Safe Input Guide
What’s safe to enter into AI tools — and what never is.
The cheapest piece of AI governance is a written line everyone can point to. This guide gives you three plain-language bands — generally fine, handle with care, never — and space to make the list your own.
Free
- No email
- Works offline
- ~15 minutes
Download my filled copy
Nothing you type here is stored or sent. Download a filled copy before closing the page.
The two questions that settle most cases
Would I put this in an email to a stranger? If not, it doesn’t go into an AI tool you haven’t deliberately checked and approved.
Do I know where this tool sends what I type? If you can’t answer, treat the tool as public until you’ve looked. Many consumer AI plans may use what you enter to improve their models, and settings differ between personal and business accounts.
- Three bands, in plain language
Generally fine
Your own writing, submitted to be improved, with names and identifying details removed.
Questions about publicly known things, regulations you’re trying to understand, how a tool works, an industry norm.
Drafts, brainstorms, and outlines that contain nothing about a specific client or person.
Your own published material, website copy, public posts, marketing you’ve already released.
- Handle with care — only in tools you’ve deliberately approved, with settings checked
Internal processes and how-we-work documents.
Pricing thinking, margins, and strategy that isn’t public.
Unreleased plans, drafts of announcements, anything embargoed.
Anonymized client scenarios — where removing the name genuinely removes the identity. If the industry plus the situation would identify them anyway, treat it as never.
Never enter
Client names tied to their situations, or any detail that identifies a specific client without their agreement.
Passwords, API keys, access codes, security answers — anywhere, ever.
Financial account numbers, card numbers, full statements, payroll files.
Government IDs, and health or personal information about other people.
Anything covered by an NDA or a client’s own confidentiality terms.
Anything you’d be uncomfortable seeing quoted back to someone else.
A useful habit for gray areas: paste the shape, not the substance, “a client in retail with a late-paying customer” carries the problem without carrying the client.
Make it yours
Generic lists get ignored; your list gets followed. Edit the lines below to name what “never” means in your business, then put the finished list where everyone who uses AI tools will actually see it.
Our never-enter list
- Never entered into AI tools, in our business
- Why
- + Add a line
- Tools we’ve approved for handle-with-care material
- Tool
- Approved for
- Checked by
- Settings checked on
- + Add a tool
Educational operating guidance only. Not legal advice, cybersecurity certification, or regulatory compliance certification.
This page runs entirely in your browser and sends nothing anywhere. It summarizes, in its own words, ideas from public frameworks, the NIST AI Risk Management Framework and Playbook, the NIST Generative AI Profile, and CISA’s small-business resources. The originals are linked in Official Sources, part of this same free set.
- Free Sunnyworks guide