AI in PhD Research

C1 Academic English · Practical workflows · Integrity boundaries

Where AI fits in PhD work

Used well, AI tools collapse the time cost of routine cognitive work — paraphrasing, summarising, translating, simplifying — and free you to spend more time on the parts of research that actually require you: judgement, originality, defending claims, deciding what matters.

Used badly, AI flattens your voice, hides gaps in your understanding, and produces text you can't defend in a viva. The difference between the two is mostly about how you frame the prompt and how much you leave the AI in control.

The single rule that makes everything else easier

Use AI to think faster, never to think for you. Every output it gives you needs to pass through your own understanding before it becomes part of your work.

Five places AI is genuinely useful in PhD work

  1. Reading — accelerating comprehension of dense theoretical texts, especially when language is the bottleneck rather than the ideas.
  2. Note-taking — paraphrasing, summarising, generating linking questions between sources.
  3. Drafting — overcoming a blank page; generating a first structural skeleton you then heavily rewrite.
  4. Editing — checking register, spotting awkward phrasing, suggesting alternatives.
  5. Stress-testing — asking AI to argue against your position so you can see weaknesses.

Before going further — a check

Find out what your specific institution and supervisor permit. Policies vary widely. Some allow AI for editing but not drafting. Some require disclosure. The workflows in this lesson are conservative, but always check.

AI for reading dense texts

Theoretical papers in your second language are often slow not because the ideas are hard, but because the language is dense. AI can help you separate the two — see the argument first, then come back to the prose.

1

Paraphrase a difficult paragraph

Paste in one paragraph. Ask the AI to rewrite it for clarity. Then read both. The AI version helps you decode the original — it doesn't replace it.

Rewrite the following paragraph in clear, simple academic English at roughly C1 level. Preserve every claim and every nuance — do not simplify the argument, only the language. After the rewrite, list any technical terms in the original that I should know. [paste paragraph]
2

Identify argument structure

For longer sections, ask the AI to map the argument — claim, evidence, counter-argument. Then go back and verify against the original yourself.

Read the following section. Identify: 1. The author's main claim (in one sentence). 2. The two or three pieces of evidence the author offers. 3. Any counter-argument the author addresses. 4. Anything the author seems to assume rather than argue. Quote one short phrase from the text for each item, with a rough location (paragraph 1, paragraph 4, etc). [paste section]

Watch out for

Hallucinated content. AI sometimes fabricates citations, page numbers, and even quoted text. Always verify quotes against the source. Never cite anything from an AI summary without confirming it yourself in the paper.

Question:

What does this kind of AI-assisted reading not teach you that traditional close reading does? Where do you lose something, and is the speed worth it?

AI for note-taking and synthesis

Note-taking is where AI is genuinely transformative — but only if you stay in the loop. The risk is producing a folder of AI-generated summaries that sound right and aren't yours.

1

The "first pass" note

After you've read a paper, ask AI to draft a Cornell-style note. Then rewrite it in your own words. The AI's version is a scaffold, not the final note.

I have just finished reading a paper. Please draft a Cornell-style note with three sections: CUES (4-6 key terms or questions) NOTES (the main argument and evidence in 6-10 bullet points) SUMMARY (one paragraph, 3-5 sentences, capturing why this paper matters) Use simple academic English. Do not add anything that is not in the text I am about to paste. [paste paper or detailed summary]
2

The synthesis prompt

Once you have notes on several papers, AI is excellent at finding connections you missed.

I am going to paste notes from three different sources. Please: 1. Identify any claims that appear in more than one source. 2. Identify any direct disagreements between the sources. 3. Suggest two or three questions that arise from holding these sources together that none of them addresses individually. Treat this as a thinking exercise — I want to push the synthesis, not just summarise. [paste 3 notes]
3

Atomic notes for Zettelkasten

If you're building a Zettelkasten, ask AI to help you split a long summary into atomic, single-idea notes.

I have a long summary that contains several distinct ideas. Please split it into 3-5 atomic notes, where each note contains exactly one idea. Each atomic note should have: - A short, claim-style title - One short paragraph stating the idea in clear academic English - A list of the other notes it should link to (use placeholders like [author + idea]) [paste summary]

The translation rule

Whatever the AI gives you, translate it back into your own voice before saving. If you can't paraphrase the AI's version without losing the meaning, you don't yet own the idea.

AI for writing and refining

This is the highest-stakes area. AI-generated paragraphs are easy to spot, easy to challenge, and risk integrity violations. The goal here is not to use AI to write, but to use AI to help you see your own writing more clearly.

1

The register check

After you've written a paragraph yourself, ask AI to comment on register and clarity — without rewriting it for you.

I have written this paragraph for a PhD-level academic paper. Please give me feedback on: 1. Register — is anything too informal, too journalistic, or too dense? 2. Hedging — are claims appropriately qualified, or too flat? 3. Clarity — are there sentences that are difficult to parse? 4. Cohesion — does it flow as a paragraph? Do NOT rewrite the paragraph. Just give me the feedback so I can revise it myself. [paste paragraph]
2

The "stress-test" prompt

Ask AI to argue against your claim. Use the result to anticipate objections in the viva.

I am defending the following claim in my thesis: [paste claim] Please act as a sceptical examiner. Give me: 1. The three strongest counter-arguments to my claim. 2. The two empirical or theoretical weaknesses my position has that I might not have noticed. 3. The kind of question a viva examiner would ask to test whether I have thought this through. Be tough. I want to be challenged, not flattered.
3

The unblock prompt — for when you're stuck

When you can't find a way into a paragraph, ask AI to generate three structurally different opening sentences. Pick none of them — but use them to see what shape the paragraph could take.

I am trying to write a paragraph about [topic] that needs to make the following point: [point]. Please give me three structurally different opening sentences I could use. Do not write the paragraph — just the openings. I will write the paragraph myself once I see the options.

✅ Use AI to

  • Get feedback on your own writing
  • Stress-test claims you've already made
  • Generate options you'll choose between
  • Check grammar, register, hedging
  • Spot logical gaps

❌ Don't use AI to

  • Write a draft you'll lightly edit
  • Generate paragraphs you'll paste into the thesis
  • Produce arguments you can't defend yourself
  • Skip the actual reading
  • Hide gaps in understanding

Integrity, limits, and what you can defend in a viva

The viva is the test that should shape every AI decision. Anything you can't defend live, in a room, under questioning — should not be in your thesis.

The viva test

For every paragraph, ask:

  • Could I explain why I made this claim, in my own words, without notes?
  • Could I name the sources behind it and summarise them?
  • Could I respond to a follow-up question that pushes against the claim?
  • Could I describe how I arrived at this paragraph?

If the answer to all four is yes, the paragraph is yours, regardless of what tools helped you write it. If the answer to any is no, the paragraph isn't ready.

Six concrete integrity rules

  1. Never paste AI output verbatim into your thesis. Even short phrases. Translate everything into your own voice.
  2. Never cite anything an AI told you without finding and reading the actual source. AI hallucinates citations routinely.
  3. Disclose AI use as required. Most institutions now have a policy. Find yours, follow it, document what you used.
  4. Don't share your unpublished work with public AI tools if your university has confidentiality policies. Use institutional or local tools where possible.
  5. Keep a record of your prompts. If asked, you should be able to show what you used AI for.
  6. If in doubt, ask your supervisor in writing. Get the answer on email so you have it.

The honest test

If you removed AI from your workflow tomorrow, could you still produce a thesis? If yes, you're using it well. If the answer is no, you've outsourced part of your thinking — and that's the part the viva will find.

Reflection:

Where in your current research process is AI doing work you should still be doing yourself? Where is it doing work that genuinely speeds you up without weakening the thesis? Be honest — the viva will be.

Notes