what else to add
- You wrote a long paragraph but the key points are jumbled together
You can review the rewritten version first and edit it if it's not right.
Give it the materials directly
Click "Add" on the left of the input box → "Upload file", and drag your files straight in.
Screenshots: as design references
See an interface you like? Drop the screenshot into the chat with a note like "follow this style". It will reference the layout and colors.
Images are a Beta feature: they guide layout and design but may not be reproduced exactly, and can't be used as in-app assets directly. For precise results, add a one-line note describing the key requirements.
Documents: as requirement materials
| Uploadable types | Good for |
|---|---|
| PDF / Word / PPT / Excel | Requirement docs, competitor research, data tables, design specs |
| md / txt | Outlines, copy, specification notes |
| env | A list of environment variables you'll need |
After uploading, it reads the document content and works from it together with your description.
Typing is a hassle: just talk
Click the microphone icon to start speaking, click stop when you're done, and your words are transcribed into the input box automatically. Speaking as you think is much faster than typing.
Pick the right type first: App / Slides / General
You can switch the type near the input box — the wrong choice produces something completely different:
| Type | What you get | When to use |
|---|---|---|
| App | An interactive app with a UI, optionally with a backend and data | Tools, websites, mini games, internal systems (pick this in most cases) |
| Slides | A complete presentation | Reports, pitches, training materials |
| General | A freer blank canvas | When you just want it to write something or run experiments first |
The Slides type also comes with a presenter mode, fullscreen presentation, and "Export PDF".
Plan first, build later (Plan mode)
Not sure what it should look like? Turn on "Plan" next to the input box first.
- While it's on, the first round only analyzes and proposes a plan — no code is written
- If the plan looks right, turn off "Plan mode" in the chat and say "build it according to this plan"
- If it doesn't, just say "change point 2 to…" and ask for a revised plan
This step costs very few credits but saves a lot of rework later.
It may ask you questions back
Before starting, it sometimes shows "follow-up questions" about points it's unsure of, like "do you want sign-in?" or "where should the data be stored?". Answer what you can — the result will be closer to what you want; if you'd rather skip, click "Skip all questions" and it will decide on its own.