Does ChatGPT train its models on what I type into it?

It depends on the tier. On the free and Plus consumer tiers, OpenAI has historically used conversations to improve its models unless you turn that setting off or opt out. On the Team, Enterprise, and API tiers, business content is not used for training by default. The exact terms change over time, so the safe assumption for a consumer login is that a human or a training process could see what you paste unless you have explicitly configured otherwise.

Is a paid ChatGPT plan enough to keep my business data private?

A paid business plan (Team or Enterprise) removes training on your content and adds administrative controls, which is a real improvement over a personal account. It does not make the data local: your documents still travel to and are processed on the provider's servers, and staff can still paste something into the wrong window. It reduces the risk; it does not remove the category of risk.

What counts as sensitive enough to keep out of a public AI tool?

Customer lists and contact details, pricing formulas and margins, vendor agreements and rates, employee information, anything covered by a contract's confidentiality clause, and anything you would not want a competitor or a customer to read. For most service businesses that is the majority of what makes the business valuable, which is why "just be careful what you paste" is not a workable policy.

How is Xandite different from using ChatGPT with my documents?

Xandite is a private workspace per business. Your documents are stored in your workspace, and in production the answering model runs on private infrastructure — your content is not sent to a public AI API. Staff ask questions against your material and get answers with citations back to the source document. It is an internal tool for the owner and crew, not a chatbot for your customers.

Guide

Is it safe to put your business data into ChatGPT?

What public AI services do with the data you give them, which paid tiers actually protect business content, and what private should really mean.

Published August 31, 2026

Short answer: it depends entirely on which version you are using, how it is configured, and what you paste in. A personal ChatGPT login is not built to be a safe home for your pricing formulas, your customer list, or your vendor agreements. A paid business tier is better, but "better" is not the same as "private," and it is worth knowing exactly where the line is before your team starts pasting job notes into a chat window.

This guide walks through what public AI services do with the data you give them, where the real risks are for a small service business, and what a genuinely private setup looks like.

What happens to the text you type into a public AI tool

When you send a message to a hosted AI service, three things happen to it that matter for a business.

It travels to the provider's servers. The model does not run on your laptop. Your message, and any document you attach, is transmitted to and processed on infrastructure the provider controls. That is true of every hosted service, on every tier.

It may be retained. Providers keep conversation history so you can go back to it, and they keep logs for abuse monitoring and debugging. Retention windows vary by tier and by setting. On some plans, deleted conversations are purged within days; on others, copies persist longer for policy or legal reasons.

It may be used to improve the model. This is the part that changes most between tiers, and the part people get wrong. On consumer tiers, providers have historically used your conversations as training material unless you opt out or switch off a data-controls setting. On business and API tiers, your content is excluded from training by default — OpenAI states this explicitly for ChatGPT Business, Enterprise, and the API.

None of this means the providers are careless. It means a hosted consumer product is designed around a consumer's expectations, not a business's confidentiality obligations.

Consumer tier versus business tier

The gap between a personal account and a business plan is real, and if your team is going to use one of these tools anyway, the business plan is the one to use.

Consumer (free / Plus): Assume that what you type could be used to train future models unless you have explicitly turned that off, and assume there is no contract governing how the provider handles your data beyond the standard terms of service. There is no admin oversight — each person's account is their own. This tier is fine for drafting an email or rephrasing a paragraph. It is not fine for anything a competitor or a customer should never see.

Business (ChatGPT Business / Enterprise / API): Your content is excluded from training by default. You typically get a data processing agreement, administrative controls, single sign-on, and audit logging. This is a legitimate improvement and, for many companies, an acceptable answer for general-purpose AI use.

What the business tier still does not do:

It does not move the processing onto your own infrastructure. Your documents still leave your control to be answered.
It does not stop a staff member from pasting the wrong thing into a personal account on their phone.
It does not ground answers in your specific documents — you are still copy-pasting context into every conversation, and the model still answers from its general training the moment you forget to.

What "sensitive" actually means for a service business

Owners often assume they have nothing confidential worth worrying about. Then you list it out:

Pricing logic — your flat-rate book, your markup formulas, the modifiers for stairs, access, disposal, and after-hours, the judgment calls on when to hold a quote and when to walk.
Customer information — names, addresses, phone numbers, payment terms, the history of who is difficult and who pays late.
Vendor and supplier terms — who you call first for which part, your account pricing, your rebate and warranty arrangements.
Employee information — pay rates, performance notes, who is certified for what.
Anything under a confidentiality clause — commercial accounts, property-manager contracts, and fleet agreements frequently include one.

For most owner-operated trades businesses, that list is the business. "Just be careful what you paste" is not a workable data policy when the useful answer requires pasting exactly the material you should not paste.

What private should mean

A genuinely private setup for business knowledge has a few properties worth insisting on:

Your documents stay in a workspace you control. Not a shared chat product — a space that is yours, isolated from every other customer.
Real content is not sent to a public AI API. In production the answering model runs on private infrastructure, so your material is not handed to a third-party model provider to generate the response.
Answers are grounded and cited. The tool answers from your uploaded documents and shows you which one it used, so staff are not getting a confident guess from general training data.
It is an internal tool, not a customer channel. The audience is the owner and the crew looking something up — not your customers, and not a lead-capture bot.
There is an audit trail. You can see what was asked and what changed.

Where Xandite fits

Xandite is built to be that private option. Each business gets an isolated workspace. You upload the documents that hold your pricing logic, your vendor map, and your procedures once. Your staff ask questions in plain language and get answers grounded in those documents, with a citation back to the source. In production, real client data is not sent to a public LLM API — the isolation and data-handling rules are enforced in the code, not left to a settings page.

It is deliberately an internal tool for how your business runs — the same knowledge you would otherwise be answering by phone from a roof or a truck. It is not a chatbot for your customers and not a missed-call or booking widget.

If you want to see how this maps to your trade, the trade-specific pages go into detail: junk removal, HVAC, auto repair, and plumbing. To talk it through, email the founder directly.

Related guides

Private ChatGPT alternatives for small business — the full range of options, with honest tradeoffs.
How to stop being the bottleneck in your service business — the operational problem this privacy question sits inside.
How to document the way you price jobs — what to write down first so a private AI has something to answer from.

Frequently asked questions

Does ChatGPT train its models on what I type into it?: It depends on the tier. On the free and Plus consumer tiers, OpenAI has historically used conversations to improve its models unless you turn that setting off or opt out. On the Team, Enterprise, and API tiers, business content is not used for training by default. The exact terms change over time, so the safe assumption for a consumer login is that a human or a training process could see what you paste unless you have explicitly configured otherwise.
Is a paid ChatGPT plan enough to keep my business data private?: A paid business plan (Team or Enterprise) removes training on your content and adds administrative controls, which is a real improvement over a personal account. It does not make the data local: your documents still travel to and are processed on the provider's servers, and staff can still paste something into the wrong window. It reduces the risk; it does not remove the category of risk.
What counts as sensitive enough to keep out of a public AI tool?: Customer lists and contact details, pricing formulas and margins, vendor agreements and rates, employee information, anything covered by a contract's confidentiality clause, and anything you would not want a competitor or a customer to read. For most service businesses that is the majority of what makes the business valuable, which is why "just be careful what you paste" is not a workable policy.
How is Xandite different from using ChatGPT with my documents?: Xandite is a private workspace per business. Your documents are stored in your workspace, and in production the answering model runs on private infrastructure — your content is not sent to a public AI API. Staff ask questions against your material and get answers with citations back to the source document. It is an internal tool for the owner and crew, not a chatbot for your customers.