The client's brief arrived as a voice note: "Something fresh, like Apple but friendly, you know what I mean, and can we have it by Friday?" Three days later you send a logo concept and get back, "Hmm, not quite feeling it. Can it pop more?"
Freelance design goes wrong in a predictable place: the words between you and the client. AI prompts for freelance designers help most before and after the design work, not during it. Before, they turn vague requests into a brief the client can confirm. After, they turn vague feedback into specific options. They don't pick the typeface or judge the layout. Those are your job.
All prompts below are untested templates, and the worked example is invented and labelled. This post is different from the one on architects and interior designers, which handles scope and change requests for building projects. Here the focus is graphic, brand and web designers: intake, feedback and revision limits. For proposals and fees, see AI prompts for consultants and freelancers. Nothing here is legal advice; contract wording is for your own agreement or a professional.
How do I turn a vague client request into a one-page brief?
Run the request through a prompt that forces structure and separates what was said from what is guessed.
You are helping a freelance designer write a one-page creative brief from a
client's message. Output:
1. Project in one sentence
2. Business objective (what should change if this works)
3. Audience (only what the client said)
4. Deliverables and file formats [STATED / NOT STATED]
5. Deadline and what drives it
6. Must-haves, must-avoids (including brand rules, colours, legal lines)
7. References the client likes, and what exactly they like about each
8. Budget or scope limits [STATED / NOT STATED]
9. Decision maker and who else will give feedback
10. Open questions, ordered by how much each could change the work
Rules:
- Label each item [STATED] or [INFERRED]. Use as few inferences as possible.
- For every vague word (fresh, clean, modern, friendly, premium, pop,
"like Apple"), list under VAGUE TERMS with three concrete readings I can
ask the client to pick from.
- Do not invent numbers, dates, competitors or audience facts.
CLIENT MESSAGE (names removed):
[PASTE]
Send the output back to the client as "here's what I understood, please correct". The goal is agreement on words before pixels. A client who replies "yes, exactly" has just given you something to point to later; one who replies with six corrections has saved you a redesign.
What should my intake questions be?
Questions that make the client choose, not describe. People find it hard to describe taste and easy to react to examples.
I'm a freelance [LOGO / WEB / SOCIAL / PACKAGING / DECK] designer. Write an
intake questionnaire for a [TYPE OF CLIENT: new small business / startup /
agency / nonprofit] project.
Group questions as:
- Business and goal (3 to 4 questions)
- Audience (3)
- Taste: "show me" questions, such as pick three brands you admire and one
you'd hate to look like, and say why (3)
- Practical: existing assets, fonts, colours, approvals, deadlines (4 to 5)
- Process: who approves, how feedback will be given (2)
Keep every question answerable in two sentences or one pick-from-list.
Mark the five that most often change the scope of the work.
Total 20 questions maximum.
Cut it down further before you send it. A 20-question form sent to a small client often gets half answered; the five scope-changing questions are the ones worth chasing.
How do I turn "make it pop" into something I can act on?
This is the highest-value use. Feed the model the exact words and the design context, and ask for readings and a question, not a verdict.
Client feedback, verbatim: "[PASTE]"
What I sent: [DESCRIBE THE DESIGN: layout, colours, type, purpose]
Brief highlights: [PASTE 3 to 5 LINES OF THE AGREED BRIEF]
1. List 3 to 4 plausible readings of this feedback. For each:
- the specific change I'd make
- a one-line risk ("this might make it look too loud for the audience")
2. Write one short question I can send that tells me which reading is meant.
Offer the options as a pick-from-list so the client can answer in a word.
3. Say which parts of the feedback, if any, conflict with the agreed brief,
quoting the brief line.
4. Do not guess what the client "really" means beyond these readings.
Item 3 is where this pays off. If the client now wants a muted palette and the brief says "bold, high-energy", you can say so politely, with the line quoted. Then it's a conversation about the brief, not a verdict on your taste.
For longer feedback threads, ask the model to produce a table of every comment, tagged ACTIONABLE, NEEDS CLARIFICATION, CONFLICTS WITH BRIEF or OUT OF SCOPE. Check the tags yourself; they are suggestions.
How do I keep revision rounds under control?
Write the revision policy in your agreement, then use AI for the reminders, not the rules.
My agreed terms (paste from my contract; do not alter them):
[NUMBER OF REVISION ROUNDS, WHAT COUNTS AS A ROUND, WHAT COUNTS AS NEW WORK]
The client just sent: "[PASTE]"
Rounds used so far: [N] of [M].
1. Is the request a revision within the current concept, or new work?
Quote the term that decides it. If unclear, say UNCLEAR.
2. Draft a polite reply (under 100 words) that confirms what I'll do,
says where this sits in the round count, and states what happens if it
is outside the agreed terms.
3. Do not state any fee, legal consequence or rights position. Mark
[CONTRACT WORDING: take from my agreement] where one would normally appear.
The model is told not to invent terms. If your agreement doesn't define a "round", that's the thing to fix next, with a professional's help if the project is large.
A short reply to chase feedback or payment
Write a friendly three-sentence follow-up. Situation: [I sent the proofs on
DATE and haven't heard back / invoice N is due on DATE]. Next step I need:
[DECISION / PAYMENT]. Say what happens to the timeline if I don't hear back
by [DATE], using only what I've stated. No guilt language.
Keep payment-term specifics out of the prompt unless they're in your contract, and send the message yourself.
A worked example (illustrative)
An invented scenario, not a real client. A café owner sends: "Make the menu board feel more cosy and premium, but still affordable. Like the places on Instagram."
Running the brief prompt should surface, under VAGUE TERMS, something like:
- Cosy: warm colours and textured paper, or softer lighting-style imagery, or rounded friendly type?
- Premium: restrained palette and generous white space, or gold-and-black?
- Still affordable: does that mean price points visible, or an approachable tone?
- Like the places on Instagram: which three accounts, and what do you like about each?
That last question usually changes everything, because the client's references tell you more than their adjectives. In this invented case, the owner replies with two accounts that both use a cream background and hand-drawn type. Now you have a direction to design toward, and you've written it down.
Later, when the owner says "can it pop more?", the feedback prompt offers four readings: bigger prices, a stronger accent colour, more contrast between sections, or more imagery. You send the pick-from-list question and get "stronger accent colour" in one line.
Failure modes to check for
| What goes wrong | Why | What to do |
|---|---|---|
| Brief states audience facts nobody gave | Model fills gaps with plausible detail | [STATED] / [INFERRED] labels; client confirms |
| "Pop" gets one confident answer | Model picks the commonest reading | Ask for several readings and a pick-from-list question |
| Reply sounds like a template | Generic phrasing | Add your own voice sample; edit before sending |
| Revision terms invented | Model improvises policy | Paste real terms; forbid altering them |
| Client files in unapproved tools | Convenience | Placeholders; check tool terms |
| Design critique treated as authority | Model can't see your work as a client does | Use it for questions, not verdicts |
What not to automate
- Design decisions and taste. The model can list options; choosing and defending one is the job you're paid for.
- Contracts, fees and rights. Ownership, licensing and liability wording come from your agreement or a professional. For rights on AI-generated images specifically, read AI image licensing and commercial use by tool.
- Telling the client how you work. Check what your client's rules allow and say plainly how you use AI; what to tell clients about AI use has wording to adapt, and the AI use policy template helps if you work with a team.
- Sending feedback replies unread. A reply with the wrong round count or a promise you didn't make is worse than a slow one.
- Client assets in unchecked tools. Unreleased brand work is confidential.
For the basic habit behind all of this, giving context explains why the brief matters more than your phrasing. If you're building a brand kit for yourself or a client, the logo and brand kit workflow is the practical next step, and the prompt library has more templates to adapt.
