How to Write a Thesis Conclusion: What Belongs In It, and What Must Not

The short answer

A thesis conclusion does four things: it answers the research question directly, ideally in one sentence; it condenses the findings that support that answer; it states the limitations of the study; and it points to what should be examined next. Five to ten percent of the thesis is the usual length. Nothing new appears in it — no new sources, no new arguments, no results the reader has not already seen.

The commonest reason a conclusion is weak is that it circles the research question instead of answering it. If you cannot state the answer in a single sentence, that is a finding about the thesis rather than about your prose, and the fix is in the body, not here.

The four jobs of a conclusion

1. Answer the research question. Early, directly, in one sentence. Qualify it afterwards if you must — but the answer comes first, and a reader should never have to hunt for it.
2. Condense the supporting findings. Two to four sentences, each visibly tied to the answer. Not a chapter-by-chapter recap; if your conclusion follows the order of your table of contents, it is a summary rather than a conclusion.
3. State the limitations. Sample, method, timeframe, access. This does not weaken the thesis — it demonstrates methodological judgement, and examiners look for it explicitly. A limitation you name is far harder to attack than one they discover.
4. Give an outlook.What question follows from yours? One paragraph, not a page, and specific enough to be actionable: “more research is needed” is true of everything ever written.

In applied disciplines a fifth element is often expected: practical implications, stated as concretely as your evidence allows. In a doctoral thesis, the contribution to knowledge is usually stated explicitly here too, sometimes as its own numbered list.

What must never appear in a conclusion

• New sources. A citation appearing for the first time in the final chapter says the argument was not finished in the body. Either it belongs in the body, or it does not belong in the thesis. This is the clearest rule on the list and it is broken constantly.
• New arguments. The conclusion closes; it does not open. If a thought genuinely deserves space, it goes in the discussion where it can be supported.
• New results. A number that appears here for the first time belongs in the results chapter. Examiners notice this immediately because it is impossible to verify against anything.
• A restatement of the introduction. Both chapters concern the same question: one asks it, the other answers it. If they read alike, the answer is missing.
• Apologies. “Unfortunately it was not possible within the scope of this work to …” State limits factually, not regretfully.
• Overreach.Eight interviews support nothing about “the industry”. This is the most common substantive deduction in final chapters, and the last paragraph is where the temptation peaks.

How is the conclusion different from the discussion?

Three terms get tangled here, so take them one at a time on the same finding.

• Results report, without evaluating: “Six of the eight participants named unclear coordination as the main obstacle.”
• The discussion interprets and relates to existing research: “That coordination rather than technology dominates is consistent with Schmidt (2023); the two divergent cases both work largely alone on separable modules.”
• The conclusion draws the consequence and answers the question: “In the cases studied, hybrid arrangements affect self-reported productivity through coordination rather than through technology.”

A practical test on your own draft: if a sentence in your conclusion contains a citation or an explanation of whya finding came out as it did, it is doing the discussion's job. Move it.

Empirical theses usually keep the two chapters separate. Literature-based theses frequently merge them into “Discussion and Conclusion”, with the interpretation first and the answer last. Both are normal — agree it with your supervisor rather than deciding alone, since some departments require the split.

The conclusion is also the matched partner of your first chapter. If they no longer address the same question, the fix goes in the introduction; see how to write a thesis introduction.

How long should the conclusion be?

Five to ten percent of the text is the standard rule of thumb, and it scales with the level:

• Bachelor's thesis (30–50 pages): roughly two to four pages.
• Master's thesis (60–100 pages): four to eight, often with implications treated as their own section.
• Doctoral thesis: a full chapter, typically stating the contribution to knowledge explicitly and revisiting each research question in turn.

Departmental guidance overrides all of this. A conclusion that is too short usually fails to answer the question completely; one that runs long almost always contains material that belongs in the discussion — that is the usual reason a conclusion swells to eight pages.

A worked example, paragraph by paragraph

Same study as the introduction example: hybrid working and self-reported productivity in small and medium-sized firms, eight semi-structured interviews.

Question and answer. “This thesis asked how hybrid working arrangements affect the self-reported productivity of developers in small and medium-sized firms. In the cases studied, the effect runs through coordination rather than through location: participants attributed gains to uninterrupted individual work and losses to the additional effort of aligning with others.” — One sentence of reference, then the answer. No suspense.

Supporting findings, condensed. “Three findings support this. Perceived gains clustered in roles with a high share of separable work; perceived losses clustered where tasks required frequent negotiation with colleagues; and no participant attributed a change in productivity to technical equipment, although all four firms had invested in it during the same period.” — Three findings in one sentence each, all pointed at the answer.

Limitations. “These conclusions rest on eight interviews in four firms and support analytical rather than statistical generalisation. All participants had worked in the same team for at least three years, which plausibly explains the absence of the reduced team belonging reported elsewhere, and which limits transfer to newly formed teams. Productivity was self-reported throughout; no objective measure was available.” — Three real limits, each with its consequence attached.

Outlook. “Whether the coordination mechanism identified here also holds for newly formed teams, and whether self-reported gains correspond to measurable output, are the two questions this design cannot address and that a larger mixed-methods study could.” — Specific, and derived from the limitations rather than invented.

Practical implication. “For firms setting attendance policy, this suggests that fixed office days are more usefully tied to the coordination intensity of particular work than applied as a uniform quota.” — One sentence, concrete, and clearly within what the data supports.

Writing the answer sentence

The answer sentence is the hardest sentence in the thesis, and it is hard for a good reason: it is the only place where you cannot hide behind description.

Three patterns that work, at different strengths:

• “The study shows that …” — for direct, unambiguous findings only.
• “In the cases examined, X affects Y through …”— states the scope inside the claim, which is usually the right level for a bachelor's or master's thesis.
• “The findings indicate that …, although …” — for a qualified answer, where the qualification is substantive rather than nervous.

Note the difference between qualifying and hedging. “In the four firms studied” is a qualification: it is information about scope. “It could possibly be argued that there may be a certain tendency” is hedging: it removes the claim. One qualifier per sentence, chosen to match your evidence, and delete the rest.

If the honest answer is that your data does not settle the question, say so — but say it as a finding: “The data does not support a general effect; instead, the direction varies with task structure.” That is an answer. Vagueness is not.

Limitations and outlook

Two failure modes appear about equally often: no limitations at all, or a regretful paragraph that talks the work down. Both read as an absence of methodological judgement.

• Weak: “Unfortunately the sample was very small, which is a serious problem and severely limits the value of the results.”
• Strong: “A sample of eight supports analytical rather than statistical generalisation: the findings identify patterns that warrant testing at scale rather than establishing how widespread they are.”

Same fact; the second shows you know what your data can carry, which is the competence being assessed. Name only genuine limitations — “more time would have helped” is not one — and attach a consequence to each, which turns self-criticism into precision.

The outlook then writes itself, because good future research questions come straight out of the limitations. If the outlook and the limitations have nothing to do with each other, one of them is invented.

Mistakes that cost marks

• The research question is not actually answered. By far the most common. Test: delete everything except your answer sentence — is there an answer left?
• The conclusion no longer matches the introduction, because the thesis moved. Revise the introduction; do not bend the answer.
• Chapter-by-chapter recap. If it reads like a walk through the table of contents, it is a summary and the conclusion is still missing.
• No limitations. Read as an absence of methodological reflection, every time.
• The outlook is longer than the answer. The proportions are wrong; readers register it as a thesis more interested in the study it did not do.
• New literature in the final paragraph. Always noticed.

A last check before you stop. Read your introduction and your conclusion back to back, and nothing else. Does the second answer the question the first asks, and is the answer supported by things the reader was shown in between? Examiners do exactly this, often first, and it is the cheapest quality check available to you.