05 · The discovery form
You write a prompt, you hit generate, and it doesn't start drawing. It puts a questionnaire in front of you first: what canvas ratio, who's the audience, how many slides, which visual direction. The official name is the discovery form.
Most people find it annoying the first time. It's also the single biggest difference between OpenDesign and asking any agent to write you a page.
Why it isn't a fixed questionnaire
The form isn't hardcoded. It's generated from your prompt, and it asks about the holes you left.
Write "make a deck about our new feature" and it will ask whether the audience is internal or customers, how many slides, whether there's data. Write "make a 12-slide Series A deck for investors, visually close to Linear" and it won't ask any of that — you already answered.
Which has a useful side effect: the more specific your prompt, the shorter the form. A long form is the tool telling you how much of the brief is still missing.
What it typically asks
Depending on the artifact type:
| Question | Why it can't guess |
|---|---|
| Canvas ratio | 16:9, 4:3, a vertical poster, and a social card are very different objects |
| Audience | Engineers and a board tolerate wildly different information density |
| Slide or screen count | Decides how the content gets cut |
| Tone and visual direction | Restrained editorial, or high-contrast launch energy |
| Brand and research positioning | Is there an existing brand, is there a reference to follow |
| Whether the data is real | If it is, use it; if it isn't, label it |
That last one matters most. One of OpenDesign's explicit rules is to use honest placeholders when it lacks data rather than fabricate numbers. Telling it in the form which numbers are real is how it knows which path to take.
Answering it usefully
One: describe the audience's state, not their job title.
Weak: product managers Better: product managers who have never heard of this project and have five minutes
The second lets it work out on its own that the first two slides have to carry all the context.
Two: give tone a reference object.
Weak: professional Better: like an internal memo, not a launch keynote
"Professional" is a word everyone hears differently.
Three: say plainly whether the data is real.
Weak: (blank) Better: the three numbers on slide 4 are real, everything else is illustrative — label it
Four: don't rewrite the whole brief in the form.
It's a supplement, not a second prompt. Anything you already said doesn't need saying twice.
What the difference looks like
Same prompt, two ways of filling the form:
Filled carelessly
Prompt: a deck about launching our AI support agent Form: audience "colleagues", 10 slides, everything else default.
Result: ten slides of generic business template, title-plus-three-bullets, abstract geometric art, and invented bar charts. Presentable, right up until someone asks where the numbers came from.
Filled properly
Prompt: a retro deck on our AI support agent, three months after launch Form: audience "support team leads plus one VP, all of them users of the system", 8 slides, tone "a retro, not a promo — say plainly what didn't work", data "ticket volume and first-contact resolution are real; CSAT isn't measured yet, mark it pending".
Result: eight slides that lead with the conclusion, three comparing before and after, two real figures charted and sourced, and a CSAT slide that says "pending" out loud.
The difference isn't the model. It's how much you told it.
When you can skip it
One case: you're exploring, and you only want a feel for the tool. Take the defaults, look at the output, then decide whether to do it properly.
Otherwise, two minutes on the form beats twenty minutes of regenerating.
Next
06 · Design systems — the form decides the content, the system decides the look.