ChatGPT Prompts for Graphic Designers: Briefs, Concepts, Client Emails

A graphic designer can turn a vague client brief into a usable one with eight ChatGPT prompts: brief interrogation, an assumptions list, a scope and price outline, concept routes, a feedback summary, a revision-limit email, an approval email and a project wrap-up. Give each prompt a redacted description of the job and your own rates and rules.

These prompts are written for a studio owner who has to protect time and margin, not for someone wanting a list of generic logo ideas. The creative direction stays yours; ChatGPT does the interrogating, structuring and email drafting that sit around it. Each prompt comes with a sample output and what I would change.

Follow me on Instagram@sagnikteaches

Keep client files out of consumer chats

Before any prompt, decide what stays out. Most designers have never opened Settings, then Data controls, where a personal Free, Plus or Pro account has "Improve the model for everyone" switched on until you turn it off. OpenAI's help page on data controls says the opposite default applies to Business, Enterprise and Edu workspaces: OpenAI does not use their content for training by default. A designer holding an NDA, an unreleased product name or a client's logo files should treat a personal plan as off-limits for that material.

Connect on LinkedInSagnik Bhattacharya

Practical rule: describe the job, never upload it. "A regional food-delivery start-up, launching in spring, needs a logo and packaging for three product lines" is enough context for most prompts. Leave out the company name, product names, prices and any file. Put placeholders such as [client], [product] and [launch date] in the prompt and fill them in your own document afterwards. A Temporary Chat adds a little more distance: OpenAI says those chats are not used to improve models, though they can be kept for up to 30 days for safety.

Subscribe on YouTube@sagnikteaches

If your contracts do not yet mention AI-assisted work, how to mention AI use in client contracts and proposals gives wording to add. It makes later conversations about tools much shorter.

Before you quote: interrogate the vague brief

Most revision rounds trace back to a brief that was never really a brief: "something modern, like our competitors but different". The first prompt turns that into a list of questions you can send the client. Ask for questions only, because a model that tries to answer will invent the client's preferences.

I am a freelance graphic designer. A client sent this brief (redacted):
[paste the brief with names replaced by placeholders]

Do not suggest designs. Write 12 questions I must ask before I quote or start, grouped under:
audience, purpose, deliverables and formats, brand constraints, approval process, budget and deadline.
For each question add one line on why it matters to the work.
Flag the three questions where a vague answer would most likely cause extra revisions.

Illustration: a bakery wants a new logo and a window poster. The sample output flags questions such as "Who signs off the final design, and is it one person or the whole family?", "Which formats do you need: print-ready PDF, a social profile image, a shop-window vinyl size?" and "Is there an existing logo that must stay recognisable?" The three flagged items were approval, formats and the existing logo.

Edit before sending: models write long questions, so cut each to a sentence a busy owner can answer in a line. Send eight, not twelve; a client who faces a dozen questions answers three. Keep the remaining ones for the kick-off call.

After the kick-off call: assumptions the client must confirm

When the client answers only half the questions, you need a record of the guesses you made. This prompt turns your notes into a short assumptions list that goes into the proposal or kick-off email.

From my notes below, write a numbered list of assumptions I am making about this project.
Each assumption: one sentence, plain English, something the client can confirm or correct with a yes or no.
Notes: [paste your call notes, redacted]
Include assumptions about: number of concepts, revision rounds, who supplies copy and photos, file formats, deadline, and who approves.
Do not invent anything not in my notes; if a point is unclear write "Needs confirming:".

A sample line: "Assumption 4: you will supply the final text and product photography by the 14th; delays to these move the delivery date by the same number of days." That sentence is a contract in miniature. Clients who reply "yes to all but 4" hand you the thing you needed to know before the work started.

Quoting: a scope-and-price outline built from your hours

This prompt does not set a price. It structures one. You supply the hourly or day rate and your estimated hours; the model lays out the quote so nothing is left off. For a fuller view of how AI affects what you charge, should you charge clients less when AI speeds up your work? covers the argument.

Draft a scope-and-price outline for a design project. Use my figures exactly and do not change them.
Rate: [amount] an hour. Estimated hours: discovery [h], concepts [h], revisions [h], final files [h].
Deliverables: [list]. Revision rounds included: [number]. Extra round price: [amount].
Not included: [list, for example illustration, photography, printing, trademark searches].
Format: a table of stages with hours and cost, then a bullet list of exclusions, then payment terms: [for example 50% to start, 50% on delivery].
Add one line explaining that extra requests are quoted separately.

Say the studio rate is $80 an hour for the bakery logo and poster: discovery 3 hours, concepts 10, revisions 4, final files 3. The model should produce 20 hours and $1,600, with the stage costs ($240, $800, $320, $240) in a table. The arithmetic here is simple but still recompute it, because models occasionally mis-add. A realistic error: the table showed $1,640 in the total row after one stage was rounded. Compare the sum with a calculator every time.

The exclusions list is where the value lies. Designers often forget to write "stock image licences are charged at cost" or "printing is not included". Ask the model for ten possible exclusions for this job type, then keep the ones that apply to your studio.

Concept stage: routes you then design yourself

This is the prompt people expect, and the one to use most carefully. The aim is not to have ChatGPT design for you but to widen the set of directions you consider and to sharpen the wording you present. A route has a name, a rationale and a visual direction in words.

I am designing a logo and packaging for [business type, redacted]. Audience: [audience]. Tone words the client chose: [three words].
Constraints: [colour limits, print method, existing brand elements].
Suggest five distinct concept routes. For each: a two-word name, the idea in one sentence, who it appeals to, one risk, and the sort of typography and colour logic it implies.
Do not describe logos that copy any real brand. Make the five routes genuinely different, not variations.

A sample output for a dog groomer might include "Clean Paws: a minimal wordmark with a single paw detail, appeals to owners who see grooming as a premium service, risk: looks like a vet clinic" and "Tail Wag: a hand-lettered script with a loose, friendly feel, risk: harder to reproduce small on a van." Pick two, sketch them, and discard the rest. Presenting three polished routes is easier to justify than five loose ones.

Use the output as a prompt for your own thinking. If a route feels familiar, search for it: models reflect common design patterns and may point you towards something existing. Rights and originality questions are covered in AI for freelance designers: faster concepts without rights issues, and the safe default is to design every final asset by hand or with tools whose commercial terms you have read.

Feedback rounds: from a rambling email to a change list

Feedback arrives as a long email with contradictory comments, a forwarded message from a cousin and a vague line like "make it pop". The summary prompt separates what is actionable from what needs a question.

Summarise this client feedback for a designer.
Feedback (redacted): [paste]
Output three lists: (1) Clear changes I can make now, each as one instruction; (2) Contradictions between comments, quoting both; (3) Vague comments I must ask about, each with a suggested question.
Do not add your own opinions about the design.

Illustration: feedback on the bakery poster says "bigger logo", "more white space" and "make it pop". The summary puts "increase logo by about 20%" under clear changes, flags that bigger logo and more white space conflict, and suggests asking "When you say pop, do you mean brighter colour, larger price text, or a bolder headline?". This one step usually saves a round of revisions, because the client sees their own contradictions in writing and decides.

What to watch: the model sometimes converts "bigger logo" into a number such as 20% that the client never said. Mark invented specifics as suggestions, not instructions, before sending the list back for confirmation.

Included rounds used up: the email that keeps the client

When the agreed rounds are used up, you need a message that is clear without sounding grasping. The prompt below drafts it from your terms.

Write a short, friendly email telling a client that the [number] included revision rounds are complete.
Facts: the extra round costs [amount] and takes [days]; the current version is [version name]; I need a decision by [date].
Give two options: approve the current version, or book another round.
Tone: warm and matter of fact, under 100 words. No apology, no discount, no invented terms.
Use [first name] and [your name].

A sample output states the facts in four sentences and ends with "Reply 'approved' to release the final files, or 'extra round' and I will send a short quote." The reply-word trick matters: it gives you a one-word record of the decision that is easy to search later. Fix any softener the model adds, such as "no pressure at all", that weakens the deadline.

Sign-off: an approval email that names the exact version

Disputes at the end of a design job usually come from one question: what exactly did the client approve? The approval email answers it. It names the version, the deliverables and the exclusions, and asks for a written yes.

Write an approval request email for a design project.
Version: [name, for example Poster v4]. Deliverables included: [list with formats and sizes].
Changes since the last version: [list].
Not included: [list].
Ask the client to reply with the word "approved" if every item is correct, and explain that after approval, further changes are quoted separately.
Tone: clear and polite. Under 120 words. Use only the facts above.

Sample output, labelled illustrative: "Hi [first name], attached is Poster v4 as an A2 print-ready PDF and a 1080 by 1080 social image. Since v3 we enlarged the logo and changed the headline colour to the green in your palette. Printing and window installation are not included. If everything is right, please reply 'approved' and I will release the final files. After approval, changes are quoted separately. Thanks, [your name]". Save the reply in the project folder with the file names and dates.

A realistic mistake: the model wrote "final files in all formats", a promise the designer never made. The client later asked for vector files and an animated version. Replace any phrase like "all formats" with the exact list. Compare the deliverables line to the quote before sending.

Hand-over: rights, archive and the next job in one note

The last email thanks the client, lists what was delivered, states what happens to source files and how long you keep them, and suggests one natural next step. It is also where you set expectations about support.

Write a project wrap-up email from a designer.
Delivered: [list]. Source files: [what the client gets, what I keep]. Usage rights: [as per contract].
Next logical project: [one suggestion]. File archive period: [for example 12 months].
Under 130 words, warm, professional. No free extras, no promises beyond those listed.

For the bakery job, the next step might be menu boards or loyalty cards. The model will often suggest three or four; keep one. A note on rights: only state usage terms that are written in your contract. If AI tools touched the work, check what each tool's licence allows before you promise a client exclusivity or trademark protection; how AI can help graphic designers with briefs, mock-ups and revisions covers where AI sits safely in the process.

Pricing the same job at three levels with one prompt

Studio owners lose money when a client says "that's more than I expected" and the only response is to cut. Offering three packages turns the conversation from whether to buy into which to buy. Ask ChatGPT to structure the tiers from your own hours, not to invent the content.

Using my figures, lay out three packages for a logo-and-poster project. Do not change any number.
Rate: [amount] an hour.
Essential: [hours], one concept, one revision round, files: [list].
Standard: [hours], two concepts, two revision rounds, files: [list].
Complete: [hours], three concepts, three rounds, files plus a one-page usage guide.
For each: total cost, what is included, what is not, and the sort of client it suits in one sentence.
Present it as a table, then two sentences recommending nothing: leave the choice to the client.

At $80 an hour with 12, 20 and 30 hours, the totals are $960, $1,600 and $2,400. The sample output should show those figures and a "suits" line such as "Essential suits a business that already has a clear idea and needs it executed." Check the totals with a calculator and read the "not included" column most carefully, because that is where tiers blur. If Essential and Standard differ only by one concept, the gap is too thin; widen it with a clearer difference such as a usage guide or an extra format pack.

Do not let the model decide the hours. If it proposes 15 hours for a logo, that is a guess based on nothing about your speed. Use your time records from the last three similar jobs. Tools that help you track time are worth setting up for this alone.

A feedback round walked through from email to change list

To see how the prompts chain, take a realistic round. A café owner, three days after receiving menu designs, writes: "Love it overall! Can the prices be bigger but also less cluttered? My business partner thinks the green looks too dark. Maybe try some other fonts? Also can we add the new catering offer, I'll send details later."

The feedback-round prompt turns that into: Clear changes: none are fully specified. Contradictions: prices bigger versus less cluttered. Vague: "too dark" (whose opinion, which green from the palette?), "other fonts" (how many, how different?), and the catering addition (the content has not arrived, so it falls outside this round). The suggested questions are short: "Shall I enlarge the prices by moving the descriptions to a smaller size, or remove the descriptions?" and "Which green would you like, one lighter shade from your brand palette or something outside it?".

The catering addition is the interesting one. It is new scope, so route it through the change-request wording from your assumptions list: new content supplied after the round starts is quoted separately and may move the date. A line in the style of the rounds-used-up email does the job, and the client keeps the choice. Notice that in this example the AI did no design at all. It cut a forty-word email into three decisions and a boundary, which is where most studios lose their evenings.

When the client is difficult: prompts for the awkward email

Some projects drift into disagreement. The model is useful here for one reason: it is not irritated. Paste your draft reply, redacted, and ask for a calmer version that keeps your position.

Rewrite my draft reply so it is calm and professional but keeps every point I make.
Draft: [paste, redacted]
Rules: do not concede anything, do not add apologies I have not written, do not add threats, keep all dates and amounts exactly as written.
Then list any sentence that could be read as an accusation, with a neutral alternative.

An illustration: a client insists the logo "was always meant to include a mascot" when the brief says nothing about it. The calmer draft keeps the facts (the brief of the 3rd, the approved concept of the 17th) and drops the sarcasm. The list of accusation-like sentences is useful: "as I already told you" becomes "as noted in the brief of the 3rd". Always reread the result for concessions that crept in; models like to be agreeable, and "I understand and will happily adjust" can undo your position in one line.

What not to ask ChatGPT in a design business

A few jobs are better kept away from the tool. Do not ask it to check whether a logo infringes an existing trademark; a chat answer is not a trademark search, even with web search switched on, and it will sound confident anyway. Use the official registers or a trademark professional. Do not ask it for colour accessibility results from a description; use a contrast checker on the actual hex values. Do not ask for legal wording on licences or ownership; draft your terms with a qualified adviser and use the model only to tidy tone. And do not ask for the "best" concept: the audience, not a language model, decides that.

One final habit is worth adopting: keep a short written rule for each tool in the studio, naming what it may be used for and what is off-limits. It turns an unspoken habit into a policy you can show a client who asks, and it helps any freelancer you subcontract to stay within the same lines.

A studio-owner workflow that strings the prompts together

For a typical small job, such as a logo and poster for a bicycle repair shop, the sequence runs like this. Before the call, run the brief interrogation to prepare questions. After the call, turn your notes into the assumptions list, then lay out the quote with your own hours. Send both together.

During the project, use the concept-routes prompt once at the start of concepts and the feedback summary after each round. When the included rounds run out, send the rounds-used-up email, and at sign-off send the approval request followed by the hand-over note. In total, a four-week job might involve eight prompts and about an hour of drafting time; the saving is not in the writing but in the clarity, because fewer misunderstandings reach the design stage.

A rough sum, illustrative: if unclear briefs cost you one extra revision round on one job in three, and a round takes four hours at $80, that is about $107 of lost time per job on average. Brief interrogation and approval emails will not remove all of it, but removing even half pays for itself quickly.

Four searches to run on every studio email draft

Read each draft against a fixed checklist. First, every number: rate, hours, dates, version names and totals should match your own records. Second, every promise: look for "unlimited", "all formats", "full rights" and "guaranteed", and delete any that you did not write into the contract. Third, tone: if a sentence would sound odd on the phone, rewrite it. Fourth, placeholders: search for square brackets so that no [client] reaches the recipient.

Keep a prompt file with one sample output per prompt that you liked, with the redactions left in. Next time, paste that sample as a model of the style you want. After a few jobs the drafts need little editing, and you will have a consistent voice across briefs, quotes and approvals. For help choosing which of these steps to automate further, that is the sort of thing my AI implementation consultation looks at with studio owners.

Questions designers ask about using ChatGPT with client work

Can I paste a client's brief or logo files into ChatGPT?

Not on a personal plan if the work is confidential or under NDA. Paste a redacted description with placeholders instead. Business workspaces are not used for training by default, but read your client contract first. Never upload unreleased artwork unless the contract allows it.

Should the concept routes come from ChatGPT or from me?

From you. Use ChatGPT to widen the list of directions and to test your reasoning, then design the routes yourself. A route you cannot defend in your own words is a route you should not present.

Can ChatGPT set my prices?

It can structure a quote and flag what is missing, but it does not know your costs, your market or your capacity. Give it your hourly or day rate and your hours, and treat its output as a layout for your own numbers.

How do I record client approval properly?

Send a short email that names the version, lists what is approved and what is not, and asks the client to reply with the word approved. Keep that reply with the project files. A chat message or a nod on a call is hard to prove later.

Further reads

• Midjourney for Graphic Designers: Client Work, Licences and Limits — Check licence limits before AI images reach client work.
• Adobe Firefly Commercial Use: What Designers Can Promise Clients — What you can promise clients about Firefly output.
• AI-Generated Logos and Trademarks: What Designers Can Promise Clients — Logo rights and trademark limits for AI-assisted work.
• AI Image to Vector: Clean Logos for Print and Signage Clients — Turn AI images into clean vectors for print.
• Why Is AI Bad at Maths? What to Check in Quotes and Invoices — Why quote totals from a chat always need recomputing.
• AI Automation for Graphic Design: Resizing, Handoffs and Proofing — Which studio production jobs can run on presets, scripts and AI, and the checks a designer keeps for resizing, handoff folders and proof rounds.
• AI for Graphic Design Layouts: Presenting Options for Client Approval — A freelance workflow for AI-assisted layout options: moodboard, three directions, capped revision rounds, licence notes and a signed approval.
• How to Use AI for Graphic Design: Brand Kits and Templates for Clients — A studio workflow for AI-assisted brand kits and templates: licence checks per tool, what to build, what to hand over and the clauses that protect you.
• Design Studios Using AI: Briefs, Moodboards and Client Revisions — A studio-side workflow for moving from a messy brand brief to approved visual directions and controlled revision rounds.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: OpenAI Help Center page on data controls in ChatGPT, including Temporary Chat retention (checked 4 November 2026).