How to humanize a discussion board post
Respond to somebody. Discussion posts read as generated because they answer the prompt in general terms and never engage with what a classmate actually said. Name the post you are replying to, disagree with something specific in it, and the format fixes itself.
Typical length · 200–400 words, which makes every weak sentence a large share of the whole. · Last reviewed 16 August 2026
Before and after
Before
I found this week's readings very interesting. The author raises several important points about organisational culture. I agree that culture plays a significant role in employee retention. This is something that many organisations struggle with today.
After
Priya's point about retention being a culture problem rather than a pay problem is the one I keep getting stuck on. At my last job we got a raise and three people still left that quarter, and the exit interviews all said the same thing about the Monday meetings. The reading would call that culture. I am not sure the distinction survives contact with a mortgage.
Four sentences addressed to nobody became a reply to a named classmate, a specific experience, and a disagreement stated plainly. Length barely moved; the post now has an author.
The format punishes generic writing more than any other
A discussion post is 250 words with no room to hide. In an essay a flat paragraph sits among fifteen others; here it is the entire submission, and the reader is a tutor who is about to read thirty more posts answering the same prompt.
That comparison is what exposes the generated ones. Thirty posts that each find the readings interesting and each agree that the topic is important are indistinguishable from each other, and the tutor notices the pattern across the set rather than in any single post. Nobody needs software for this.
It also means the fix is unusually cheap. One specific detail — a number, a name, an incident, a disagreement — is a large proportion of a 250-word post, so a single concrete sentence changes the whole thing.
Reply to a person, not to the prompt
The structural error is answering the assignment when the format is a conversation. A post that begins by restating the question is writing an essay introduction in a thread, and it has already lost the thing the format is for.
Start from the last post you actually read. Name the person, quote or paraphrase the specific claim you are responding to, and say where you land on it. That gives you an opening sentence nobody else in the class can write, because nobody else is responding to that claim in that way.
Disagreement works better than agreement, and not for rhetorical reasons. Agreeing requires restating; disagreeing requires you to say what you think instead, which is the part that carries your voice. It does not have to be a strong disagreement — "I am not sure that follows" is enough to force a real sentence.
The one specific thing
Every good discussion post has at least one detail that could not have been generated, because it came from your life or your reading rather than from the prompt. A figure from the paper with the page number. Something that happened at work. A question the reading left you with that it did not answer.
This is also the honest version of what people mean by writing that sounds human. It is not a style. It is information the writer had, which is the same reason concrete detail lowers a detection score without anyone setting out to lower one.
If you genuinely have no such detail, that is worth noticing before it is worth disguising, because it usually means you have not finished the reading. The post is a symptom.
Formats with the same problem
Essays — the long form of the same argument problem, with a worked before and after.
Emails — short, frequent, and judged on whether it sounds like a person wrote it.
Questions
- Are discussion posts even scanned for AI?
- Sometimes, though most detectors publish a minimum length well above a typical post, and a score on 250 words is much less reliable than a score on an essay. The realistic risk in this format is a tutor reading thirty near-identical posts, which is not a software problem.
- How specific is specific enough?
- One thing a classmate could not have written. A page number, a figure from your own experience, a named example from your field. Not a stronger adjective — a fact.
- Is it acceptable to disagree with a classmate?
- It is usually what the format is asking for, and most rubrics reward engagement over agreement. Disagree with the claim and not the person, and quote what they actually said rather than a version of it.
- What if I am posting first and have nobody to reply to?
- Then reply to the reading in the same way: name the specific passage, say what you think it gets wrong or leaves open, and give the detail that made you think so. The move is the same — respond to something particular rather than to the topic.
What to watch for
- Naming a classmate means quoting them accurately — check what they actually wrote before responding to it.
- Short posts are below most detectors' minimum length, so a score on one means even less than usual.
- If your course requires disclosure of AI assistance, disclose it. A rewrite does not change that obligation.
If your writing gets flagged
Rewriting for rhythm and specificity tends to lower detection scores, because that is what detectors read as human. It is not a guarantee — detectors disagree with each other and change without notice, and we do not promise a result from any of them.