How to Write the Discussion Section
The discussion is where a competent paper becomes a convincing one — and where most drafts quietly fall apart, usually by restating the results in longer sentences. Here is a five-move structure drawn from the standards editors actually apply, with an example of the same paragraph written badly and then well.
What the discussion is for
Your results section answers what happened. Your discussion answers so what. The ICMJE Recommendations, which several thousand journals follow, put it plainly: the discussion should emphasise the new and important aspects of the study and place the findings in the context of the totality of the relevant evidence — and should not repeat in detail data already given in the introduction or results.
That single instruction kills the most common failure mode. If a reader can delete a paragraph of your discussion and lose no information that isn't already in Table 2, that paragraph is padding.
The five moves
1. Answer the question you asked
Open with your main finding stated as an answer, not as a summary of procedure. APA's reporting standards for quantitative research ask the discussion to open with a clear statement of support or non-support for each hypothesis, distinguishing primary hypotheses from secondary ones. Do this in the first two or three sentences, before any hedging.
A non-significant result is still an answer. Write it as one: the data did not support the hypothesis — not "the hypothesis was partially supported in some conditions," which is how a null finding gets dressed up as a positive one.
2. Explain why you got that result
Offer a mechanism. ICMJE advises beginning the discussion by briefly summarising the main findings and then exploring possible mechanisms or explanations for them. This is the part reviewers read most carefully, because it is the only part that shows you understand your own data rather than merely having collected it.
Give a genuine alternative explanation as well, then say why yours fits better. A discussion that entertains no rival account reads as advocacy.
3. Compare with what is already known
APA's standards ask explicitly for the similarities and differences between your results and the work of others. The differences matter more. "Consistent with Author (2023)" is one clause of work; explaining why your effect is half the size reported in a study with a different sample or measure is a contribution.
This move is also the most citation-dense part of your paper. Every comparison you make is a claim about someone else's findings, and it needs a reference that a reader can open and check — which is exactly where drafts written quickly, or with an AI assistant, tend to break.
Paste a DOI, URL or book details into our free citation generator and get a formatted reference and in-text citation in APA, MLA, Chicago, Harvard or IEEE — for the sources you actually read.
4. State your limitations
Name them yourself, specifically, and say what each one does to the interpretation. Vague self-criticism ("the sample was small") buys nothing; a limitation tied to a consequence ("with 42 participants the study was underpowered to detect the interaction, so the null result there should not be read as evidence of no effect") buys credibility.
Reporting guidelines will tell you which limitations count in your field. The CONSORT 2025 statement for randomised trials, for instance, gathers them into one checklist item covering sources of potential bias, imprecision, generalisability, and — where relevant — multiplicity of analyses. Those four headings work as a prompt list well beyond trials.
5. Say what follows
Close with implications for practice, policy or future research, and link your conclusions back to the aims you set in the introduction. Then stop. ICMJE warns against unqualified statements and conclusions not adequately supported by the data, and against blurring the line between statistical and clinical significance — a p value below .05 is not a recommendation to change anything on its own.
"Future research should investigate this further" is not an implication. Name the study you would run next and what it would settle.
Before and after
Here is a comparison paragraph as it usually arrives in a first draft:
Our results showed a significant improvement in the intervention group (M = 4.2, SD = 0.8) compared to the control group (M = 3.1, SD = 0.9), p < .001. This is a very important finding which proves that the intervention works. Similar results were found by previous research. Future research should investigate this further.
The same paragraph doing the five moves:
Self-reported engagement was higher after the intervention, supporting our primary hypothesis. The most plausible mechanism is the weekly feedback cycle rather than the content itself, since the effect appeared after week two, when feedback began, and not after week one. The direction matches Author (2023), although our effect was roughly half the reported size — likely because their participants were volunteers and ours were enrolled by default, which plausibly lowered baseline motivation. Because engagement was self-reported and the trial ran for eight weeks, we cannot say whether the gain persists or reflects reporting bias; a follow-up using platform log data at six months would settle both.
Same study, same numbers, no new data. The second version explains, compares, qualifies and points forward — and it never uses the word proves.
A short checklist before you submit
- The first paragraph states an answer, not a method recap.
- No statistic appears here for the first time.
- At least one difference from prior work is explained, not just noted.
- Every limitation is paired with what it changes.
- No sentence claims causation the design cannot support — suggests, is consistent with and indicates are not interchangeable with proves.
- Every source named in a comparison is one you have opened and can find again.
That last point is worth a paragraph of its own. The discussion cites more heavily than any other section except the literature review, and it is written last, under deadline — the exact conditions under which people paste in a plausible-looking reference from a chatbot. Language models generate references that look real and are not, and a fabricated citation in your discussion is the one place a reviewer who knows the literature is guaranteed to look.
Draft from sources that actually exist
CiteGood writes from real papers and formats every citation for you — so the comparisons in your discussion point at references a reviewer can open.
Try CiteGood free →Frequently asked questions
How long should the discussion section be?
There is no universal word count — the only binding answer is your target journal's author guidelines or your department's rubric. A useful working test is structural rather than numeric: if all five moves (answer, explain, compare, limit, imply) are present and none of them is a restatement of your results tables, the length is about right.
Should the discussion and conclusion be separate sections?
That depends on the journal. Many combine them into a single 'Discussion' that ends with a short concluding paragraph; others require a separate 'Conclusions' heading. Check the author guidelines of the journal you are submitting to, or your unit guide, before you decide — and when they are separate, the conclusion should link back to the aims stated in your introduction rather than introduce new argument.
Do I have to include limitations?
Yes, in almost every context. The ICMJE Recommendations tell authors to state the limitations of their study, and reporting guidelines make it an explicit checklist item — CONSORT 2025, for randomised trials, folds sources of potential bias, imprecision, generalisability and multiplicity of analyses into a single trial-limitations item. Naming limitations is a mark of competence, not a confession.
What tense should the discussion be written in?
Mix them deliberately. Use the past tense for what you did and what you found ('participants completed', 'scores increased'), and the present tense for interpretation and implications — APA Style's Discussion Phrases Guide recommends the present tense for discussing what the results mean, because the meaning still holds now.
Can I introduce new results in the discussion?
No. Every number that appears in the discussion should already have appeared in the results. The ICMJE Recommendations also warn against repeating in detail data given elsewhere in the manuscript, so the discussion should reference findings, not re-report them — one summary figure to anchor an interpretation is fine, a second results section is not.