marketing-content-reader-test

Content Reader Test

Checks whether the text works on someone who has never seen the idea, the sources, or this conversation. The writer sees what they meant to write; a fresh reader sees only what is written. Of the four passes, this one tends to find the most.

Round, revision, and waiver rules: docs/content-generation.md, section "Review passes".

Inputs and output

Procedure

1. Make the reader's copy at tmp/reader-test/{slug}.md: strip HTML comments and all frontmatter except title. Keep [ASK: …] markers — the reader should notice them. Write the file rather than pasting text into the prompt: retyped text loses links and table cells, and the reader then reports problems that are not in the draft.
2. Run a separate agent (Agent tool / subagent) with no conversation context. The prompt holds only: the persona as a reader (role, company type, situation — from the persona file, without the angle or buyer question), the path to the reader's copy (and no other paths), and the questions below.
3. Questions — answer point by point, quoting the text: 0. The title is "{title}". If you had typed this into a search engine, does the text answer it? Quote the sentence with the answer and give its paragraph. Is the text more about the question, or about the author's firm and its offer?
   1. What is the main point, in one sentence?
   2. What would you do differently in your company after reading it?
   3. What is unclear, or assumes knowledge you do not have?
   4. Where would you stop reading, and why?
   5. Which claim is the least supported?
   6. Does anything sound machine-generated? Quote it.
   7. Does the text contradict itself anywhere?
   8. Is the mention of the firm helpful or pushy?
   9. Score it 1–10 as something you would forward to a colleague.
4. Compare the angle. Set answer 1 against the idea's unique_angle and buyer_question: match (same meaning), partial (narrower, broader, or shifted emphasis), mismatch.
5. No subagent available (a chat in a browser, for example): give the user a ready prompt to paste into a new conversation, and wait for the answers.
6. Write the report. Do not edit the draft.

Report

## Reader test

- Round: N · Date: YYYY-MM-DD · Reader: {persona as described} · Fresh agent, no context
- Angle (idea): "…" · Angle (read): "…" · Match: match | partial | mismatch
- Score: n/10
- Verdict: pass | revise

### Reader's answers
### Fixes
| # | Place (quote) | Problem per the reader | Proposal |
|---|---------------|------------------------|----------|

### Earlier rounds

Set reader_test: pass or reader_test: revise in the frontmatter.

Gate

Passes when, in question 0, the reader found a concrete answer to the title's question in the first two paragraphs and did not read the text as being about the firm's offer; the angle is match; the reader found no contradiction; and no [ASK:] is left. A missing answer to the title, or "I would stop here because there is no number", always blocks — it may not be moved to "for the author". Points from questions 3–6 go to the drafting skill as fixes; the human decides which to apply.

A score that stalls around 7/10 over several rounds usually means the draft lacks a real case, not better wording. Say so in the report, and ask for an author or proof source rather than spending another round on phrasing.