Key takeaways
- Whether a public AI tool keeps or learns from what you type depends on the plan and the settings. Read the provider's terms for your plan.
- "Private" should mean your documents stay in an environment reserved for you, and are not used to train anything.
- For general writing and brainstorming a public tool is usually fine. For contracts, customer data and internal documents, think twice.
The short answer
If you only need help phrasing an email or generating ideas, a public AI assistant is a reasonable choice. If you want an AI to read your contracts, quotes and internal documents, the question changes from "is it clever?" to "where does my information go?" That is where a private setup earns its place.
What happens to your documents
When you paste text or upload a file into an online AI tool, it travels to the provider's servers so the model can read it. What happens next is set by the provider's terms and by the plan you are on: some keep conversations for a period, some allow staff to review them, and some offer settings that stop your data being used to improve their models.
None of this is sinister, but it is easy to miss. The practical rule is to check three things before you share business documents: how long it is kept, who can see it and whether it is used for training.
What "private" should mean
The word is used loosely, so here is a useful test. A private AI for company documents should give you:
- A reserved environment: the model that answers your questions is not shared with other customers.
- No training on your files: your documents are used to answer you, not to improve anyone's model.
- A private connection: your offices reach it over an encrypted link, not the open internet.
- Respect for permissions: people only get results from files they could already open.
When a public tool is fine
Not everything needs a private setup. Public tools work well for drafting and rewriting text that contains nothing sensitive, brainstorming, summarising public information, and learning a new topic. A good habit is to keep a short list of what staff may and may not paste into public tools.
A simple rule of thumbIf you would not be comfortable emailing it to an outside consultant without an agreement, do not paste it into a public AI tool.
What it takes to run it privately
At a high level, a private setup needs three things: dedicated hardware to run the model, a private connection from your offices, and a search layer that understands your folders and your permissions. Running it yourself means looking after all three. Having a provider look after it means you simply ask questions and get answers from your own documents.
We are testing exactly this in the Labs, and we write up the results as we go.
Common questions
Is a private AI as good as a public one?
For answering questions about your own documents, usually yes, and often better, because it is looking at your files. For open-ended creative tasks, the largest public models can still be stronger.
Does a private AI need a lot of setup on our side?
Very little. It needs a connection to the places where your documents live, and a decision about who may see what. The rest is done for you.
Can staff still use public tools?
Yes, for the right tasks. A private AI covers the work that involves your documents, and a short guideline covers everything else.