Chapter 5: Thesis

What does Chapter 5 have to do?

Three things: present your results in the context of the prior literature, make recommendations for future research, and state your own conclusions about what your results mean.

This is the first chapter where your own thinking is the point. Everything before it reports — in the literature review you report what others found, in the results chapter what you found. Here you argue.

It is also where institutional review looks hardest, because the most common structural failure in a dissertation is a conclusion that claims more than the results support.

Chapter 5: Conclusion

Finally, you are here, one step from the promised land and in a position to air your own thinking on what your results mean. This is the step in which you paint on the canvas you have created. The gold frame was your review of literature, the canvas was your results, and painting on the canvas is done by discussing your results in the context of the previous literature and providing your conclusions. However, as with all chapters there are certain 'bells you must ring'…

Chapter introduction

Restate the problem, the research questions, your methodology and theoretical framework briefly, and recapitulate your findings.

All of this is only to refresh the reader and set the context for what follows. Nobody reads Chapter 5 having just finished Chapter 4 — your committee reads it weeks later, and your external reviewer may read it first.

Having refreshed the reader's understanding of all that has gone before, you move on to the beating heart of the chapter.

Summary of your results

A concise summary of your results, without interpretation.

This is a recapitulation, not a re-analysis. The interpretation comes next, and keeping the two apart is what makes the next section persuasive.

Discussion of your results

Here you discuss what you think your results mean.

At this point you are discussing your own results, not those of other scholars in relation to yours.

You address how well your results turned out, how they fit your hypothesis or research questions, and whether they answered what you set out to ask — and if so, how well.

If your results fail to support all or part of what you expected, explain why, in detail. Remember: whether negative or positive, results are equally important to the field.

In sum, this section explains how the results came to be.

Results based conclusions

Did your research confirm or support previous work? Strongly, or weakly? Does it contradict previous work, and if so, why?

To do this you contextualize your findings against the theoretical framework you used and against previous research. You interpret what you found and tell the reader why you think you found it.

Limitations

Some are resource-based — time, duration, money. Others are inherent in the parameters of a study, a theoretical or analytical approach, natural physical laws, or the technology used.

Here you tell the reader what limitations your study faced. It is also a place to discuss how, in retrospect, the study could have been strengthened, and what its flaws were — avoidable and unavoidable.

All studies have limitations, including hugely expensive longitudinal ones. The limitations you identify contextualize your study's validity, its reach, and its generalizability.

Your research's implications

Who might benefit from your research, and how?

Recommendations for future research

However many ideas you have, keep this to four at most.

Recommendations often stem from limitations that future research might address. They may also arise from your data — or from neither, provided they are directly relevant to your topic and field.

Conclusion

Once again you summarize what has gone before, as the foundation for the conclusions you are about to draw.

Then you tell your reader what you think your results mean.

Do not over-reach, and do not claim more than your study can support. This is still a place for calm, rational, clear writing and argument.

The two things that sink a Chapter 5

Everything above is structure, and structure is the easy part. These two are what institutional review and committees actually catch.

Alignment: do your conclusions follow from your findings?

Among the things School Review checks — a different test from your committee's assessment of the scholarship — is whether your conclusions actually follow from your findings, and whether claims outrun the evidence.

Both are Chapter 5 problems, not Chapter 4 problems. Results that fit what you expected are not the same thing: every conclusion must trace to what you actually found, not to what you hoped to find.

See what School Review actually checks for the full list.

Chapter 5 is often written last and in a hurry — where uneven voice across chapters shows worst. The First Paragraph Test takes ten minutes.

Why you must not ask an AI whether your conclusions hold

Chapter 5 has one characteristic failure: claiming more than the evidence supports.

Your findings were narrower than you hoped, or messier, and the conclusion quietly stretches to cover the gap. Committees are extremely good at spotting this, because it is the thing they are specifically reading for.

It is also the exact failure an AI will not catch — and worse, will actively encourage.

A 2026 study in Science tested eleven leading models and found they affirmed users' positions roughly 50 percent more often than humans did — even where the user's actions involved deception or clear harm. The study examined interpersonal advice rather than research reasoning, but the cause is not domain-specific: these systems are trained on human feedback, and humans consistently rate agreement more highly than accuracy. The same study found people trusted the agreeable responses more, not less.

So the tool many candidates reach for to check whether their conclusions hold is built to tell them yes — and the reassurance raises their confidence in a chapter nobody has actually checked.

Ask an AI does my conclusion follow from my results? and you will almost certainly be told it does. That answer is worth nothing. Not because the model is careless, but because agreement is what it optimizes for.

What to do instead.

Use AI for the mechanical half of the check, and do the judgment half yourself.

• List every claim your Chapter 5 makes. One line each.
• List every finding your Chapter 4 actually reports. One line each.
• Put the two lists side by side and draw the line from each claim back to the evidence that supports it.
• Any claim without a line is an overreach. Cut it, or soften it until it matches what you actually found.

An AI can produce both lists quickly and accurately — that is a bounded extraction task it does well. Drawing the lines, and being honest about which ones do not connect, is yours. More on the distinction on our Human-in-the-Loop Structural Refinement page.

Where quantitative and qualitative differ

Less than most candidates expect. The chapter does the same work either way, and the sections above apply to both.

Three differences are worth knowing.

What a claim has to trace back to

In a quantitative study, a conclusion traces to a number. There is arithmetic behind it, and either the arithmetic supports the claim or it does not.

In a qualitative study, a conclusion traces to coded data — which participants, which passages, and how many.

That makes the qualitative version harder to check and easier to overstate. A theme carried by two of fourteen participants is a finding. Stated as though it were general, it is an overreach, and it is the commonest fault in a qualitative Chapter 5.

Three questions find most of it:

• Which participants support this claim, and how many?
• Could a different analyst working from my codes reach a different conclusion? If yes, say so. That is a limitation, not a weakness — and hiding it is what a reviewer will find.
• Am I claiming transferability I have not established? Qualitative work does not generalize in the quantitative sense, and a conclusion behaving as though it does will be caught.

Why the AI risk is worse in qualitative work

There is no arithmetic to fail.

A model asked whether a p value supports a claim has something to check against. Asked whether a theme is well evidenced, it has only your account of the theme — and it will agree with your account.

These tools are also fluent about interpretation. Ask one whether your reading of a transcript is defensible and you will get a confident, well-written yes, built from nothing but the words you supplied. That reads as validation and is not.

The claim-and-evidence exercise above still works — with one change. Your second list is themes with the number of participants behind each one. That is what makes the lines drawable.

The language you are working in

A quantitative Chapter 5 discusses hypotheses supported or not supported.

A qualitative Chapter 5 discusses research questions answered, partly answered, or reframed by what emerged.

The distinction matters in the writing. A qualitative chapter that talks about hypotheses being confirmed reads as though the candidate has not settled what kind of study they ran — and that is a methodology doubt raised in the last chapter, which is the worst place to raise one.

You write your last sentence and look up. Suddenly … the long journey is over and it is time for final editing and submission of your master work for your school's review. After that? Your dissertation defense, graduation and, at long last, a new life … or at least the grinding burden of writing a dissertation over with!