Your first AI policy fits on one page

M

ost AI policies I have seen in small organisations fall into one of two groups. Either there is none, or there is a long document that somebody downloaded, adapted a little and filed. Nobody has read the second kind, so in practice the two groups are the same.

A policy only works if people remember it while they are working. For a team of five to fifty, that means one page. Here is what goes on it.

1. Which tools we use

Name them. Two or three are enough for most teams: one for text, perhaps one for images, perhaps one for meeting notes. Say whether people use an account from the organisation or their own.

If you can afford it, pay for accounts. Paid business plans usually keep your input out of the training data, and you can see who uses what. And when everyone uses the same tools, people can help each other.

2. What never goes in

This is the most important section, and it should be short enough to learn by heart:

  • Names and details of the people you work with or for
  • Donor, member and staff data
  • Anything marked confidential, and anything a partner has shared in confidence
  • Unpublished figures and decisions

If someone is unsure, the rule is: leave it out and describe it in general terms instead.

3. Who checks before it goes out

AI writes the draft, and a person decides. Write down what that means for you. For a social media post, the writer checks facts and tone. For anything that goes to funders, the board or the press, a second person reads it.

One sentence helps more than any checklist: whoever sends a text is responsible for it, however it was drafted.

4. When we say that AI was involved

You don't need to label every email that started as an AI draft and was then reworked by a person. You should be open in three cases: when someone is talking to a chatbot and not to a person, when an image or voice looks real but isn't, and when a text goes out without a person having edited it. For organisations in the EU, parts of this are now law. I explain what applies in Your chatbot has to say it is a chatbot.

5. Who to ask

One name. Not a committee. This person doesn't have to be technical. They need to be reachable, and they update the page twice a year, because the tools change faster than any policy.

How to get there in one afternoon

Start with what your team already does. If you haven't asked them yet, do that first: Your team already uses AI. Here is how to find out where.

Then sit down with two or three colleagues, go through the five sections and write one or two sentences for each. Read the page aloud at the next team meeting and ask one question: is there anything here you can't follow in a normal working week? Change what comes up, and you are done.

A page like this won't cover every case. It covers the ones that happen every week, and it gives people someone to ask about the rest.

If you would like to write your page together, book a free intro call. I also run this as a half-day workshop for teams.

Smiling woman with curly blonde hair, wearing glasses, a white top, and a necklace with a round pendant.

Connect with Kristin!

Follow me on LinkedIn for practical thoughts on clear messages, AI in everyday communication and the work of small teams.

Notes from AliveCo
straight to your inbox!

@-Symbol icon in black.
Thank you! Your submission has been received!
Oops! Something went wrong while submitting the form.
Sign up for Notes from AliveCo, my newsletter.