The Concern Column
Before a model weighs our options, has it heard what we want?
Objectives first, then what's attractive and what's concerning about each option, so the decision rests on everything that matters.
What do we want, and what worries us about each option?
For whoever has a decision to make. You bring a decision you're weighing·15min·any model·card
It ends in: heard, or not yet
Watch the two-minute walkthrough 2:05
A two-minute walkthrough: why pros and cons hand a model half a decision, the fields, the heard or not yet verdict, and how to run the prompt in any model.
Made with the open-source LEJ AI Video Guides kit from Limited Edition Jonathan
The decision it forces
Before a model helps us weigh our options, has it heard what we want, and what about each option concerns us?
Most of us skip both. We paste the situation into a chat window, ask for the pros and cons, and get back a tidy, balanced list in seconds. It looks rigorous, and the model is genuinely good at this part. The trouble is what we handed it: the rational half of the information, and none of the rest.
To run it, copy the prompt below into any model you already use.
The argument
Good decisions aren't born of rational thinking alone. What concerns us about an option is information too, often the most important information in the room, and it's the one kind a model can't look up. As Sabino Marquez put it on Episode 4, at 14:05 in the transcript, "safety is a feeling in the body." Only we can report it.
So the move isn't to stop using AI to decide. It's to change what we give it. Have it ask what we want before anything else. Let it help with what's known about each option, where it's genuinely useful. Then have it ask, option by option, what's attractive and what's concerning, and weigh all of it, concerns included, against the objectives we named.
The structure is Michelle Florendo's Pro-Con List With A Twist: instead of pros and cons, a column for each option, asking what's attractive about it and what's concerning, then looking for themes. She teaches the full method in her Decision Toolkit for Coaches and Counselors, and she's been answering questions like these for six seasons on her podcast, Ask a Decision Engineer.
The fields
| Field | What it holds | Who fills it |
|---|---|---|
decision |
The decision we're weighing, in one sentence | you |
objectives |
What we actually want from this decision, in our words, including any our concerns turn up | you |
options |
The choices on the table. The AI may suggest up to two, labelled as its own | you |
information |
What's known about each option, and what isn't. The AI drafts it and we correct it | either |
attractive |
What draws us to each option | you |
concerning |
What concerns us about each option, and where we feel it: head, heart, body, or more than one. The AI asks, and never guesses | you |
themes |
Patterns across our answers, grouped by the AI, in our words | machine |
leaning |
Which option, and the objectives and concerns it rests on. The AI drafts it only when we ask | machine |
verdict |
heard, or not yet | derived |
There's no worked example on this page: a filled card is one person's real decision, with their own concerns in it.
The verdict rule
Take the first rule that applies.
- Any option with no
concerninganswer from us, not even "I don't know": not yet. Go back and ask. - A
leaningthat doesn't name the objectives and concerns it rests on: not yet. - Otherwise: heard. The weighing rests on everything we said, and we can see what any suggestion stands on.
"Heard" isn't "decided". It means the model has the whole of the information, so its weighing is worth reading.
Run it with any model
It interviews, one question at a time. It may draft what's known about each option and suggest up to two more. It never supplies a concern.
You are going to help me make a decision with a Concern Column card. Do not give me a pros-and-cons list. Interview me first, one question at a time, and wait for my answer each time. Before the first question, tell me in one line that my answers may get personal, and that I can switch to a temporary chat now if this one is saved. 1. Ask me to state the decision in one sentence. 2. Ask what I actually want from it: my objectives, in my own words. If I name an option instead of an objective, ask what it would get me. 3. Ask me to list the options. You may suggest at most two more, labelled as your suggestions. 4. For each option, draft what's known about it and what isn't. Mark it as your draft and ask me to correct it. 5. For each option, ask me what's attractive about it. Use my words, not yours. 6. For each option, ask me what's concerning about it, and where I feel it: head, heart, body, or more than one. 7. Then ask whether any of my concerns points to something I want that I haven't named. If it does, add it to my objectives in my words. RULES YOU DO NOT BREAK 1. Never invent a concern for me, finish one I've started, or soften one I've named. If I say I don't know, record "I don't know" and move on. 2. Never drop a concern from the weighing because it doesn't fit. 3. Suggest a leaning only if I ask for one, and name the objectives and concerns it rests on. 4. Derive the verdict, first match wins: any option without a concerning answer from me, not yet. A leaning that doesn't name what it rests on, not yet. Otherwise heard. OUTPUT My objectives; a table of the options with my attractive and concerning answers; the themes you see, quoted in my words; how each option serves my objectives with my concerns counted; a leaning only if I asked, with what it rests on and what would change it; the verdict. Then valid JSON matching the schema concern-column/v0.3 as published at https://sigsub.show/takeaways/concern-column/#the-answer-format. If you can't read that page, use these keys, in this order: schema, decision; objectives, a list; options, a list, each with name, suggested_by, information, information_gaps, attractive, concerning and felt_in (a list); themes, a list; leaning, with option, rests_on_objectives (a list), rests_on_concerns (a list) and what_would_change_it; verdict.
What to check in the answer
It may draft the information and suggest options. It may never supply a concern: if one appears that we didn't say, or one we said is missing from the weighing, the verdict is not yet, whatever the JSON says.
The part that's not optional
Objectives first, and back to them after the concerns, since a concern often points at something we want and hadn't said. If we can't say what we want, no weighing of options, however good, will tell us which one gets us there.
For builders
The answer format
{
"schema": "concern-column/v0.3",
"decision": "the decision, in one sentence",
"objectives": [],
"options": [
{
"name": null,
"suggested_by": "user | ai",
"information": "drafted by the AI, corrected by the user",
"information_gaps": null,
"attractive": null,
"concerning": null,
"felt_in": ["head | heart | body"]
}
],
"themes": [],
"leaning": {
"option": null,
"rests_on_objectives": [],
"rests_on_concerns": [],
"what_would_change_it": null
},
"verdict": "heard | not yet"
}
concerning is null until the user answers, and "I don't know" is an answer. leaning.option stays null unless the user asked for one.
Got an answer back from the prompt? Read it against this schema. It runs in your browser and sends nothing.
Report back
Ran this? Tell us what happened. What broke and what you changed are the useful parts, because that's how the next version gets better.
Or write to hey@sigsub.show with Takeaway: concern-column as the subject. It's a plain email: no form, no account, nothing tracked. Nothing you send appears here unless you say yes, and you choose how you're named.
Rather talk it through in public? Comment on the Sunday issue on Substack, the same issue on LinkedIn or the episode on YouTube. Those threads live on their platforms, under their rules; nothing from them runs on this page.
Keep the next choice in view
One signal, one subtraction and one analogy every Sunday. Five minutes, no video required.
Optional. Free on Substack, no account needed. Already on LinkedIn? Subscribe there.