ChatGPT Prompts for Designers: Briefs, Critique and Research

Good ChatGPT prompts for designers turn a vague assignment into explicit questions, connect research themes to evidence and evaluate options against agreed criteria. Supply the goal, constraints and source material, then ask for a defined output. Use the result to support your design judgment and real research.

The reusable part is usually the procedure rather than the conversation. fenbs can keep prompts and approved design decisions available to connected assistants. For the wider workflow across a design project, see ChatGPT for designers.

What should a design prompt include?

Name the decision the output should help you make. “Improve this screen” gives almost no criteria. “Identify obstacles to completing this appointment form on a phone” tells the model what to examine and gives you a way to judge whether its response is useful.

Add the audience, intended outcome, constraints and known evidence. Explain what is settled and what remains open. If the team already approved the navigation structure, a critique of labels should not silently reopen that structure or invent a new product strategy.

OpenAI recommends specific instructions and iterative refinement↗ (opens in a new tab). Apply that by asking for a bounded first pass, reviewing it and correcting one problem at a time. Give an example of the response format when different teammates need comparable results.

• Goal: the user action or design decision this work supports.
• Audience: who needs the design, including relevant access needs and use conditions.
• Evidence: approved research notes, requirements or design descriptions with stable labels.
• Constraints: required content, platform, brand rules and decisions already made.
• Output: questions, a structured comparison or a critique with evidence and uncertainty.

How do you clarify an incomplete design assignment?

Ask ChatGPT to surface uncertainty before proposing solutions. It is easy for an attractive concept to hide an unanswered business question. Separate facts supplied by the requester from assumptions about users, content, delivery and success.

Help clarify this design assignment before proposing concepts.
Assignment: [paste approved, nonconfidential project summary]
Audience: [known audience and use setting]
Constraints and prior decisions: [list]
Evidence available: [label E1, E2...]

Return: confirmed requirements; assumptions; contradictions;
missing information; five questions ranked by their effect on
the design. For each question, say who should answer it and
which decision depends on it. Do not invent user needs, metrics,
approvals or research. Keep solution ideas in a separate list
and label them as proposals.

For example, a fictional community workshop asks for an easier registration page. The most useful first question may be whether people need to register several family members together, rather than which button color to use. Have the project owner answer before treating the question as a requirement.

Use the general work prompt templates for other intake formats. Preserve the distinction between confirmed requirements and suggestions when adapting them. That distinction matters more than asking the assistant to adopt an impressive professional persona.

How can you synthesize research without fabricating findings?

Use only research material you are authorized to share, with participant information removed under your research process. Participant labels do not automatically make sensitive narratives anonymous. For practice, use fictional notes and make their fictional status explicit.

Organize the supplied research notes without adding findings.
Research question: [fill in]
Material: [approved notes labeled N1, N2...]
Sample and method limitations: [supply known limitations]

Return a table: theme; note IDs; supporting observation;
contradictory evidence; confidence limitation; next question.
Separate observed behavior, participant opinion and your
interpretation. Preserve disagreements. Do not create quotes,
participants, percentages or claims about all users. Use
"Not established" where the supplied notes cannot support
a statement. End with hypotheses to test, labeled as hypotheses.

Trace every theme back to the original notes. Check whether a theme represents several independent observations or one person’s repeated comment. Do not turn a small qualitative sample into a population estimate. A polished summary should preserve the limitations you would explain in a live research review.

Keep the synthesis open to disagreement. A note that contradicts the main pattern may reveal a different use setting rather than bad data. Ask what additional observation would distinguish competing interpretations, then plan actual research instead of letting the model settle the question.

Which prompt makes design critique more actionable?

Anchor critique to a task and agreed criteria. Generic requests for a “modern” or “clean” design tend to produce taste-based advice. Ask for observations, their possible effect on the user task and a way to check the concern.

Compare design options A and B for [specific user task].
Options: [describe each option precisely]
Audience and setting: [fill in]
Approved criteria and constraints: [numbered list]
Known evidence: [source IDs]

Return columns: criterion; observation about A; observation
about B; evidence; uncertainty; suggested test. Do not invent
user preferences or assume a visual choice proves usability.
Keep accessibility concerns separate from aesthetic opinions.
Do not declare either option validated. Finish with the
smallest practical test that would inform the open decision.

If the assistant cannot inspect a design directly, provide an accurate description and state that limitation. If you supply an image through an available product feature, check which details the response actually recognized. Never infer that it saw an interaction, keyboard behavior or hidden error state from a static screen.

Accessibility needs actual evaluation. W3C’s evaluation guidance↗ (opens in a new tab) explains that tools alone cannot determine accessibility and knowledgeable human evaluation is necessary. Use a prompt to organize checks, then inspect the implemented experience with the appropriate tools and people.

How do you turn critique into a clear handoff?

Summarize the person’s decision after the review, including the reason and any condition. “Use option B” leaves the next designer guessing which details were accepted. “Use B’s shorter form; keep A’s help text; error states still need review” gives the next person a concrete boundary.

Organize these approved meeting notes into a draft handoff.
Notes: [nonconfidential notes with speaker and item labels]

Return: decisions explicitly made by a person; proposals still
open; follow-up actions; unanswered questions; checks remaining.
Preserve the named decider and any conditions. Do not turn
silence, a suggestion or a positive comment into approval.
For each action, use only an owner and date stated in the
notes; otherwise write "Owner/date to confirm." Identify
conflicting statements for a person to resolve.

Have the participants review the handoff. An assistant can organize what they said, but cannot know whether an unclear remark was intended as authorization. Keep links to approved artifacts in the design tool and record the current decision separately from discarded alternatives.

Save that approved decision with the reusable procedure in fenbs’s work record. Connected assistants can start from the same instruction instead of reopening a choice because its reason disappeared into yesterday’s chat.

How should you review a reusable design prompt?

Test the procedure with a normal example and an awkward one. Give the research template contradictory notes. Give the critique template two options that both fail a required constraint. The answer should expose those problems rather than invent a winner.

The system prompt examples provide ways to organize standing instructions. Keep one current version of each procedure, with its purpose and review date. Avoid storing every alternate wording just because one response sounded good.

How to do this with fenbs

1. Create a free account and add one design project.
2. Save a prompt in Skills and prompts with the audience, required inputs and example output format.
3. Record a person’s approved constraint in Decisions and rules, including its reason and scope.
4. Follow the integration guide for your assistant and ask it to read the prompt and project rules before critiquing.
5. Save progress in Where we left off, including which design question needs research or a person’s decision next.

fenbs does not replace a design canvas, conduct user interviews or certify accessibility. It fits the reusable instructions, decisions and next steps around those activities. The AI prompts hub has additional examples for that work.