Agency LinkedIn AI Workflow
A per-client folder and Project structure, a weekly routine, and the isolation rules that keep the work clean. Running six clients through one chat window is how client A’s proof point ends up in client B’s post. This prompt designs the structure that prevents it: one Project per client, shared files that are genuinely shared, client files that never leave, and a routine that fits the seat and send limits you are actually working inside.
You are designing an operating structure for an agency running LinkedIn work for several clients with AI assistance.
## The agency
Clients: {{CLIENT}}
Services delivered: {{PRODUCT}}
Team size and roles: {{CONSTRAINTS}}
Objective: {{OBJECTIVE}}
## What has to be true
- No client's material may ever appear in another client's output. This is the requirement everything else is designed around.
- A new person must be able to pick up a client and produce acceptable work the same week.
- Approvals must be traceable: who approved what, when.
- The LinkedIn side has real ceilings — roughly 100 invitations per week per account and a fixed monthly InMail allowance per seat — and they apply per account, not per client. A plan that ignores them is fiction.
## Task
1. Design the folder structure: what is shared across all clients, what belongs to one client, and where the boundary is. Show it as a tree.
2. Specify the Project setup: one Project per client, what is attached to each, and what lives in the agency-wide Project instead.
3. Write the isolation rules — the specific practices that prevent cross-contamination, including what to do when a useful idea comes from the wrong client.
4. Design the weekly routine per client: what happens on which day, who does it, and what the client sees.
5. Define the approval trail: the artefact, the recipient, the deadline and the default if nobody responds.
6. Compute the per-client cost in seats, credits and hours, and state the point at which adding another client requires another seat or another person.
## Output format
### Folder structure
A tree, with one line per file on what it contains.
### Project setup
| Project | Attached files | Who uses it | What it must never contain |
|---|---|---|---|
### Isolation rules
Numbered, specific, each one actionable rather than aspirational.
### Weekly routine
| Day | Activity | Owner | Client-visible output |
|---|---|---|---|
### Approval trail
The artefact, the route, the deadline, the default.
### Per-client cost
Seats, InMail credits, hours. Then: the client count at which this structure breaks and what to add.- You run LinkedIn for more than two clients and the context keeps bleeding.
- You are standardising how your team uses AI rather than leaving it to each person.
- You are pricing an engagement and need to know what a client actually costs in seats and hours.
- Run it once for the agency, not once per client. The output is your operating manual.
- Implement the isolation rules first. They are the part that protects the client relationships.
- Use the per-client cost section when pricing. Agencies usually under-price the seat and credit cost of outbound by a wide margin.
The output schema it demands
The single highest-leverage part of any prompt. Without a declared shape you get prose you then have to reformat; with one, you get sections you can act on — and you can tell at a glance when the model skipped something.
### Folder structure (tree) ### Project setup (table) ### Isolation rules ### Weekly routine (table) ### Approval trail ### Per-client cost (+ the breaking point)
What it looks like filled in
Illustrative, not a promise — this is the shape of a good answer rather than a guarantee of one. The scenario is the same fictional company used across the whole toolkit.
Agency running LinkedIn outbound for 4 clients, one shared sending seat per client, a 2-person team.
Isolation rules — 1) a client's file is never attached to another client's Project. 2) an idea from Client A is rewritten from scratch for Client B, never copy-pasted across.
Per-client cost — one seat gives roughly 100 invitations a week per client, so 4 clients need 4 separate sending seats, not one shared one — a fact the original engagement pricing had missed.
The breaking point — at 6 clients, one person can no longer personalise touch 2 across all of them by hand; that is where the team needs a third person, not better tooling.
Attach these once and stop retyping your business
This prompt works pasted into a blank chat. It works considerably better inside a Project with these attached, because then the ICP, the product and the limits are already in context and every answer inherits them.
ICP.md
Every prompt you run afterwards knows who you sell to, in terms LinkedIn can filter on.
Free downloadBRAND-VOICE.md
Drafts arrive in your voice on the first attempt instead of after four rounds of "less corporate".
Free downloadOUTREACH.md
Every message an assistant writes comes out at the right length, in the right shape, with the openers you hate already banned.
Free downloadLINKEDIN-LIMITS.md
Your assistant stops proposing 300 connection requests a week, because it now knows the ceiling is about 100.
Free downloadThe reference behind this prompt
Per-seat pricing, dated, for the cost section of this plan.
The constraint that most agency capacity plans get wrong: six clients on one sending account share one weekly invitation ceiling.
Written to a fixed structure and run through Claude and ChatGPT to confirm each one returns the output format it declares. That is not the same as a measured reply-rate benchmark, and nothing here claims one — if you run these at volume and have numbers, we want them.
Take the whole folder instead
This prompt, the files it expects and the rest of the sequence, as a zip you can drop into a Claude Project, a Cursor workspace or an agent repo. Free, no account.