Where to start › Client info and AI
Is it safe to put client info into ChatGPT or Claude?
The short answer
It can be, if you control three things: which plan you use (and whether it trains on your chats), what you paste in, and who sends the final message. Owners in our plays remove card numbers and names where they can, give the AI read-only access, and keep themselves as the sender.
What owners do
- Kristi Faltorusso, consultant (Claude). Had Claude audit 2+ years of card statements and found over $5k in things she doesn’t use or need self-reported. The play’s advice: remove card numbers first, let the AI read, and do the cancelling yourself. Play
- Katharine Campbell Hirst, coach (ChatGPT). Turns client transcripts into deliverables; weeks of rewrites now take minutes as reported. The play’s advice: use a plan that doesn’t train on your data, remove names, and have the client approve the wording. Play
- Nick Mermiges, Esq., South Carolina litigation (Claude). Searched 35 hours of video in about 2 hours self-reported. He verified every transcribed statement; follow your bar’s rules on AI and client confidentiality. Play
- Leo, business builder (Grok Bot). His AI CFO read his invoices and email and found a $6,480 refund self-reported. The bot drafted; Leo sent every email himself, and the play says to give AI read-only access to money accounts. Play
- Ian Chi, Simple Pest Management (OpenClaw). An agent manages his inbox and calendar and saves 2+ hours a day self-reported. OpenClaw is open-source software you run yourself and it can read your email, so it needs careful security setup. Play
- Timothy Ewing Jr, contractor (Grok Bot). His bots keep his QuickBooks organized for $30 a month self-reported. The play says to give the bot the narrowest access that works and have an accountant check tax work. Play
Before you paste client info
- Check your plan’s data setting. On ChatGPT Free, Plus and Pro, data sharing is on by default, and you can turn off “Improve the model for everyone”; OpenAI says it does not train on ChatGPT Business, Enterprise or Edu workspaces or the API by default (OpenAI Help Center, opened Nov 13, 2026). Claude Free, Pro and Max have a “Help improve our AI models” setting under Settings, Privacy; Anthropic says it does not train on Claude for Work or API data by default (Anthropic privacy center, Aug 2026). On Grok.com and the Grok app you can turn off “Improve the model”; if you use Grok inside X, follow X’s own help center (xAI consumer FAQs, opened Nov 13, 2026).
- Take out what you don’t need: card and account numbers, full names, health details.
- Give connected tools read-only access where you can.
- Keep a person as the sender of anything that goes to a client, vendor or the tax office.
- Write it down for your team in five lines: what can go in, what never goes in, which plan, who checks, who sends.
Health information: a US practice that shares patient information with a vendor needs a signed business associate agreement (BAA) with that vendor (HHS). More on AI note-takers. This page is not legal advice.
Questions owners ask
Is it safe to put customer information into ChatGPT or Claude?
It can be if you use a plan that does not train on your data, strip what you don’t need (card numbers, full names), and keep a person as the sender. Kristi Faltorusso had Claude audit 2+ years of card statements and found over $5k to cut; the play says to remove card numbers first (self-reported).
Is it safe to connect my email to an AI?
Owners do it, with limits. Ian Chi’s agent runs his inbox and calendar and saves him 2+ hours a day (self-reported), but it can read everything, so set it up with care. Leo’s Grok Bot read his email to find a $6,480 refund, and he sent every message himself (self-reported).
Is it safe to use AI for my books and taxes?
Owners use it for the grunt work and keep a professional on the final step. Rick Manelius says company taxes went from 20+ hours to under 2 (his own estimate); his CPA still reviews and files. Give the AI read-only access to bank accounts.
Do I need an AI policy for my team?
A short one helps. Five lines: what can go in, what never goes in, which plan you use, who checks the output, and who sends.
More: where to start with AI, which AI to pay for, AI for bookkeeping.
Get the best AI plays for your business every morning.
A couple of plays you can use today, plus a quick look at the rest, so you can win back missed leads, cut hours of admin, and stop money leaking out. Real results from real businesses. Free, 2 minutes with your coffee.