Claude Prompts for Agency Creative and Production Teams

It’s 4:15 on a Thursday and the client has replied to the proof.

Nine notes, buried in three paragraphs. Two of them contradict each other. There’s a “quick thought” that landed in Slack twenty minutes before the email, and a comment left directly on the file that nobody has opened yet. Somewhere in that pile is a decision about whether the headline changes, which determines whether tomorrow is a normal day.

Someone on your team now has to read all of it, work out what the client actually wants, decide who does what, and put dates on it.

That job has no name. It isn’t on the SOW. It isn’t in anyone’s title. It’s just where four or five hours a person go every week.

The most useful Claude prompts for agencies all do the same thing in three different places: they take a messy input and hand back a structured artifact. Kickoff notes become a brief. A feedback round becomes a deduplicated revision list. A client email becomes a task list with owners and dates. The move underneath all of them is one line: give Claude the raw material, then name the exact output format you want back.

The short answer

• The one rule that makes any of these work: name the output format and the sections. Ask for “a summary” and you get prose. Ask for a deduplicated list grouped by copy, design, and layout and you get something a designer can start on.
• The instruction most people leave out: tell Claude to park anything the source doesn’t cover under “Open Questions” instead of filling the gap. That single line is the difference between a brief and a brief with invented facts in it.
• The limit worth knowing: Claude sorts. It doesn’t know the client’s history, your margin, or why the CD killed the last round.
• Start here: run the revision-summary prompt on your next feedback round, then compare it to what you would have written by hand.

Bottom line: the value isn’t that Claude writes for you. It’s that the twenty minutes between a client email and a plan somebody can execute stops being a person’s evening.

An ER Has a Triage Nurse. Your Agency Has Whoever Opened the Email.

Emergency departments figured out a long time ago that the hardest part of the job happens before treatment starts. Everything arrives at once, unsorted, and some of what’s loudest isn’t what’s urgent. So triage became a discipline. A defined role, a protocol, a scoring system, training, and a person whose entire job is to look at the unsorted pile and turn it into an ordered one.

Agencies do triage forty times a week and have never once called it that.

Client feedback lands in four channels at different hours from three people with different amounts of authority, and the sorting gets done by whoever happened to open the email first, usually at the end of a day they’d already scheduled. There’s no protocol. There’s no scoring. There’s frequently no record of the decision anywhere except in that person’s head, which is a problem you discover about eleven days later when they’re on PTO.

It is, to be clear, structurally odd that this is the default. The agency has a process for timesheets. It has a process for onboarding a new hire. The thing that determines whether the next round of work is correct gets handled ad hoc, by whoever’s around.

And the part that’s uncomfortable: nobody on your team has ever logged this work as what it is. It goes down as account management, or it doesn’t get logged at all. Which means when the retainer conversation comes around and someone asks why the hours look like that, you have no way to point at it. The work you can’t name is the work you can’t bill.

Most of the AI-for-agencies pitch aims at the creative. Generate concepts, generate copy, generate whole campaigns. That’s aimed at the wrong end of your P&L. The creative isn’t what’s eating your margin. The forty minutes before the creative starts is.

Raw input	What you ask Claude for	What comes back
Kickoff notes, call transcript, client goals	A structured creative brief	Objective, audience, deliverables, tone, mandatories, and a list of what’s still unanswered
A round of markup, comments, and change requests	A revision summary	One deduplicated list, grouped and prioritized, with contradictions flagged
A client email, Slack thread, or meeting recap	A task list	Discrete tasks with owners, dates, and the ambiguous items pulled out separately

None of this touches creative judgment. It clears the administrative drag around creative judgment, which is the part that burns out account managers and adds three days to every project nobody can account for. Anthropic’s own prompt engineering guidance lands on the same principle these are built on: the more specifically you name the output you want, the more useful the result. It’s the same reason our Claude email prompts and the vertical sets for healthcare marketing and higher-ed marketing teams all open with constraints instead of a topic.

A Brief Dies When Someone Has to Write It From Scratch at 6pm

The thinking usually happened. It happened on the call, in the room, with the client saying the thing they only say once, forty minutes in, after the deck is closed. Then the call runs twelve minutes over, everyone drops, and the brief gets written that evening by someone reconstructing it from six lines of notes and a memory.

That’s where briefs go vague. Not from bad thinking. From the gap between the thinking and the writing down.

Here are my notes from a client kickoff: [paste notes or transcript].
Turn these into a creative brief with these sections: Objective, Target
Audience, Key Message, Deliverables, Tone and Voice, Mandatories and
Constraints, Success Metrics, and Open Questions. For anything the notes
do not cover, list it under Open Questions instead of guessing.

Draft a creative brief for a [social campaign] for [client/industry].
The goal is [goal]. Give me two versions: one concise one-pager and one
detailed version. Flag any assumption you made so I can confirm it with
the client.

Review this draft brief and pressure-test it: [paste brief]. What is
vague, what is missing, and what would a designer or copywriter need to
ask before they could start? Rewrite it to close those gaps.

The “Open Questions” line is the one to keep, and it’s the one everybody deletes because it makes the output look incomplete. It makes the output look incomplete because the input was. A model told to fill gaps will invent a target audience that sounds plausible enough to survive three rounds before somebody notices it was never discussed. A model told to flag gaps hands your account lead a to-do list for the follow-up call.

The Revision Round Is Where the Margin Goes

Nobody loses money on the first round. They lose it on rounds three through six, on changes small enough that asking for a change order feels petty, which is exactly why the fourth one arrives.

Here is a round of client feedback and markup on [deliverable]: [paste all
comments, notes, and change requests]. Summarize it into one deduplicated
list of changes. Group by type (copy, design, layout, other), note any
comments that contradict each other, and flag anything that needs a
clarifying question before we can action it.

Compare the client's requested changes against the current version and
tell me which are quick fixes, which are substantial, and which would push
scope beyond the original brief: [paste feedback] [paste brief]. Draft a
short, friendly note to the client about anything that looks out of scope.

Write a plain-language revision recap for our internal team from this
feedback: [paste]. Lead with what changed and why, so the designer does
not have to read the whole thread.

If you only take one prompt out of this article, take the middle one. Catching scope creep at the revision stage, before the work is done, is the entire difference between a healthy account and one that quietly loses money one small tweak at a time. Most agencies discover scope creep during reconciliation, which is roughly like discovering a leak by reading your water bill.

The contradiction flag earns its place too, because client feedback has a dialect and the model is good at reading it flat.

“Can we make it feel a bit more premium?” “Not loving the direction, but I can’t put my finger on it.” “I shared it with a couple of people and got some thoughts.”

Translation: someone senior has a note, it hasn’t been articulated yet, and the next round will be judged against a standard that doesn’t exist on paper. Claude can’t fix that. It can put those three lines next to each other in one list, which is usually enough for whoever runs the account to pick up the phone instead of guessing at a fourth round.

The Translation Nobody Scoped

This is the one that gives back the most time, because it’s the step that exists in every project and appears in no estimate.

Convert this client feedback into a task list: [paste email/thread/recap].
For each task, give me a clear action, a suggested owner by role (account,
design, copy, production), a rough effort estimate, and a due date working
back from [final deadline]. Put anything ambiguous in a separate "Needs
clarification" list.

Here is a meeting recap: [paste]. Pull out every commitment, decision, and
action item. Format as a table with Task, Owner, Due Date, and Status set
to Not Started. Separate client action items from agency action items.

Take this list of tasks [paste] and draft the client-facing status update:
what is done, what is in progress, what we are waiting on from them.
Keep it short and confident.

That last instruction, separating client action items from agency ones, does more than it looks like. Half of what stalls a project is sitting on the client’s side, and the reason nobody says so is that nobody has a clean list of it. Put those four items in their own section, paste it into the status email, and the conversation about the timeline changes shape.

You still assign the work and set the dates. What you skipped is the reading and the sorting and the rewriting. If you’re running lean enough that this is basically the whole job, AI prompts for a marketing team of one covers the same move at smaller scale.

Your Client’s NDA Doesn’t Care How Good the Prompt Is

Same judgment as any other external tool. Client work sits under agreements, some clients have explicit rules about where their material can go, and a few have language about AI specifically that was added in the last eighteen months and that nobody on the account team has read since signing.

Check the agreement and your own data policy before anything sensitive, unreleased, or confidential goes into a consumer AI tool. Strip what shouldn’t leave your walls. When a client has a hard rule, follow it, including the ones that seem excessive.

Every prompt above works on de-identified inputs. Swap the brand name for [CLIENT], pull the unreleased product name, generalize the numbers. The structure is what you’re asking for, and the structure doesn’t need the confidential parts to come out right.

Use It on Inputs. Never on Decisions.

Claude can tell you two notes contradict each other. It cannot tell you which one the CMO actually cares about, or that this is the third “small copy tweak” this month and the retainer is already underwater, or that the client’s new VP is testing whether you push back.

Three specific holdouts:

The creative call. A structured brief is not a concept. The thinking still happens in the room, with people who know the brand.

Any estimate you’ll commit to. Claude’s number is a starting point for a conversation with production. It is not a figure to put in front of a client, and production will remember if you do.

Anything client-facing that ships unread. A generated status update is a draft. Somebody who knows the relationship reads it before it sends, because tone is a relationship decision and the model has never been in the room.

The line holds across all three. Inputs, not decisions.

A Task List Sitting in an Email Thread Is Just Homework

Watch what happens to a clean list. It gets pasted into an email. It goes to four people. Two of them read it. One of them starts a task. Nobody knows what happened to the other six items until somebody asks in a status meeting eight days later, and the answer is that they’re fine, probably.

The artifact was never the hard part. The handoff is.

Workzone started as an agency more than two decades ago, so briefs, rounds, approvals, and production are the workflow the product was built around rather than features added later. The brief becomes a project. The task list becomes assigned work with owners and dates. On the Team and Enterprise plans, Proofing & markups and Document approvals move revisions to reviewers and clients for sign-off inside the platform, so chasing approvals stops being an account manager’s second job. Because the proof and the project live in the same place, an approved asset moves into the next stage instead of getting downloaded from one tool and re-uploaded into another. Portfolio View gives leads and clients one picture of where things stand, so status stops being a meeting, and Workload & capacity planning shows who’s underwater before they say so, which is the actual job of capacity planning.

The Prompt Was Never the Bottleneck

Claude can hand you a clean brief, a deduplicated revision list, and a well-formed task list in about ninety seconds each. That’s a genuine hour back on a Thursday afternoon.

Then the list has to go somewhere.

If it lands in an inbox, you spent ninety seconds to create one more thing to chase, and by Monday you’re back to the version of this job that has no name. If it lands in a system where the work already lives, the triage you just did actually holds.

The sorting was always the cheap part to fix. The chasing is the expensive one.

See how agencies run briefs, approvals, and production in one place. Explore Workzone for agencies.

Frequently asked questions

What can creative agencies use Claude for? The highest-value uses turn messy inputs into structured outputs: drafting creative briefs from kickoff notes, summarizing rounds of client feedback into a clean revision list, and converting client emails or meeting recaps into task lists with owners and due dates. Claude handles the administrative translation around creative work, which frees account and production teams to spend time on judgment rather than sorting.

What is a good prompt for writing a creative brief? Paste your raw kickoff notes or call transcript and ask Claude to structure them into a brief with fixed sections: Objective, Audience, Key Message, Deliverables, Tone, Mandatories, Success Metrics, and Open Questions. The key instruction is to place anything the notes do not cover under Open Questions instead of guessing, so you get a ready list of what to confirm with the client.

How do you turn client feedback into a task list with AI? Give Claude the full email, thread, or meeting recap and ask for a task list rather than a summary. Request a clear action, a suggested owner by role, an effort estimate, and a due date for each item, working back from the final deadline, plus a separate “Needs clarification” list for anything ambiguous. You then assign and confirm, skipping the manual sorting.

Can AI catch scope creep in client revisions? It can flag it early. Ask Claude to compare requested changes against the original brief and label each as a quick fix, a substantial change, or out of scope, then draft a friendly client note about anything beyond the agreement. Catching scope creep at the revision stage, before the work is done, protects the project margin.

Is it safe to use client work in Claude? Use the same caution as any external tool. Client work is often under an NDA, so check your client agreements and data policy before pasting anything confidential, unreleased, or sensitive, and follow any client rule about AI use. The prompts work well on generalized or de-identified inputs, so you rarely need to share material you would hesitate to send.

What should agencies not use AI for? Anything that requires knowing the client, the history, or the stakes. Claude can tell you that two pieces of feedback contradict each other, but not which one matters more to the client, and its effort estimates are a starting point for a conversation with production rather than a number to commit to. Use Claude on inputs, not on decisions, and have someone who knows the relationship read anything client-facing before it sends.