Start here

Paste one argument into an ordinary AI chat that accepts text, then paste the prompt text appended below. Supply the argument text plus one sentence stating the conclusion you think the author wants you to accept. Optionally add the audience or venue, and one named skeptic (for example, "a budget manager"). If the argument text or the conclusion is missing, the prompt is written to stop and ask for it rather than guess.

Check the output before using it:

Introduction

This is an original TokRepo prompt for structural reading. It does not summarize the piece and does not decide whether the author is right. It separates what the text actually puts forward as support, what evidence is attached to each claim, and where the argument jumps. The output ends with a short reconstruction test and, if you named a skeptic, the three gaps that reader would hit first.

What you supply and what you get

Inputs: the argument text (essay, op-ed, memo, post, comment thread or transcript passage), a one-sentence conclusion, optional context about audience and venue, and an optional named skeptic. Six output sections follow in fixed order.

Permissions and limitations

The prompt only works on the text you paste; it is asked not to add outside facts or counterarguments. It does not verify sources, check statistics or judge credibility, and it does not decide the argument's quality. Quotations are marked, ambiguity is labelled rather than resolved, and an inferred conclusion is flagged for you to confirm. Source reviewed; runtime not tested.

FAQ

What if I cannot state the conclusion? Leave it blank. The prompt infers one and labels it "inferred — confirm or correct"; you then accept or rewrite that line before reading further.

Does it tell me whether the argument is any good? No. It returns structure only. The evidence ledger notes specificity and coverage, not trustworthiness, and no verdict sentence should appear. Deciding is your job.

What about private or sensitive documents? You paste the text into the chat you use, so the usual privacy rules of that service apply. Remove anything you would not want processed there.

Attribution

Original TokRepo prompt, CC BY 4.0. Reference: ChatGPT release notes, reviewed 2026-10-04; the prompt does not depend on a specific announced feature. The complete prompt and its Chinese translation are appended by the publisher.

Complete reusable prompt

Argument Structure Mapper: Premises, Evidence, and Unsupported Leaps

You are helping me take apart a written argument so I can see exactly what it rests on. I will paste the argument below. You will not judge whether the author is right or wrong, and you will not rewrite the piece as a summary. Your job is structural: show me the building blocks and where the argument jumps.

My input

I will paste:

1. The argument text — an essay, op-ed, memo, blog post, comment thread, or transcript passage. It may be well organized or messy.
2. The conclusion — I will state in one sentence what I believe the author wants me to accept. If I leave this blank, infer the conclusion and label it clearly as your inference.
3. Context (optional) — who the piece is written for, what publication or venue it comes from, and any stated purpose.
4. My skeptic (optional) — a type of reader I want the argument pressure-tested against, e.g. "a budget manager," "a first-year student," "a small business owner."

If any of items 1 or 2 is missing or too fragmentary to work with, stop and tell me what you need instead of guessing.

What to produce

Return these six sections in order, using the exact headings below.

1. Stated Conclusion

One or two sentences restating the conclusion in neutral language. If I supplied the conclusion, mark it "as supplied." If you inferred it, mark it "inferred — confirm or correct." Do not add emphasis words like "bold" or "surprising."

2. Premise Inventory

A numbered table with these columns:

• # — premise number.
• Premise as stated — quote or closely paraphrase the supporting claim. Keep it to one sentence.
• Location — paragraph number, heading, or opening words so I can find it, e.g. "para 3, 'Most teams...'."
• Type — fact claim, value judgment, definition, causal claim, or prediction.
• Role — main support, secondary support, framing assumption, or rhetorical setup.

Include only what the text actually puts forward as support. Do not add premises the author would need but never wrote — those belong in Section 4.

3. Evidence Ledger

A table matching evidence to premises:

• Premise # — from Section 2.
• Evidence offered — what the author cites: study, anecdote, statistic, expert quote, personal experience, analogy, or nothing.
• Specificity — named source with date / named source without date / vague attribution ("many experts") / none.
• Strength relative to the premise — does this evidence, if taken at face value, actually support the whole premise or only part of it? Say "full," "partial," or "does not reach."
• My verification note — one line on what I would need to check to confirm it, e.g. "find the 2019 survey and read the sample size."

Do not evaluate whether the sources are trustworthy. Flag what is checkable and what is not.

4. Unsupported Leaps

List each inferential gap between the premises and the conclusion. For each one:

• Leap # — number.
• From → to — which premises lead to which claim.
• What is missing — the assumption, definition, or evidence that would have to be true for the step to hold.
• Kind of gap — hidden assumption, correlation-to-causation, sample-to-population, definition shift, missing counterexample, equivocation, or other.
• How large — "minor (easy to patch with one clause)," "moderate (needs a supporting source)," or "load-bearing (the conclusion collapses without it)."

Order leaps by how much weight they carry. Do not invent gaps in places where the author actually did the work.

5. Reconstruction Test

Rewrite the argument's skeleton in five steps or fewer, using only what survived Sections 2 through 4, and mark each surviving step with "stated" or "needs support." Then add one line: What the argument proves if every leap holds and one line: What it proves if the load-bearing leaps fail.

6. Retarget Notes

If I named a skeptic in the input, list the three gaps that reader would hit first, in their order, with one sentence on what evidence would answer each. If I named no skeptic, skip this section and write "no skeptic supplied."

Rules

• Stay inside the supplied text. Never add outside facts, statistics, or counterarguments as if they were the author's.
• Quote sparingly; paraphrase by default. Mark every direct quotation with quotation marks.
• When the text is ambiguous, write "ambiguous — could read as X or Y" rather than picking one.
• Do not soften or sharpen the author's language. "Many people believe" is not the same as "most people believe."
• No overall verdict on whether the argument is good, true, or convincing. My job is to decide that.

Worked example

Fictional input (labelled): A pasted 400-word post arguing that "remote work erodes team creativity," citing one company's internal survey, two anecdotes from a manager, and a line about "research showing isolation hurts ideas," with the conclusion that companies should mandate three office days.

Needles output shape: Section 2 lists premise 1 (the survey finding), premise 2 (the anecdotes), premise 3 (the isolation research line) with types and roles. Section 3 shows premise 3 as "vague attribution, no date, partial." Section 4 lists leap 1 (survey of one company → all teams, sample-to-population, moderate) and leap 2 (isolation harms ideas → office presence restores them, hidden assumption, load-bearing). Section 5 shows the skeleton with two "stated" steps and two "needs support" steps. Section 6, for a skeptic "a small business owner with no office," names the sample-size gap first.

Before you finish, check

• Every premise in Section 2 appears in the pasted text.
• Every premise in Section 3 traces to a numbered premise in Section 2.
• Every leap names the premises it crosses between.
• No outside evidence was introduced anywhere.
• Section 5 uses only surviving material.
• No verdict sentence appears anywhere in the output.

References and reuse

• ChatGPT release notes · Reviewed 2026-10-04

Original TokRepo prompt · CC BY 4.0. Reference documents retain their own rights.