Twelve rules for using AI in your business.

The twelve rules.

1. Every number shows where it came from.

   A number without a source is a guess. Every number the AI gives you comes with where it came from: the account, the dates and any filters, so anyone on your team can check it.

   In practice: a weekly enquiries figure arrives with the Google Analytics account, the dates and the filter used.

2. The AI uses your definitions.

   Every business counts leads, customers and sales differently. Write your definitions down once, and the AI uses them every time instead of guessing.

   In practice: “A lead is a form with a work email. Newsletter sign-ups don't count.”

3. Compare like with like.

   Compare a week with the same days a year earlier, counted the same way. A Tuesday-to-Monday week set against a calendar week shows a trend that isn't there.

   In practice: “Compare each week with the same weekdays last year, and say which comparison you used.”

4. Every fact links to its original source.

   Statistics link to the organisation that collected them, with the date. If the AI can't find the original, the fact comes out.

   In practice: a survey figure links to the research firm's own report, with its date and how many people it asked.

5. A person approves everything before it goes out.

   The AI drafts, and a named person approves. That covers articles, press releases, ads, emails and website changes, and your rulebook says who approves what.

   In practice: “Press releases need the owner's approval. Blog posts need the editor's.”

6. The AI can look, but not change.

   Connections let the AI read your data. Changes arrive as drafts or preview links that a person accepts. Nothing changes a live ad account, CRM or website on its own.

   In practice: a headline change arrives as a preview link. It goes live when someone approves it, and it can be undone in one step.

7. Your tone of voice is written down.

   Write down how you sound, somewhere the AI reads it: your tone, product names, the words you never use, and examples of writing you like and don't like.

   In practice: a list of banned words that the AI checks every draft against before anyone sees it.

8. No fake logos, customers, quotes or reviews.

   The AI never draws a logo, invents a customer, writes a review or puts words in a real person's mouth, and AI-made people are never presented as your staff or customers. Logos come from your brand files, and quotes come from the people who said them.

   In practice: image requests leave space for the logo, which is added from your real files afterwards.

9. Everything is in your accounts, on your card.

   The AI accounts, connections, data and code belong to the business that uses them. If the people who set it up move on, nothing is lost.

   In practice: every subscription is billed to you, and every login uses a company email address.

10. Measure before and after.

    Choose what you want to change before you start: hours on a task, money spent on software, pieces published a month, or enquiries with a known source. Measure it at the start and again 30 days later.

    In practice: the hours spent on the Monday report, written down in week one so nobody has to remember them later.

11. Someone on your team is in charge of it.

    AI setups go stale when nobody looks after them. Name one person who keeps the rulebook up to date, adds assistants and fixes what breaks, and give them time to do it.

    In practice: that person's name is at the top of the rulebook, with a monthly review in their calendar.

12. If it can't be checked, it doesn't go out.

    Everything gets checked: numbers against a report you trust, facts against their sources, web pages against automatic tests, images against your brand rules. Anything that fails goes back with the reason.

    In practice: a draft with a statistic that has no source goes back to the AI with that sentence marked.