How agencies keep client brands separate across AI tools
Client brands get mixed in two ways: an agent applies the wrong client’s guidance, or one client’s material turns up in chats and projects where it doesn’t belong. Here’s how agencies guard against both in the AI tools they already use.
TL;DR
Name the intended client brand in every request and check the work against its guidance. Keep client files and conversations in separate projects, with memory and sharing restricted to that account team. Use plans whose terms fit the client contract. Brand selection and data access need separate checks.
Two ways client brands mix
Check both which brand the work uses and who can access the client's material.
- The wrong brand in the work
Copy uses another client's voice, or an image follows another client's palette or references.
- Client material in the wrong place
Files reach another client's chat, people outside the account team or a tool the client has not agreed to use.
Where the crossovers happen
Check these places where one client's context can reach another client's work:
- Memory
Details from one client's chats enter another task through account memory.
- Shared references
A mixed inspiration folder gives the agent another client's style reference.
- Reused deliverables
A reused post or template carries another client's phrases or claims.
- Defaults
A vague request leaves an agent to choose among connected brands.
- Copies
Guides pasted into several tools need separate updates after a brand change.
Set up each AI tool per client
Projects hold a client's files, instructions and chats. Check these differences in the vendors' documentation, as of October 2026:
Claude
- Memory: Projects do not use each other's memory. Turn off “Use account memory” to stop exchange with non-project chats; Team/Enterprise owners control availability.
- Sharing: Team/Enterprise: invited members or the whole organisation, with view/edit permissions.
- Training by default: Not on commercial products, such as Claude for Work. Claude's thumbs-up/down feedback can store the whole conversation for training; Team/Enterprise owners can disable that button.
ChatGPT
- Memory: Project-only memory blocks exchange with outside chats; shared projects use it automatically. Default memory can cross that boundary on plans other than Enterprise/Edu, including Business. The setting can be changed later.
- Sharing: Invite people or groups, or share by link. “Anyone with a link” admits any signed-in user; workspace links on Business/Enterprise/Edu admit anyone in that workspace.
- Training by default: Not on Business, Enterprise or Edu.
If you switch an existing private ChatGPT project to project-only memory, let the change take effect before adding restricted client material. OpenAI says this can take a few hours.
Check the client contract before selecting tools: it may require the client's account or exclude particular material from AI tools.
For each client, in every tool you use:
Per client, per tool
- One project per client, plus separate ones for confidential work such as an unannounced launch, named so nobody picks the wrong one.
- Memory kept inside the project.
- Shared by invitation with the account team, not by link or with the whole organisation.
- A business plan, or the client’s own account if the contract asks for it.
- One person who removes access when someone leaves the account.
For other agents, check the same four things: file location, memory, sharing and data use. With model APIs, separate credentials too; a Claude Console workspace can scope a key to one client's resources.
Keep one brand per client
Separate projects still need current guidance. Twelve clients across three tools can mean 36 copies to update. The cross-tool update guide covers retiring them; use retrieval instructions instead:
Instead of Upload the brand guide to every project.
Maintain each client’s brand in one place.
Instead of Put everything in the project.
Keep the stable brand apart from campaign details. Put the campaign brief in the project.
Instead of Keep a shared folder of inspiration.
Keep approved references and assets per client, so an agent working on one client’s task is only given that client’s files.
Instead of The agent will work out which client this is.
Name the client’s brand in every request. Project instructions say which client the project is for and where its brand comes from, instead of holding a copy of the guide.
Rainbrand maintains each client's guidance, identity and reusable assets. With agents connected through MCP:
- List accessible teams and brands, then select the client's brand.
- Request a Brief for the task.
- Make the work and request a Review against that brand.
- For lasting changes, inspect and apply a brand edit, then request fresh context. Copied settings remain unchanged.
A new brand goes to the personal account unless you name a team. Team members share brands and credits; the docs do not describe per-brand permissions within a team. Check the client's access restrictions before placing its material there. The same selection controls apply when embedding Rainbrand in client services.
Keep the selected brand visible in the job
Name the client and brand in the task record. Inspect the returned logo, product references and voice rules; resolve mismatches before creation. Client-facing software should resolve the authorised brand on its server, since access to several brands does not identify the right one.
Follow the Claude or ChatGPT setup guide, then try:
Confirm the connected Rainbrand account and the selected brand for [client]. If the task or attached files name a different client, ask which one the work is for before retrieving guidance. Request a Brief for the attached deliverable, using its audience, format and exact copy. Show the guidance and selected assets before making the work. Use only the intended brand's files, then review the finished composition against that brand.Check the client before the brand
Before sending the work, check the selected client and references:
Right client
- Made in that client’s project, from that client’s brand.
- Every reference image and template came from that client.
- No other client’s names, products, claims or file names, including in metadata and alt text.
Right brand
- Checked against the client’s current guidance, not an old copy.
- Images reviewed against that client’s brand, with findings fixed or explained.
- Anything uncertain goes to the account lead.
Getting the client’s own sign-off is a separate step; How agencies get client approval on AI-made brand work covers it. For the wider rules on keeping any AI output on brand, see How to keep AI-generated content on brand.
When a client leaves
- Remove people from the client’s projects, then archive or delete the projects as the contract requires. In Claude, archiving a project keeps its members and their access, and deleting one also deletes its memory. In ChatGPT, deleting a project removes its chats, instructions and the files stored only in that project; files saved separately in the Library stay.
- Revoke the API keys you created for that client and archive its API workspace. In the Claude Console, archiving a workspace also archives its keys and can’t be undone.
- If the client wants its brand context, export it as a Brand Skill folder through the API first. Then delete the brand in Rainbrand if the contract calls for it. Deleting a brand can’t be undone.
Where to start
Try two clients: separate their projects and memory, select their Rainbrand brands explicitly, and run one real task for each. Check the output and access before expanding. Use the roster guide for ongoing work and the onboarding checklist for each new client.
More in Agencies
5 min read
How agencies get client approval on AI-made brand work
Agree the ground rules before the first draft, check the work against the client’s brand before they see it, tell them how it was made and what they’re deciding, and put each correction into their brand once.
5 min read
How agencies manage many client brands with AI agents
Give each client job an explicit brand, current guidance and a handoff the next person can use. Keep brand changes separate from campaign feedback without replacing your delivery tools.
4 min read
How to onboard a client brand into AI: an agency checklist
Collect the client's current sources, resolve conflicting guidance, create reusable brand context and check the first task. Keep the files, findings and unanswered questions in a handoff your team can use.
