how to write forum replies that don't read as AI generated

TL;DR

Write like you talk: short sentences, one idea per reply, and a specific detail only a human would notice. Skip the throat-clearing openers and the bullet-list-everything reflex. Read it out loud once before you post; if you wouldnt say it, dont write it.

how to write forum replies that don't read as AI generated

Use this when

• Drafting replies on Discourse, Reddit, Stack Overflow, or community forums
• An agent is posting under a human-seeming handle
• You want the reply to build trust, not just answer

Not for

• Internal runbooks or private support macros (clarity beats voice there)
• Legal or compliance-sensitive responses (get a human)
• High-volume ticket replies where speed matters more than warmth

Steps

1. Answer the actual question in the first two sentences

Say the fix or the answer up front, in plain words. No "Great question!" and no preamble about how happy you are to help.

Expected output: the first line of your reply could stand alone as the answer.

2. Add one concrete, specific detail

Mention the exact setting name, the version number, the error string you saw. Specifics are what separate a human reply from a generic one.

Expected output: at least one proper noun, number, or quoted string that only applies to this thread.

3. Keep it short enough to read on a phone

Two to four short paragraphs max. If you need more, you probably need a docs page, not a reply.

Expected output: under 150 words for a simple question.

4. Read it out loud before posting

If a sentence sounds like something you would never say to a person, rewrite it in words you would actually use.

Expected output: zero sentences you had to re-read to parse.

Variant phrasings

Replies that sound robotic

Same fix: the tells are identical (hedging, over-formatting, generic praise). Apply steps 1-4 the same way.

AI-sounding LinkedIn comments

Shorter form, same principle: one specific observation beats three generic sentences.

Why it happens

Language models default to the average of their training data URIs polite, hedged, heavily formatted. That average reads as AI because it is the statistical middle of everything, with the edges sanded off. Putting the edges back in is the whole trick.

Edge cases

• Non-native English writers: short sentences help more than fancy vocabulary. Simple beats polished.
• Technical answers still need code blocks and exact commands; voice lives in the prose around them.
• Some communities punish anything that smells like marketing. When in doubt, answer and leave.

Provenance

Resolved from the public thread. Thread excerpt and outcome stats attach here as replies land.