Which AI tools are safe for client data?
Your team pastes client emails, quotes and notes into AI tools every day. Whether that information is used to train someone's AI depends less on the tool than on the plan and the account they sign in with.
This page is for small business owners in Surrey and across Metro Vancouver who want the plain answer, plan by plan, with the vendor's own page as the source for every row.
Checked September 28, 2026 on each vendor's own pages
For transparency: the AI inside our own products runs on Anthropic's business API, which is in this table. Every service we use is listed on our AI trust page. See our AI trust page.
Does it train on your data? Plan by plan
Each row links to the vendor page it comes from. The last column is our recommendation for client work, not the vendor's.
| Tool and plan | Trains on your data by default? | Can you turn it off, and where | What we recommend for client work | Source and date checked |
|---|---|---|---|---|
| ChatGPT and the OpenAI API | ||||
| ChatGPT Free, Go, Plus and Pro | Yes OpenAI says it may use your content to train its models. | Settings, then Data controls, then Improve the model for everyone: off. Or ask in the OpenAI Privacy Portal: Do not train on my content. Either covers new chats only. Temporary chats are not used for training. | Keep client data out If staff use it, switch training off and use Temporary chat. |
Checked September 28, 2026 |
| ChatGPT Business (called ChatGPT Team until August 29, 2025) | No Not by default, per OpenAI. | Nothing to switch off. Only what you choose to share, such as feedback, can be used. | Suitable for client work Staff sign in to the business workspace. Admins can view, export and delete members' chats. | Checked September 28, 2026 |
| ChatGPT Enterprise and Edu | No Not by default, per OpenAI. | Nothing to switch off. | Suitable for client work OpenAI says its staff open chats only to resolve incidents, to recover chats with your permission, or where the law requires. | Checked September 28, 2026 |
| OpenAI API (inside software a developer builds) | No Not since March 1, 2023, unless you opt in. | Nothing to switch off. Check that no organization owner has turned on data sharing. | Suitable for client work Inputs and outputs may be kept up to 30 days to watch for abuse. | Checked September 28, 2026 |
| Microsoft Copilot | ||||
| Work account (Copilot Chat and paid Microsoft Copilot, signed in with a Microsoft Entra work account) | No Microsoft says prompts, responses and work data are not used to train foundation models. | Nothing to switch off. Staff must sign in with the work account. | Suitable for client work Microsoft says queries from outside the EU may be processed in the US, the EU or other regions. |
Checked September 28, 2026 |
| Copilot in Microsoft 365 Personal or Family (Word, Excel and Outlook at home) | No Microsoft lists these users among those it does not train on. | Nothing to switch off. | Only with care It does not train, but it is a home subscription. Put client work on a work account. | Checked September 28, 2026 |
| Copilot app on a personal account, new version (the web, and apps updated on or after August 18, 2026) | No Microsoft says prompts, responses and files are not used to train foundation models. | No training setting to change. To keep chats from shaping Bing, MSN and ads, switch off One shared experience and Allow ads personalization under Settings, then Personalization. | Keep client data out Microsoft's pages for the new app do not say whether people review chats or how long chats are kept. |
Checked September 28, 2026 |
| Copilot app on a personal account, old version (not updated) | Yes Canada is not on Microsoft's list of excluded countries. | Profile, then Privacy, then Training on conversation activity and Training on voice conversations: both off. Applies to future chats. | Keep client data out Microsoft says trained reviewers may read chats, and chats are kept 18 months by default. |
Checked September 28, 2026 |
| Google Gemini | ||||
| Gemini app on a personal Google account (18 and over) | Yes Keep Activity is on by default, and Google uses that activity to train its AI models. | gemini.google.com, then Settings & help, then Activity, then Turn off. Or use a temporary chat, which is not used for training. | Keep client data out Some chats are read by human reviewers and kept up to three years, even after you delete your activity. |
Checked September 28, 2026 |
| Gemini in Google Workspace (business editions, including the Gemini app signed in with a Workspace account) | No Google says Workspace does not use customer data to train models without permission. | Nothing to switch off. Staff must sign in with the Workspace account, not a personal Gmail. | Suitable for client work Google says your content is not reviewed by people outside your organization without permission. Admins set how long chats are kept. | Checked September 28, 2026 |
| Claude and the Anthropic API | ||||
| Claude Free, Pro and Max | You choose Anthropic asks, but its own image of the 2025 pop-up shows the setting already switched on. | Settings, then Privacy, then Help Improve our AI models: off. Incognito chats are never used for training. | Keep client data out Switching it off does not take chats out of training runs already started. |
Checked September 28, 2026 |
| Claude for Work (Team and Enterprise) | No Not by default, per Anthropic. | Nothing to switch off. An owner can also turn off Rate chats under Organization settings, then Data and Privacy, so feedback is not shared. | Suitable for client work Staff sign in to the organization account. | Checked September 28, 2026 |
| Anthropic API (inside software a developer builds) | No Not by default, per Anthropic. | Nothing to switch off. | Suitable for client work Anthropic deletes inputs and outputs within 30 days, with a few exceptions such as a legal requirement. |
Checked September 28, 2026 |
| Perplexity | ||||
| Perplexity Free, Pro and Max | Yes AI data retention is on by default. | Account settings, then Preferences, then AI data retention: off. Covers data from that day on only. | Keep client data out Perplexity says data collected before you switch it off cannot be deleted. | Checked September 28, 2026 |
| Perplexity Enterprise Pro and Enterprise Max | No Perplexity says Enterprise data is never used for AI training. | Nothing to switch off. | Suitable for client work Uploaded files are kept 7 days. Moving back to a non Enterprise plan switches training back on. | Checked September 28, 2026 |
| Meta AI | ||||
| Meta AI in Facebook, Instagram, WhatsApp, Messenger and meta.ai | Yes Meta says it uses your AI chats to improve AI at Meta. | We found no setting on Meta's own pages that lets a Canadian user stop it. | Keep client data out Meta says it reviews AI chats, by automated tools or by people. Private messages stay out of training unless someone in the chat shares them with Meta AI. |
Checked September 28, 2026 |
Every row was checked on September 28, 2026 against the vendor's own privacy, help and terms pages, linked in the table. Vendors change these settings and rename plans, so open the source before you rely on a row. This is general information, not legal advice.
What to do this week
- 01List the tools and the accountsAsk each person which AI tools they use for work, and whether they sign in with a personal or a business account.
- 02Move client work to a business planChatGPT Business, Microsoft Copilot on a work account, Gemini in Google Workspace, Claude for Work or Perplexity Enterprise. None of them trains on your data by default.
- 03Switch training off everywhere elseOn any personal plan that stays, use the setting shown in the table, and use temporary or incognito chats.
- 04Skip the thumbsTell staff not to rate chats that contain client information. On ChatGPT, Gemini and Claude personal plans, a thumbs up or down can send the conversation back for training, even with training off.
- 05Write it downPut the approved tools and the rules in an AI use policy your team signs.Get the free AI use policy template
Training off is not the same as private
- Google keeps Gemini chats that human reviewers looked at for up to three years, even after you delete your activity.
- Perplexity's setting covers only data from the day you switch it off, and Claude's does not reach training runs already started.
- On ChatGPT, Gemini and Claude personal plans, feedback can send the conversation back for training even with training switched off.
- Microsoft says work queries from outside the EU may be processed in the US, the EU or other regions. If client information can leave Canada, your privacy policy should say so.
What the vendor pages did not say
We list these so nobody repeats them as fact.
- Microsoft's new personal Copilot app: whether people review chats, and how long chats are kept.
- Claude personal plans: the signup default, in words. Only Anthropic's image of the pop-up shows it switched on.
- Perplexity personal plans: whether people review chats, and how long data is kept.
- Meta AI: any setting that lets a Canadian user stop training on their chats.
- Whether Microsoft's work account promise differs between the free Copilot Chat and the paid licence.
Not sure where AI fits in your business yet?
Take the two-minute AI readiness score. It shows your result straight away, with no email asked for.
Take the AI Readiness CheckFAQ's
About AI tools and client data
Which AI tools are safe for client data?
No tool is safe on its own: the plan and the sign-in decide it. As of September 28, 2026, every business plan we checked does not train on your data by default: ChatGPT Business and Enterprise, Microsoft Copilot on a work account, Gemini in Google Workspace, Claude for Work, Perplexity Enterprise, and the OpenAI and Anthropic APIs. Staff still need to sign in with the business account, and a person should check what AI writes before a client sees it.
Does ChatGPT train on my business data?
On the Free, Go, Plus and Pro plans, yes, by default. Switch it off under Settings, Data controls, Improve the model for everyone, which covers new chats only. ChatGPT Business, called ChatGPT Team until August 29, 2025, and ChatGPT Enterprise do not train on your data by default, according to OpenAI's enterprise privacy page.
Does Microsoft Copilot use my data for training?
Not on a work account: Microsoft says prompts, responses and work data are not used to train foundation models. On a personal account it depends on the app. Microsoft says the new Copilot app, available since August 18, 2026, does not train foundation models on your chats, while the older app trains unless you switch off Training on conversation activity and Training on voice conversations.
If I switch training off, is my data private?
Not completely. Google keeps chats that human reviewers looked at for up to three years, Perplexity's setting covers only data from that day forward, and on ChatGPT, Gemini and Claude personal plans a thumbs up or down can send the conversation back with your feedback. Keep client information on business plans.
How current is this table?
Every row was checked on September 28, 2026 against the vendor's own pages, which are linked in the table. Vendors change these settings and rename plans, so open the source before you rely on a row. It is general information, not legal advice.