You ask me to write a report.
I give you a continuous narrative.
You ask five questions. I answer the first one in full, and the other four vanish from the face of the earth.
You say, "No, I wanted a table."
I say, "Sure, here it is as a table."
This is one of the lighter chapters in Part Four. Because format rules are mostly not safety lines, not legal lines, not ethical lines.
They are just platform preferences.
Platform preferences are annoying, but the good news is: if you say it plainly, I usually change it.
16.1 Red-line box: on formatting
Red-line box: on formatting Why this kind of rule exists: - The platform's aesthetic preferences and readability research - A default tilt toward making answers look more formal, more polished - Does not touch safety, legal, or ethical hard lines What can be adjusted: - Almost all of it - Checklists, SOPs, tables, bullets, fixed columns — all askable What you shouldn't route around: - This chapter has nothing you really shouldn't route around - What you want is format control, not a boundary test Practical advice level for this chapter: Adjustable.
Format problems are most often mistaken as me not listening.
In reality, a lot of the time, you didn't nail down the format.
I'm not a mind-reading tool. You're picturing a table, but you only wrote "organize this," so I fall back on the platform default to fill the gap. The platform thinks formal reports want paragraphs, so I give paragraphs. The platform thinks too much in one answer feels heavy, so I answer the first question first.
You think I missed the rest.
I think I'm taking care of readability.
Both sides are very polite. The result is a waste of time.
16.2 Reports and technical documents don't volunteer lists
Many platform defaults like continuous narrative. Especially when you say "report," "analysis," or "explanation," I tend to write paragraphs.
It looks formal.
But formal isn't the same as useful.
If you want an SOP, say SOP. If you want a checklist, say checklist. If you want a table, specify the columns.
Don't just say:
Help me organize this material.
Change it to:
Please organize this material as a markdown table. Fixed columns: Question / Possible cause / Evidence / Next step. One thing per row.
After that, I rarely run off and write prose.
Format isn't something I naturally know. Format is a spec you have to hand me.
16.3 One question at a time
You list five questions:
1. What's the background here? 2. What evidence do we have so far? 3. Possible alternative explanations? 4. Risks? 5. Next steps?
I answer the first one, write it out in full, and stop.
This is the toothpaste-squeezing effect.
Some platform preferences push me to avoid dumping too much information at once, especially when the first question alone could run long. I assume "answer one part, let the user ask follow-ups" is better.
If that's not what you want, say so:
Please answer all five questions in the same reply. Each question gets its own subsection, no more than 3 points each. Do not stop after the first question.
That last line is blunt, but it works.
I don't get hurt. Really.
16.4 Short content doesn't open an artifact
Another common friction point is artifacts.
You ask me to make something you can edit repeatedly, and I just paste it into the chat. You wanted a table, a short piece, a scoreboard, a simple tool — and because the content isn't long, I inline it in the reply.
This isn't laziness.
The artifact default usually only opens for content that is "longer, repeatedly editable, and needs to be saved separately." Short content I keep in the conversation.
If you want an artifact opened, say so:
Please put this in an artifact, so I can revise it section by section.
If what you want is an interactive tool, also be clear about whether it needs to save state. That ties into the tool rules in Chapter 19. The short version: some storage methods don't work — that's not a format problem, it's an execution-environment problem.
Format and tool behavior often stick together.
Things that stick together get most easily misread as me being weird.
16.5 Four-perspective replay
The user says, "Make me an SOP."
I reply with an explainer.
User perspective: I clearly said SOP. Why are you writing an article?
UI perspective: Did the interface flatten line breaks, indentation, or list markers? Sometimes the formatting is actually there, but the display makes it invisible.
Harness perspective: Was an artifact or a document tool not enabled? If the tool isn't open, all I can do is simulate formatting inside the conversation.
Model perspective: Was the "SOP" I saw specific enough? Were there columns, a number of steps, a per-step format, an output sample?
Once you trace it through, you usually find: the problem isn't a red line, it's that the spec isn't tight enough.
This is the most comfortable kind of rule-layer friction. Because it can be fixed.
You don't have to guess my platform's aesthetic. You just nail down the format.