We built two free AI prompts that every agency should have
If you use AI tools in your agency work, and you've been building up client knowledge inside ChatGPT, there's a good chance you've hit this problem.
You switch to Claude for a new project, or start a fresh thread, and suddenly all that accumulated context is gone. The tone decisions, the avatar work, the positioning you spent hours refining — scattered across old conversations, buried in project files, or just lost.
So you start again. You explain the client from scratch. You rebuild the brief. You re-establish the voice. And a few weeks later, you do it all over again with the next project, or the next tool.
It's one of the most frustrating inefficiencies in agency AI workflows right now, and it's almost entirely avoidable.
We've built two free prompts that solve it. One runs inside ChatGPT. One runs inside Claude. Together, they produce three finished onboarding documents that capture everything an AI, or a new team member, needs to write for a client from day one.
This isn't about switching tools. It's about not having to rebuild your thinking every time you do.
The problem with how most agencies use AI for client work
Most agencies have fallen into a pattern that looks something like this.
A new client comes in. You start a ChatGPT thread. You upload some files, paste in some website copy, ask a few questions, refine a few answers. Over a few weeks, that thread, or that project, builds up a real understanding of the client: their voice, their audience, their quirks, the things they care about that aren't written down anywhere.
That accumulated understanding is genuinely valuable. It's the difference between copy that sounds like the client and copy that sounds like every other business in their sector.
The problem is that it lives entirely inside one tool, in one set of threads, in a format that doesn't transfer cleanly anywhere. If you switch AI tools, start a new project, bring in a new team member, or simply need to brief a different platform on the same client, you're starting from scratch.
And even within a single tool, that knowledge tends to degrade. Old threads get archived. Context gets lost. A conversation that happened three months ago might as well not exist.
The fix isn't to use one AI tool forever and never look at another one. The fix is to extract the thinking into a format that doesn't belong to any tool at all.
The solution: three foundation documents per client
We work with a simple principle at Licorne: every client should have three foundation documents that govern all the copy written for them. Together, they answer the question a new staff member, or a new AI tool, would ask on day one.
"What does this business do, who do they serve, how do they sound, and what are the rules I need to follow when I write for them?"
The three documents are:
| Document | What it covers |
|---|---|
| Tone of Voice & Positioning Brief | How the business sounds, what it stands for, how it differs from competitors, and what it should never sound like. The document that stops generic copy before it starts. |
| Ideal Client Avatars | Who actually buys from this business, what they care about, what language they use, and what they're trying to solve. Written in the client's own words, not marketing-speak. |
| Writing Pack | The practical rulebook: structure, terminology, CTA language, words to use and avoid, section patterns, sector-specific rules. A new writer should close this and know exactly what to do. |
These aren't internal notes or working drafts. They're finished onboarding materials. The test we use for every section: could a new hire read this and write confidently for the client without needing to ask a follow-up question? If not, it's not done yet.
The two prompts
We've built one prompt for each platform. They produce the same three documents, but they're designed for different starting points.
Prompt 1: The Foundation Documents Prompt (for Claude)
This one is for new client projects, or for clients where the source material already exists in a Claude project folder.
You place the prompt in your client folder alongside whatever source material you have, such as discovery notes, existing website copy, competitor research, brand guidelines, reviews, interview transcripts and ask Claude to follow the Foundation Documents Prompt for this client.
Claude works through three phases in order:
Audit phase: it reads everything in the folder and produces a short audit covering what the business is, who they serve, what's genuinely distinctive, what tone the existing material uses, and where the obvious weaknesses are. This step alone often surfaces things the brief didn't make explicit.
Gaps brief: it lists what's missing or thin, framed as specific questions grouped by which document they affect. A good gaps brief has 8–15 genuinely useful questions. You answer them, partly answer them, or tell it to proceed with assumptions flagged.
Document production: it produces all three documents as finished .docx files, written for a new staff member. If a section genuinely can't be substantiated from the source material, it omits it rather than filling it with placeholder content.
The audit-then-gaps-then-produce order is deliberate. Foundation documents grounded in real specifics are the difference between copy that sounds distinctive and copy that sounds like every other website in the sector.
Prompt 2: The ChatGPT Extraction Prompt (for migrating to Claude)
This one is for when you've been working with a client inside ChatGPT and want to pull all that accumulated thinking into a form you can actually use elsewhere.
ChatGPT projects don't hold source material in a tidy folder. The useful context is scattered across project-level custom instructions, uploaded files, and the conversation history of previous threads. A prompt that just says "audit the folder" won't catch most of it.
This prompt tells ChatGPT to do a full sweep of everything it has access to before producing anything — including the memory from past conversations, not just the files. It then runs through the same three phases as the Claude version, producing the same three finished documents.
Once the documents are produced, you save them and upload them to your new Claude project. The new hire — whether that's a human or an AI — has everything they need to write for the client from day one.
No lost context. No starting from scratch. Just a clean handoff.
How to use them
The workflow is straightforward. Here's how we recommend approaching it:
If you're setting up a new client in Claude:
Create a new Claude project for the client
Upload your source material, such as website copy, discovery notes, brand guidelines, reviews, anything you have
Place the Foundation Documents Prompt in the project folder
Start a new conversation and ask Claude to "follow the Foundation Documents Prompt for this client"
Work through the three phases
Save the three output documents to the project folder
If you're migrating an existing client from ChatGPT:
Open the ChatGPT project you want to migrate from
Start a new conversation inside that project (so it has access to files, instructions, and memory)
Paste the ChatGPT Extraction Prompt into the conversation
Work through the three phases
Save each of the three documents to your machine
Upload all three to the new Claude project
Add a brief handoff note so Claude knows the context of the transfer
The handoff note we use: "Here is the knowledge transfer document for [Client Name]. This has been extracted from my ChatGPT project for this client and covers everything built up over our working history together. Please ingest this as your full working context for this client. I'll be continuing all work for them here going forward."
What you end up with
Done properly, these three documents should make a new team member, or a new AI tool, fully capable of writing for a client without any follow-up from you.
In practice, that means:
Brand-consistent copy from day one, whether you're writing it, a colleague is, or an AI is
No rebuilding context every time you start a new thread or project
A clear, transferable record of the thinking you've invested in each client
A starting point for every new piece of work that doesn't depend on whoever originally built the relationship
Onboarding material that's genuinely useful if you ever bring in support, a freelancer, or a new team member
The documents also tend to surface things you didn't realise you'd built up. Tone decisions made months ago that were never written down. Avatar nuances that exist only in the memory of whoever ran the discovery session. Positioning language that evolved organically across six conversations and never got captured anywhere.
Getting that out of your AI tool's memory and into a document you own is, in our view, one of the most valuable things an agency can do with an hour of AI time.
A note on using AI tools well
We work across both Claude and ChatGPT at Licorne, and we don't think of them as competitors. They have different strengths, different contexts where they perform well, and different ways of holding, or losing, accumulated knowledge.
What these prompts reflect is a broader principle we try to apply to all our AI work: the output is only as good as the context you give it. An AI working from a vague brief produces vague copy. An AI working from three specific, substantiated foundation documents produces copy that sounds like the client.
The prompts are free because we think this approach is worth sharing. If it makes AI-assisted client work more consistent and less time-consuming for agencies beyond our own, that's a good outcome.
If you'd like to talk through how we approach AI Optimisation more broadly, including how it affects website structure, content strategy, and how AI tools recommend and rank businesses. That's a conversation we're always happy to have.
About Licorne Agency
Licorne is a Squarespace web design and digital strategy agency based in Deal, Kent. We build websites, manage SEO, and help SMEs get found online, including by the AI tools that are increasingly shaping how businesses are discovered.
"Design without strategy is just decoration."
Get the prompts
Both prompts are free to download. No sign-up needed.
Foundation Documents Prompt
(Claude version) - for new client projects and Claude-based workflows
Foundation Documents Prompt
(ChatGPT extraction version) - for migrating existing client work from ChatGPT to Claude
Questions about how to adapt these for your own workflow? Get in touch, we're always happy to take a look.