Journal

How to use AI to turn messy client calls into a clear design brief

A practical workflow for turning a rambling kickoff call into a brief you can design against, with AI doing the sorting and you making the decisions.

How to use AI to turn messy client calls into a clear design brief

Most design projects start with a call that wanders. The client talks about their competitors, a website they saw last week, a launch date that may or may not be real, and three different ideas of who the customer is. You leave with forty minutes of recording, a page of half-sentences, and the job of turning it into something you can design against.

AI is good at the boring half of that job: sorting, grouping and spotting gaps in a long, messy transcript. It is bad at the other half: deciding what actually matters for this client. This guide shows how we split that work, so you get a clear brief in an hour instead of an afternoon, without handing your judgement to a chatbot.

Record and transcribe the call properly

Everything downstream depends on a clean transcript. Ask for permission to record at the start of the call and say why: "so I can listen to you instead of taking notes." Most clients are happy to agree.

  • Use the built-in recording and transcription of whatever tool you already meet in (Zoom, Google Meet, Teams), or a dedicated transcription app.
  • Make sure speakers are labelled. "Client" and "Designer" is enough; the model needs to know who said what.
  • Skim the transcript once and fix names, product terms and numbers the transcription got wrong. A wrong price or date will travel straight into your brief.

Before you paste anything into an AI tool, check your contract or NDA. If the call included confidential numbers or unreleased plans, remove them or use a tool your client has approved.

Ask for extraction, not a summary

The most common mistake is typing "summarise this call." You get a tidy paragraph that sounds right and drops the details you needed. Instead, ask the model to pull specific things out of the transcript and to quote where each one came from.

A prompt that works well for us:

From this transcript, list: the business goal, the target audience, the main problem the project should solve, deliverables mentioned, deadlines, budget signals, brand constraints, competitors or references, and decision-makers. For each item, quote the exact line it came from. If something was not discussed, write "not discussed". Do not guess.

The quotes are the important part. They let you check every claim in seconds, and they stop the model from filling silence with plausible inventions.

Find the contradictions and the gaps

Clients often say two incompatible things in the same call: "we want it to feel premium" and, ten minutes later, "it should feel friendly and cheap to try." Neither is wrong, but your brief has to choose or explain the balance.

Run a second pass on the same transcript:

  1. Ask the model to list statements that conflict with each other, with quotes for both sides.
  2. Ask it which standard brief items were never discussed, such as success metrics, content ownership or who signs off.
  3. Ask it to turn each conflict and gap into one short, neutral question you could send to the client.

This step is where AI saves the most time. Spotting a contradiction between minute 4 and minute 37 is tedious for a person and easy for a model.

Write the brief yourself, using the extraction as raw material

Now you have a list of facts with sources, a list of conflicts and a list of open questions. This is the point to close the chat window and write. The brief is where you show the client you understood the business, and that only works if the reasoning is yours.

A structure we use for most website and product projects:

  • Context: who the client is and why this project exists now.
  • Goal: one measurable outcome, for example more demo requests or fewer support emails.
  • Audience: the primary person you are designing for, in two or three sentences.
  • Scope: what is in, and just as clearly, what is out.
  • Constraints: brand rules, platform (for example Framer), content readiness, deadlines.
  • Open questions: the gaps and conflicts from the previous step.

Keep it to one or two pages. A brief nobody rereads is not doing its job.

Use AI as a second reader, not a co-author

Once your draft is written, give the model both the brief and the original transcript and ask a narrow question: "Does anything in this brief contradict or go beyond what the client said? Quote the transcript for each issue."

This catches the places where you projected your own preferences onto the client. It is also a good check on tone. Ask whether any sentence would read as a promise you have not agreed to, such as a delivery date or a result you cannot guarantee.

Do not ask it to "improve" the writing. You will get longer sentences and vaguer words, and the brief will stop sounding like you.

Send the brief back for confirmation

A brief only protects you once the client has agreed to it. Send it within a day or two of the call, while the conversation is fresh for both of you. Put the open questions at the top of the email, not buried at the end of the document, and ask for a simple reply: "Does this match what you have in mind? Anything missing?"

When the answers come back, update the brief and save the final version with the date. If scope questions come up later, this is the document you point to, and it is far easier to point to than a forty-minute recording.

A simple workflow you can reuse

Put the steps above into a repeatable routine so you are not rewriting prompts for every client:

  1. Record, transcribe and correct the transcript.
  2. Run the extraction prompt and check the quotes.
  3. Run the conflicts-and-gaps prompt.
  4. Write the brief yourself in your standard template.
  5. Run the second-reader check against the transcript.
  6. Send it to the client with the open questions first.

Save your prompts in a note or a text expander so the whole routine takes one paste per step. After a few projects, you will also start to notice which questions you keep forgetting to ask on calls, and you can add them to your call agenda.

Conclusion

A messy kickoff call is normal. What matters is how quickly you turn it into something clear enough to design against and agree on. Let AI do the sorting, quoting and gap-spotting, and keep the decisions, the priorities and the writing for yourself. You will start projects faster, with fewer surprises in week three, and clients will notice that you listened.

Oybek Ruziev

Designer & Framer developer · AidanCreates

Keep reading

All posts