Products SEO & GEO Agent Paid Ads Agent Website Agent Conversion Agent
How it works Pricing FAQ Blog Book a call

AI assistant data: what to share and what to keep private

Updated · 7 min read

Short answer Give an AI assistant only what the task in front of it needs: numbers and page text, not names. Connect accounts at the lowest role that does the job instead of sharing a password. Never paste passwords, card or bank numbers, ID numbers, customer health details or a contact list into a chat app.

Want help applying this to your business? Talk it through on a free 30-minute call.

Book a call

What data does an AI assistant actually need?

Less than most owners think. Most marketing tasks run on two kinds of input: your own public material (web pages, service list, prices you already publish) and totals (clicks, calls, cost per lead). Neither names a person. Names, phone numbers and account numbers almost never change the answer, so leave them out.

Match each task to its smallest slice:

TaskWhat the AI needsWhat to leave out
Write or fix a service pageThe current page text, your services, area served, real pricesCustomer names, job addresses
Reply to a reviewThe review text and what actually happened, in general termsThe customer's phone, invoice, health or legal details
Find wasted ad spendA search terms report and campaign totalsYour login, billing and card details
Explain a traffic dropMonthly report exports from Analytics or Search ConsoleRevenue by named client
Draft a follow-up emailThe offer, your tone, one made-up sample customerYour real contact list
Plan next month's postsYour calendar of events, seasons and promotionsUnannounced deals, staff matters

The test before you paste: would the answer be worse without this detail? If not, cut it.

What should you never share with an AI chat app?

Some data is risky in any chat window, whatever the settings:

  • Passwords and sign-in codes. No AI task needs them. Connect the tool instead (next section).
  • Card numbers, bank and routing numbers.
  • Government ID numbers: Social Security numbers in the US, Social Insurance Numbers in Canada, driver's licence or passport numbers.
  • Customer health details, if you are a clinic, dentist or other health provider (see below).
  • Full customer or patient lists with names, emails or phone numbers.
  • Employee records: pay, reviews, discipline, medical notes.
  • Anything under a confidentiality agreement, plus contracts and deals not yet public.

Google says it plainly about its own chat app:

"Please don't enter confidential information that you wouldn't want a reviewer to see" (Gemini Apps Privacy Hub)

The reason: "Human reviewers (including trained reviewers from our service providers) review some of the data we collect," and "Reviewed chats are retained for up to three years," even after you delete your history (Gemini Apps Privacy Hub).

Does ChatGPT use what you type to train its models?

On personal plans, yes by default. On business plans, no by default. The setting you're on matters more than the brand:

Product and planUsed for training by default?Source
ChatGPT Free, Plus, ProYes: "data sharing is enabled by default." Turn off "Improve the model for everyone" in Settings, Data controlsOpenAI Help Center: Data controls in ChatGPT
ChatGPT temporary chatNo, but it "may be retained for up to 30 days for safety purposes"Same
ChatGPT Business, Enterprise, Edu, APINo, not by defaultSame
Claude Free, Pro, MaxOnly if you allow it, plus chats flagged for safety reviewAnthropic Privacy Center (consumer)
Claude for Work, Anthropic APINo: "By default, we will not use your inputs or outputs from our commercial products"Anthropic Privacy Center (commercial)

One catch: on personal ChatGPT plans, rating an answer with thumbs up or down lets OpenAI use that whole conversation for training, even after you opt out (OpenAI Help Center). Anthropic says the same about feedback. Turning training off also doesn't delete anything; delete chats separately. For which plan fits a small team, see AI tools for small business.

Training is only one risk. A setting doesn't stop a staff member from pasting a client list, and it doesn't stop human review or a hacked account. Keeping data out is the only control you fully own.

How do you give an AI tool access to your accounts safely?

Connect it through the platform's own access screen, at the lowest role that does the job. Never hand over your password. Google's ad, analytics and search tools each have roles built for this:

AccountLowest role for reading and reportingRole to let it make changesKeep for yourself
Google AdsRead-only: "Can view campaigns and use planning tools"Standard: "Can edit campaigns"Admin and Billing
Google AnalyticsViewer, plus the "No Cost Metrics" and "No Revenue Metrics" restrictions if it doesn't need themMarketer (audiences and events)Administrator
Search ConsoleRestricted user: "Has simple view rights on most data"Full user: "Has view rights to all data and can take some actions"Owner

Role wording is from Google Ads Help, Google Analytics Help and Search Console Help.

Admin can add and remove users. Billing, and also Google Ads Standard, can edit billing information. An AI tool rarely needs either. Start with read-only, and raise one level only when the tool has to make a change you've agreed to. The bigger question of when a tool should act on its own is in AI agent tool vs chatbot.

Turn on two-step verification for the AI account itself: anyone who gets in can read every saved chat.

Check what's connected every few months. In your Google Account, open the linked apps page, choose Access to your Google Account, pick the app, open its details, then Remove access. Google: "If you remove access, the app can't access your Google Account" (Google Account Help).

How do you strip personal details before you paste?

Five steps, under a minute:

  1. Export totals, not rows. Monthly counts by service or city say as much as a row per customer.
  2. Delete name, email, phone and address columns before you upload a spreadsheet.
  3. Swap names for roles: "Customer A, a homeowner in the north end."
  4. Round anything that could identify someone: "about $2,000 job," not the invoice.
  5. Read the paste once before you press enter. Signatures and email footers hide phone numbers.

Canada's federal privacy regulator tells organizations using generative AI to, where possible and reasonable, "use anonymized or de-identified information within prompts" (Office of the Privacy Commissioner of Canada, December 2023). It also says prompts should not be "retained, used for secondary purposes, or disclosed" unless required.

Stricter rules apply. In the US, a clinic or dental office covered by HIPAA (the federal health privacy law) that lets an outside online service handle patient data needs a signed contract first:

"the covered entity (or business associate) and the CSP must enter into a HIPAA-compliant business associate agreement (BAA)" (HHS: Guidance on HIPAA and Cloud Computing)

A CSP is a cloud service provider, which includes AI software that stores what you type. Without that agreement, keep patient details out entirely. The same risk shows up in website tracking: see Meta pixel and Google Analytics on a clinic website.

Law firms, accountants and financial advisers: check your professional body's confidentiality rules before any client file goes into an AI tool. Until then, no names and no documents.

Example

A plumbing company tidies its ad account

The owner wants AI to find wasted Google Ads spend of about $1,500 a month. He exports a 90-day search terms report and gives the tool Read-only access. It flags 40 irrelevant searches; he adds them as negatives himself. No login, no billing access shared.

A dental office replies to reviews

The front desk pastes a 2-star review into a chat app and writes "patient waited 40 minutes; we were short one hygienist that day." No patient name, no treatment details. The draft apologizes for the wait and invites the reviewer to call the office.

A bookkeeper plans content

She gives ChatGPT her service list and published prices, plus "most calls come from restaurants in March." Her client list stays in her accounting software. She uses a Business plan, so training is off by default, and keeps client files out anyway.

FAQ

Is it safe to upload a spreadsheet to ChatGPT?

It's safe once the columns that identify people are gone. Delete names, emails, phone numbers and addresses first, or export totals instead of rows.

Does turning off training make ChatGPT private?

No. It stops new chats being used to train models, but chats stay in your history, and feedback you give can still be used. Delete chats you don't want kept.

Should I give an AI tool my Google password?

No. Grant access through Google's own screens (a role in Google Ads, Analytics or Search Console), which you can remove at any time without changing your password.

Can staff use their personal ChatGPT accounts for work?

They can, but the business doesn't control those accounts, and personal ChatGPT plans train on chats by default. Write down what staff may paste and what they may not.

Different for you?

Keep learning

NextHow to vet AI tools: a safety checklist before your team uses oneEight safety and compliance checks to run on any AI tool before staff use it, with the Grok image-edit case showing what happens when a vendor skips them.

Ask us one question

One line is enough. We reply by email.

Sources

  1. OpenAI Help Center: Data controls in ChatGPT
  2. Gemini Apps Privacy Hub
  3. Anthropic Privacy Center: Is my data used for model training? (consumer)
  4. Anthropic Privacy Center: Is my data used for model training? (commercial)
  5. Google Ads Help: About access levels in your Google Ads account
  6. Google Analytics Help: Access and data-restriction management
  7. Search Console Help: Manage owners, users, and permissions
  8. Google Account Help: Manage connections between your Google Account and third parties
  9. HHS: Guidance on HIPAA and Cloud Computing
  10. Office of the Privacy Commissioner of Canada: Principles for responsible, trustworthy and privacy-protective generative AI technologies (December 7, 2023)