How to Humanize Results and Discussion Sections

A results section reports that the intervention group scored higher than the control group, a specific number sitting in a table two pages earlier. The discussion section, covering the same finding, says the intervention group scored higher, which may suggest a possible positive effect that could warrant further investigation. The second sentence adds nothing the first did not already establish, only four hedges and a longer sentence.

This is the real reason that these two sections start to sound alike when they are both drafted by a language model. Learning how to humanize AI discussion section drafts means restoring a sentence that commits to a specific claim about what the finding means, instead of another hedge stacked on the one before it.

Why Results and Discussion Start to Sound the Same

The purpose of a results section is to tell us what happened: either with numbers or with a description specific enough to count as numbers. The purpose of a discussion section is to tell us what that result means, how it relates to other results, and what it doesn't resolve. When you collapse those two sections together into one, you have a paper in which there're two sections that do the same job twice: once with a number attached and once without.

A language model collapses it by default, because both jobs get drafted from the same instinct: describe the finding cautiously and move on. The result is a results section that hedges findings it should just be reporting, and a discussion section that reports findings it should be interpreting. Neither sentence commits to anything a specific reader could disagree with, and that is the actual tell, more than any single word choice.

Question being answered	Results section	Discussion section
What happened?	States the finding directly: a number, a comparison, a description	Assumes the reader already has the number and moves past it
What does it mean?	Not this section's job; a number does not interpret itself	States what the finding indicates, and what it does not settle
Typical failure mode	Hedging a number that needs no hedge, such as 'results appeared to show a difference'	Restating the number instead of interpreting it, such as 'the results showed a higher score'
A sentence that passes	The treatment group improved by nine points; the control group by two.	The gap looks driven by the shorter interval, not the method itself.

The failure-mode row is worth checking first in any AI-assisted draft, because the other rows tend to follow once that one is fixed.

What a Results Section Should Actually Do

A results section states what was found, in the order the analysis produced it, without arguing for what it means yet. 'Group differences were observed across conditions' is not a results sentence. 'The treatment group improved by nine points; the control group improved by two' is a results sentence, because a reader can check it against the table and move on.

Vary how each finding is introduced rather than repeating one template for every result. Open one sentence with the number, another with the comparison, another with the condition that produced it. A results section that reports an unexpected finding in a plain, direct sentence, without hedging it in advance, almost always reads as more human than one where every result arrived exactly as predicted.

What a Discussion Section Should Actually Do

TextPulse's broader guide to humanizing academic writing already makes the case that the discussion is where your own voice has to show up. What is worth adding here is the mechanism: a discussion section exists to say something a table cannot say on its own, why the finding matters, what it rules out, and where it stops applying. That requires committing to an actual interpretation rather than gesturing at one.

'This finding aligns with the broader literature and suggests several possible directions for future work' interprets nothing, because a version of it could close almost any discussion section in almost any field. A discussion sentence can be specific and still carry real uncertainty: 'the effect likely reflects the shorter training period rather than the method itself, though the study cannot rule out a cohort difference' names a mechanism, states a preference between two explanations, and admits what it cannot settle, all at once.

Where the Line Between the Two Gets Blurry

Not all papers separate these sections. In some journals, especially those in the qualitative and case study traditions, the result section is often combined with the discussion section. A combined result/discussion section, though, doesn't eliminate the distinction above. It just removes the marker (i.e., the heading) for that distinction. Inside a combined result/discussion section, a paragraph either reports a finding or interprets one, and the hedging tell works the same way at the sentence level.

Qualitative work shifts the ratio further. Presenting a theme from interview data is already partly interpretive, since choosing what counts as an instance of the theme is itself an analytic decision, so the reporting-only paragraph a quantitative results section relies on may not exist in the same form. The hedging tell still applies: a theme description stacked with 'may potentially suggest' qualifiers is exactly as machine-sounding in a qualitative paper as it is in a quantitative one.

The Hedging Tell: Sentences That Commit to Nothing

Hedging has a real job in academic writing: flagging genuine uncertainty a reader needs to know about. A single hedge is normal caution. The tell is a hedge stacked on another hedge until the sentence stops claiming anything a reader could push back on. 'This may potentially suggest a possible link' hedges the same idea three times in five words, and a discussion section built from sentences like it reads as machine-generated regardless of the finding underneath.

Count the hedges in a single sentence as a rough check. One qualifier is normal scientific caution: 'this suggests', 'this may indicate', 'this appears to'. Two in the same sentence is worth a second look. Three is the pattern a checker, and a supervisor, will both notice, because the sentence has stopped saying anything specific enough to be wrong.

How to Humanize AI Discussion Section Drafts Without Overclaiming

Start by finding every sentence that restates the results section instead of interpreting it. If a discussion sentence could be moved into the results section unchanged, it has not done its job yet. An AI humanizer helps here only if it is pointed at the hedging and the restated sentences specifically, rather than run blindly across the whole section, since the numbers never needed rewriting in the first place.

Then remove the second and third hedge from every sentence that has more than one, and see what is left. Often what remains is a perfectly defensible claim the hedging was burying. Running the paragraph through an academic tone converter afterward can catch the vaguer phrasing that survives a first pass, though deciding what you actually think the finding means is not something a tool can do for you.

One more check worth doing by hand, before any tool: read the results section and the discussion section back to back and ask whether the second could have been generated from the first without reading the rest of the paper. An essay checker run over the finished draft catches leftover restatement a close read misses, but the question above is the one that actually finds it first.

At full document scale the problem changes shape, and humanizing a thesis is covered on its own page. If the discussion has run long, reducing word count in a thesis is a separate skill worth reading first.

A results section and a discussion section can report the exact same finding and still do completely different jobs on the page. The first earns its place by being checkable. The second earns its place by making a claim someone could actually disagree with, which is exactly what a stack of hedges is built to avoid.

Related research: the humanizers discussed above are compared head to head in A Controlled Comparison of AI Text Humanizers on Academic Writing, a TextPulse Research working paper with open data and code.