AI Guides › Creative & Design
Using AI To Get Unstuck On A Design, Not To Finish It For You
By Nigel Guy · 2 min read
Staring at a blank canvas or a design that's gone flat, the fastest relief is
to hand the whole thing to a tool and ask for a finished version. It works,
in the sense that you get something back quickly — and that's usually the
problem, because what you get back is a generic answer to a question you
hadn't actually finished asking yourself.
The rule: use AI to break a specific stall, not to skip past it — the
moment you're asking a tool to finish the whole piece rather than answer
one question, you've stopped designing and started outsourcing.
The habit worth building: the named stall
- Name the actual stall before opening any tool. Not "I'm stuck on the
design" — "I don't know what goes in the empty space top-right" or "the
colour palette feels flat but I can't say why." A vague stall gets a
vague, whole-piece answer; a named one gets a targeted one.
- Ask for options against the named stall only. Generate five
variations of the one element you're stuck on, not five versions of the
whole design. Keep everything you've already decided fixed.
- Take the reaction, not the output. The point of the variations isn't
to pick one and paste it in — it's to notice which one you react against
and why, which usually tells you what the actual answer was all along.
- Return to your own hand for the fix. Once you know what you're
actually looking for, make the change yourself. The tool's job was
unsticking you, not finishing the piece.
What to skip
Skip asking for "a few directions for this project" as an opening move — at
that scope, there's no specific stall to answer, and you'll get generic
direction because that's the only kind a request that broad can produce.
And skip repeating the loop past the first stall you clear — if you're
running it again ten minutes later, check whether you've quietly slid from
unsticking to outsourcing.
Guardrails
- This works because the stall is specific; a request this narrow against a
problem this vague won't help, and the fix there is to define the
project's direction first, by hand or in conversation, before asking a
tool for options at all.
- The variations a tool produces are drawn from patterns it's seen widely
before — useful for jarring you out of a rut, not a guarantee of
originality. Treat them as prompts for your own reaction, not as
finished candidates.
- If you notice you can no longer tell which decisions in a piece were
yours, that's a sign the tool has drifted from unsticking into
authoring — worth stepping back and finishing the next section by hand.
All 751 AI guides · JulieMango plans from £17/mo