Give each client its own container in every AI tool that touches their footage. That container should also be the unit you search in and the unit you delete. In practice this means one workspace or project per client in anything you upload to. It also means never putting two clients in the same chat or the same index, and knowing, for each tool, where an uploaded file ends up and how to remove it. Most of the work is setting things up per client before the first upload. The rest is choosing tools whose structure makes that habit easy to keep.
Why AI tools make this harder than folders did
With folders and drives, separation came from location. A client's footage lived in that client's folder, and access followed the folder. AI tools add copies. A clip dropped into a chat is one copy. A transcription job makes a transcript. A search tool builds an index. Every copy is a place where client footage exists outside your folder structure, and an index that holds two clients' footage will return both.
So for every tool, ask four things before anyone uploads. Where does the file go? What gets derived from it (transcripts, summaries, an index)? What does one search cover? What happens when you delete it?
The routes, and what each one costs
Folders, shared drives and naming rules
Per-client top-level folders, a shared drive per client, and access limited to the people on that account. This is cheap, and most agencies already do it. Its limit is that AI tools don't inherit it. Once someone drags a file from the client folder into a chat window, the folder rule no longer applies to that copy.
AI chats with file uploads
Uploading a clip to an AI chat to get a summary or a rough log is the most common way footage leaves the folder. If you do it, keep to one conversation or one project space per client and never mix clients in a thread. Read the vendor's data terms for what happens to uploaded files and how long they are kept, because these often differ by plan. The practical limit is that uploads work per conversation. Long recordings may not fit, and you start over the next time.
Editing and transcription tools with workspaces
Many of these tools have team workspaces. Make one per client if the tool allows it. If it doesn't, use one project per client with consistent names. Remember that transcripts and captions are derived copies. Deleting the source video doesn't always delete them, so check how the tool handles that before you rely on it.
A search layer with one project per client
The fourth route is to upload a client's footage to a search tool, index it once, and search it by description. Here the separation comes from the project boundary: if every search runs inside a single project, a search on one client's work only runs over that client's footage. In Vivu, each search stays inside one project, so an agency that keeps one private project per client gets the boundary from how the account is set up, not from people remembering which folder they're in. When an engagement ends, deleting that client's project removes its media from the cloud, which gives you one step to take at offboarding.
The cost is that footage has to be uploaded first, searches draw on an allowance, and you have one more place to account for at offboarding. It also means comparing work across clients takes one search per project. That is the boundary doing its job. The setup is covered step by step in setting up one library per client, and choosing search software as an agency covers what else to check.
Reference and third-party footage
Separation isn't only between clients. Agencies collect reference ads, competitor spots and UGC that nobody on the account owns. Before uploading any of it to an AI tool, confirm you have the right to use it that way. Keep references in their own project instead of mixing them into a client's material, so a search for the client's shots doesn't turn up someone else's ad.
When you don't need any of this
If you have one or two clients, footage never leaves the editing machine, and your AI use is limited to scripts and emails, folder discipline is enough. The same goes if a client contract bars third-party tools. Then the answer is that their footage doesn't go into AI tools at all, and no setup changes that.
How to tell which side you're on
Count the places each client's footage currently lives: folders, chats, transcription tools, indexes. If you can name every one and say how to delete from each, your separation holds, and a new tool only needs its own per-client container. If you can't, sort that out before adding another AI tool.
FAQ
Is it safe to upload client footage to an AI chat?
It depends on the vendor's data terms for your plan and on your contract with the client. Check both before uploading. Keep each client in their own conversation or project space so one thread never holds two clients' material.
If the contract is silent on AI tools, ask the client. That conversation is easier before an upload than after one.
Can I search all my clients' footage at once and still keep it separate?
Not in tools that separate clients by project, because each search runs inside one project. Searching across clients means running the same search in each client's project in turn.
That extra step is what keeps one client's results from showing up in another client's search.
What should I delete from AI tools when a client engagement ends?
Delete every copy, including derived ones: uploads in chats, transcripts and captions in editing tools, and projects in search tools. Deleting the original file in one tool doesn't always remove what was derived from it in another.
Keeping a simple list of which tools hold each client's footage makes offboarding a checklist.
Do I need a client's permission to put their footage into AI tools?
Often the contract answers this, and some contracts restrict third-party processing. If yours doesn't address it, ask the client and write down the answer. For footage you don't own at all, such as another brand's ads, confirm your usage rights before uploading.