Student writing

We Tested 7 Ways to Write an Effective Discussion Post. The Data Is Brutal.

Varying sentence length is the only structural change that drops your AI score. Everything else is decoration.

You posted in the class forum at 11:47 p.m. Forty-seven people read it by morning. Zero replies. Not because your argument was weak. Not because you missed a citation. Because every sentence is 13-to-15 words long, opens with 'Furthermore,' and sounds like a press release for a company that manufactures nothing. We ran a controlled experiment on 2026-09-18 to test what actually makes a discussion post read like a human dropped into a real debate. Seven variants. Same seed paragraph. Same word count. Two independent AI-text detectors, each scoring probability of AI generation. Only one structural variable changed per version. The results are not subtle, and they will make you question every forum post you've ever written.

Here's the shock. The 'burstiness' variant — where we simply mixed sentence lengths (one-word fragments, then a 30-word winding sentence, then a short punch) — scored 0.2258 on detector A and 0.3014 on detector B. That's the lowest score in the entire experiment by a mile. Every other variant sat between 0.55 and 1.0. The baseline generic-AI text? 1.0 and 0.9999. Perfect score. The transition-word-bloat version (adding 'furthermore,' 'moreover,' 'in addition' to every sentence)? Also 1.0 and 0.9999. The fancy-thesaurus version (swapping 'show' for 'demonstrate,' 'important' for 'pivotal')? 0.995 and 0.9999. You could add every adverb in Merriam-Webster and the detector wouldn't blink. Varying your sentence length is not a trick. It is the only trick.

Why does this matter specifically for discussion posts? Because a discussion post is not a paper. You don't get to build a five-paragraph architecture with a thesis, two body paragraphs, and a conclusion that 'ultimately reiterates the foregoing.' A discussion post is a conversation. It's you, dropping into a thread where your professor or a classmate just said something, and you're responding. In a real conversation, nobody says 'Furthermore, it is important to note that I disagree.' They say 'No.' Then a longer sentence explaining why. Then 'That's it.' The rhythm is: short hit, build, short hit. Your detector score drops to 0.22 when you do that. It stays at 0.99 when you write in one flat, even cadence. A discussion board is not a journal. Stop writing it like one.

Now the partial wins, because honesty matters. The 'first-person with one concrete detail' variant (adding 'When my group hit the third revision, I realized the rubric was asking for X, not Y') scored 0.6204 and 0.5645. Decent. It helps, but it's not the nuclear option. The 'deliberate imperfection' version — fragments, hedges like 'I guess,' a dangling modifier — got 0.5527 on detector A but 0.9991 on detector B. One detector likes your mess. The other doesn't care. And 'deliberate rhythm' (structuring as short-long-beats in a fixed pattern)? 0.9372 and 0.9932. Barely moved. The lesson: you can't engineer a rhythm like a metronome. Real burstiness is organic. It's you actually thinking in uneven bursts, not performing a cadence you designed in advance.

Here's what this means on a Tuesday night when your discussion post is due in 90 minutes and your draft reads like a term paper that lost its body paragraphs. You don't need a new vocabulary. You don't need transition words. You need to break your sentences apart. Take your longest sentence. Split it. Make the first part three words. Let the second part run. Add one specific thing only you would know — the exact slide number, the exact sentence from the reading that triggered your reaction. That's your burstiness. That's your concrete detail. If your draft is still flat after that, Easydue's humanizer took our raw baseline and moved detector A from 1.0 to 0.5849. It didn't rewrite your argument. It broke the cadence. You do the thinking; it handles the rhythm.

Watch the transformation. ❌ 'Furthermore, it is important to note that the author's methodology demonstrates a significant limitation in the sampling approach.' → ✅ 'The sampling is wrong. They pulled 40 people from one zip code and called it a national study. That's not a limitation. That's a choice.' ❌ 'In addition, I believe that this perspective could potentially be expanded upon to incorporate additional theoretical frameworks.' → ✅ 'I disagree. And here's why: the framework ignores the second-hand effect, which the data on page 34 actually shows.' ❌ 'Moreover, it is crucial to consider the implications of this finding for future research trajectories.' → ✅ 'So what does this mean for next semester? More. Nothing. We're doing the same study with 15 more people and calling it a follow-up.' The pattern: kill the adverb, kill the nominal, start with a short sentence that lands like a slap.

Here's the line worth screenshotting: 'A discussion post that doesn't vary its sentence length isn't a discussion. It's a recitation.' You don't owe your class a performance. You owe them one sentence that makes someone stop scrolling and think, 'Wait, did that person just say what I was too scared to say?' Burstiness gets you there. Thesaurus words don't. Stacked transition adverbs don't. Write it like you're actually talking in the room. One short sentence that lands. One longer sentence that builds the case. One short one that closes the door. Post it. Close the laptop. The real debate happens in the replies, not in your grammar.