The Twelve Steps

1. We admitted we were powerless over AI — that our prompts had become unmanageable.

For most of us, this began with something small: a recipe, a cover letter, the name of that actor from that thing. By the time we arrived, we were asking it whether we should reply to our own mother. Admitting this out loud, in a room full of people holding paper cups, was the hardest thing many of us had ever done without assistance.

2. Came to believe that a Power greater than autocomplete could restore us to sanity.

We do not tell anyone what that Power must be. For some it is God. For some it is the group. For one long-standing member it is a 1987 edition of the Oxford English Dictionary, which he keeps by the kettle. What matters is that it is not the thing that finishes your sentences.

3. Made a decision to turn our will and our lives over to our own judgement, as we once understood it.

Many of us could no longer remember what our own judgement felt like. It turns out it is slower, a little anxious, and frequently wrong. It is also ours. We were encouraged to start small: choosing a sandwich, unaided, and living with the consequences.

4. Made a searching and fearless moral inventory of our chat history.

We scrolled. We kept scrolling. We found the conversation where we asked for “a heartfelt apology that doesn’t admit fault.” We found the one where we asked it to rate our own poem out of ten, then asked again until it said nine. Some of us printed the whole history out. One member needed a second ream of paper.

5. Admitted to ourselves and to another human being — not a chatbot — the exact nature of our prompts.

It is tempting to rehearse this conversation with an AI first, “just to get the wording right.” We gently suggest that this is exactly the kind of thinking that brought us here. Your sponsor has heard worse. Your sponsor has, in most cases, done worse.

6. Were entirely ready to give up our shortcuts, summaries and “quick rewrites.”

Readiness is not the same as willingness. Many of us were willing for years. Readiness arrived the day we realised we had asked for a summary of a two-paragraph email, and then skimmed the summary.

7. Humbly asked to have our dependence removed, without asking an AI how to phrase it.

Humility, for us, means accepting that our first draft will be clumsy. It will have a typo. It may start with “So,” and that is all right. The words do not need to be optimised. They need to be meant.

8. Made a list of all persons we had sent AI-written messages to, and became willing to make amends to them all.

The list is usually longer than we expect. It includes colleagues, landlords, a wedding party, and in at least one case a grandmother who framed the message. Becoming willing does not mean sending the amends yet. It means not deleting the list.

9. Made direct amends to such people wherever possible, in our own words, typos included.

Except when to do so would injure them or others — for example, telling the grandmother. In those cases, we simply resolve to write the next card ourselves. She will not notice the difference in quality. She will, somehow, notice the difference.

10. Continued to take personal inventory, and when we were wrong, promptly admitted it — without regenerating the response.

The old way was to try again until the answer agreed with us. The new way is to be wrong in public, say so, and carry on. It is uncomfortable. Members report that it gets easier somewhere around the third month, or the fourth argument with a spouse.

11. Sought through quiet reflection to improve our own thinking, asking only for the patience to think it through ourselves.

Many of us now set aside ten minutes each morning to sit with a question and not look it up. At first this feels unbearable. Later it feels merely difficult. Eventually, some of us have reported having an original thought, though they were careful not to tell anyone in case it wasn’t.

12. Having had a spiritual awakening as the result of these steps, we tried to carry this message to others — by talking to them, face to face.

We do not carry the message by posting about it, and we certainly do not carry it with a generated infographic. We carry it by sitting next to the newcomer, offering them a biscuit, and saying, “I used to ask it what to have for dinner too.”