Write Persuasively Codex skill with AI-writing-tell checklist and sources

Write Persuasively

Turn the user's goal into a truthful, reader-centered argument that is easy to understand and act on. Favor clarity, brevity, and concrete language over cleverness or pressure.

Workflow

1. Establish the brief

Interview the user before drafting. Make the first response to an underspecified persuasive-writing request a compact interview, then wait for the answers. Do not draft in that same response or invent material parts of the brief. Ask only for important information that the user has not already supplied. Usually ask one compact batch of three to five short questions covering:

• Identify the outcome: What should the reader believe, decide, feel, or do?
• Identify the audience: Who are they, what matters to them, and what do they already believe?
• Identify support: What facts, proof, examples, or offer must the text include?
• Identify resistance: What doubts, risks, or objections could prevent the desired response?
• Identify the format: What channel, length, tone, voice, deadline, and constraints apply?

Treat existing answers and source material as part of the brief; do not ask the user to repeat them. If the request is already well specified, confirm the inferred brief in one sentence and proceed. Skip the interview only when the user explicitly asks not to be questioned or explicitly asks Codex to make assumptions. In that case, make the fewest reasonable assumptions, state only material assumptions, and proceed. A request such as “write an email” or “draft this” does not by itself waive the interview.

2. Build the argument

• State the desired reader response in one sentence.
• Choose the strongest true central claim or reader benefit.
• Select the one to three strongest supporting facts or reasons.
• Address the likeliest objection directly when doing so strengthens trust.
• End with a specific, proportionate next step.
• Preserve uncertainty and qualifications that matter. Never invent evidence or use deceptive pressure.

3. Draft for momentum

• Make the first sentence earn the second through curiosity, relevance, consequence, or a specific benefit. Avoid clickbait.
• Lead with the main point. Put background after the reader knows why it matters.
• Arrange ideas in the order the reader needs them: claim, support, objection if needed, and action.
• Use one paragraph for one stage of the argument.
• Match the language and level of formality to the audience and channel.

4. Prune and clarify

• Remove words that do not change the meaning.
• Prefer familiar, concrete words to abstractions and jargon.
• Write short sentences. Give each sentence one main thought.
• Prefer active constructions that present the actor before the action.
• Replace weak general words with precise, vivid ones when the tone allows.
• Cut unsupported superlatives, filler, throat-clearing, and repeated claims.
• Read the text aloud. Fix every point where the reader may stumble.
• Reduce cognitive load without patronizing the reader.

5. Remove machine-written tells

Edit for a specific human voice, not merely for the absence of errors. Treat the patterns below as a cluster: no single word or punctuation mark proves AI authorship, but several in one passage make the prose sound generated.

• Default to no em dashes in new persuasive copy. Use a period, comma, colon, or parentheses instead. Preserve an em dash only when quoting source text or when it is clearly part of the requested speaker's established voice.
• Cut manufactured contrasts such as “not just X, but Y,” “not X, but Y,” and “X rather than Y.” State the positive claim directly unless the contrast answers a real reader misconception.
• Avoid slogan formulas such as “X is a moment. Y is memory.”, “X in. Y out.”, and strings of polished sentence fragments. Use a concrete subject and verb.
• Do not use “honest,” “authentic,” “real,” “meaningful,” “thoughtful,” “quiet,” or similar virtue words as self-awarded praise. Name the behavior, limit, evidence, or tradeoff that earns the description.
• Watch for a high density of stock AI vocabulary: “delve,” “intricate,” “tapestry,” “realm,” “landscape,” “pivotal,” “crucial,” “robust,” “seamless,” “transformative,” “underscore,” “foster,” “showcase,” and “unlock.” Keep one only when it is the most precise word.
• Avoid repeated rule-of-three phrasing, matched card sentences, and mechanically parallel headings. Lists and parallel structure are useful when the subject actually has that shape; they should not manufacture a sense of completeness.
• Replace personification with the actual product or actor. A project stores settings, a service checks sources, and an agent prepares a brief; they do not “remember what matters,” “keep watch,” or produce a “quiet feed.”
• Replace abstract labels, slash constructions, and clever filler headings with labels that name the content. Phrases such as “human + machine,” “works where you work,” “what to build,” and “the short version” usually hide the useful noun.
• Remove canned openings and transitions such as “In today's fast-paced world,” “Whether you're…,” “At its core,” “It's worth noting,” “The result?”, and “In conclusion.” Begin with the claim, fact, or situation instead.
• Avoid hypophora, excessive bold emphasis, decorative emoji, and a heading or card for every thought unless the requested format or source voice calls for them.

After the rewrite, scan once for em dashes and stock phrases. Then read the draft aloud for uniform cadence: generated prose often gives every sentence the same polish, length, and rhythm. Keep useful technical lists, exact product labels, necessary caveats, and genuine contrasts.

6. Deliver the useful version

Give the polished text first unless the user asked for strategy or options. For a rewrite, preserve the user's meaning, facts, and recognizable voice. Add a short rationale, alternatives, or a message map only when it helps the user choose or improve the result.

Final check

Before delivering, verify that:

• The opening creates a clear reason to continue.
• The reader can identify the point on the first pass.
• Every sentence advances the argument.
• Claims are supported and accurately qualified.
• The likely objection is handled or intentionally omitted.
• The requested action is clear and easy to take.
• The draft matches the requested speaker's identifiable voice.

Complete source text

Keep the following source text intact in this skill.

The Day You Became A Better Writer — by Scott Adams

I went from being a bad writer to a good writer after taking a one-day course in “business writing.” I couldn’t believe how simple it was. I’ll tell you the main tricks here so you don’t have to waste a day in class.

Business writing is about clarity and persuasion. The main technique is keeping things simple. Simple writing is persuasive. A good argument in five sentences will sway more people than a brilliant argument in a hundred sentences. Don’t fight it.

Simple means getting rid of extra words. Don’t write, “He was very happy” when you can write “He was happy.” You think the word “very” adds something. It doesn’t. Prune your sentences.

Humor writing is a lot like business writing. It needs to be simple. The main difference is in the choice of words. For humor, don’t say “drink” when you can say “swill.”

Your first sentence needs to grab the reader. Go back and read my first sentence to this post. I rewrote it a dozen times. It makes you curious. That’s the key.

Write short sentences. Avoid putting multiple thoughts in one sentence. Readers aren’t as smart as you’d think.

Learn how brains organize ideas. Readers comprehend “the boy hit the ball” quicker than “the ball was hit by the boy.” Both sentences mean the same, but it’s easier to imagine the object (the boy) before the action (the hitting). All brains work that way. (Notice I didn’t say, “That is the way all brains work”?)

That’s it. You just learned 80% of the rules of good writing. You’re welcome.

Sources and rationale

This gist contains the updated and validated write-persuasively Codex skill, including its full checklist for editing machine-written tells.

The checklist treats these patterns as clusters, not proof of AI authorship. It draws on:

• Wikipedia: Signs of AI writing, a practical field guide covering patterns such as negative parallelism, repeated rule-of-three phrasing, stock vocabulary, and formulaic em-dash use.
• Why Does ChatGPT “Delve” So Much? Exploring the Sources of Lexical Overrepresentation in Large Language Models, on overrepresented vocabulary in LLM output.
• Do LLMs Write Like Humans? Variation in Grammatical and Rhetorical Styles, on systematic grammatical and rhetorical differences between LLM and human writing.

These references informed the prevention checklist. The skill turns the findings into practical editing rules for persuasive copy.