48 Free Claude Prompts for Agencies
A free pack of 48 Claude prompts for agency work, from client onboarding to invoices, and how to make any prompt work for your clients.
48 Free Claude Prompts for Agencies
Most prompt lists for agencies read the same: "Act as an expert copywriter. Write a blog post about [topic]."
Then the output reads the same too. Generic, a little too enthusiastic, and nothing like the client.
The problem isn't Claude. It's that the prompt gave it nothing to work with. We put together 48 prompts that do the opposite, and they're free.
Why most prompts produce generic work
Anthropic's own prompting guidance puts it plainly: think of Claude as a brilliant but new employee who lacks context on your norms and workflows. The more precisely you explain what you want, the better the result.
It also suggests a test worth stealing: show your prompt to a colleague with minimal context on the task and ask them to follow it. If they'd be confused, Claude will be too.
So what? A prompt for agency work needs three things a one-liner doesn't have: a clear task, a structure for the output, and a marked place for the client's context: voice, audience, offer, constraints. Leave out the third and Claude fills the gap with something plausible and wrong.
A good prompt is a brief. It's short on adjectives and long on the facts only you know.
What's in the pack
The Agency Operations Pack covers the work between winning a client and sending the invoice, in seven parts:
| Part | Prompts | Examples |
|---|---|---|
| Client onboarding & discovery | 7 | Brand voice extractor, client context document |
| Strategy & research | 7 | Content gap analysis, audience persona builder |
| Content production | 7 | Client-voice writer, email sequence architect |
| Project management | 7 | Scope of work drafter, status report generator |
| Client communication & reporting | 7 | Monthly performance report, QBR narrative |
| Financial operations | 6 | Proposal builder, pricing strategy analyser |
| Business development & growth | 7 | Cold outreach personaliser, case study pipeline |
Every prompt sets a task, gives Claude a structure to follow and marks each place your client's context belongs with a {{PLACEHOLDER}}. Each one comes with a short tip on using the result.
You get an online library you can search and copy from, a PDF, an editable HTML version and the source files. Enter your name and email for immediate access; there's no Vythos account or payment, and the newsletter box is optional and unticked.
How to use a prompt well
Start with one task you do every week
A monthly report or a status update, not a one-off. You'll see quickly whether the prompt saves time, and you'll refine it on real work.
Replace every placeholder
The placeholders are the point. {{CLIENT_AUDIENCE}} filled in with two specific sentences does more for the output than any amount of "act as a world-class strategist".
Give it the source material
Paste the brief, last month's numbers, the client's tone notes. Claude can only use what it can see.
Review before it goes out
Treat the output as a strong first draft from someone new to the account. Check every figure and every claim against what the client approved.
Save what worked
When you've tuned a prompt for a client, keep that version. Next month should start from it, not from the original.
Where it gets harder
Prompts get you from a blank page to a draft. The friction that's left is the context: pasting the same tone notes, rate card and banned phrases into every chat, for every client, and hoping everyone on the team pastes the current version.
That's the part Vythos handles. It keeps each client's facts, brand rules and your approved decisions in one place, and gives them to Claude, ChatGPT or Cursor when a task needs them, so the placeholders fill themselves from what you've already agreed.
The honest trade-off
A prompt library won't make weak strategy strong, and it won't know your client. If your team already has well-worn prompts for these tasks, you may only want a handful of these. And any AI draft for a client needs a person to read it before it's sent.
The one thing
The quality of the output follows the quality of the context. Pick one prompt, fill in every placeholder properly, and compare the result with what a one-line prompt gave you last week.