Using AI in Academic Research Responsibly: Accuracy, Citations, Originality, and Academic Integrity

AI tools are now part of how many students draft, summarise and revise. The question is no longer whether they will be used, but whether the work they produce survives contact with the standards your department applies.

The distinction that matters is simple: AI can help with the process of research. It cannot be the source of your evidence. A claim in your dissertation must be traceable to a source you have verified, or to data you collected.

What AI is genuinely useful for

• Explaining a concept you have already found in a source, so you can read the source better.
• Reworking your own sentences for clarity while the meaning stays yours.
• Generating questions you might ask about a document you have read.
• Checking internal consistency: does your objective two appear in your findings?
• Formatting a reference entry from metadata you have supplied.
• Summarising a document you provide, with the understanding that summaries are lossy and must be checked against the original before you rely on them.

What AI cannot do for you

It cannot tell you what a source says if it has not read it. It cannot be relied on to confirm that a reference exists. It cannot know your department's rules. It cannot decide what your findings mean.

When a tool answers a request for a citation, it may produce a plausible author, a plausible journal and a plausible year that do not correspond to any real article. This is not a rare malfunction; it is a predictable consequence of how these systems generate text. A fabricated citation in submitted work is a serious integrity problem, and it is the single fastest way to turn useful assistance into a disciplinary matter.

Verification is the whole discipline

Adopt one rule and apply it without exception: no citation enters your document until you have seen the source.

That means:

1. Open the source itself — the article, the book, the report.
2. Confirm the author, year, title and container match what you are about to write.
3. Read enough of the source to confirm it supports the claim you are attaching to it.
4. Record the DOI or stable identifier where one exists.
5. Note whether you read the full text, or only an abstract, and do not attribute detail that was not in what you actually read.

If you cannot open the source, do not cite it. Find something you can open, or describe the state of the evidence honestly.

Originality and the line between assistance and authorship

Originality in academic work means that the argument, the structure and the wording are yours, and that the ideas of others are attributed. Using a tool to improve your sentence does not breach that; submitting text you did not write, without understanding or disclosure, does.

Three practical tests:

• The explanation test. Could you explain every sentence to your supervisor, without the tool open?
• The defence test. If an examiner asked why you structured the argument that way, could you answer?
• The removal test. If you removed the tool's contribution, would a defensible piece of work remain?

If a passage fails any of these, it should not be in your submission in that form.

Disclosure and institutional policy

Requirements differ. Some departments permit AI assistance for language editing only. Others require a declaration describing which tools were used and for what. Others still prohibit generative tools for assessed written work entirely.

You are responsible for knowing which applies to you. Concretely:

1. Find your institution's academic integrity policy and any programme-specific guidance.
2. Check whether the unit you are submitting to has its own rules.
3. Ask your supervisor in writing if the guidance is ambiguous, and keep the reply.
4. Follow the stricter rule when two documents conflict.
5. Declare what you used, in the form your institution requires.

Do not assume that a policy you read online from another university applies to you, and do not assume silence means permission.

Privacy and data protection

Treat anything you put into a tool as leaving your control. That has consequences:

• Do not upload confidential institutional documents, unpublished datasets or identifiable participant information unless you have explicit permission and a lawful basis.
• Do not paste interview transcripts containing personal details.
• Do not enter a participant's name together with sensitive information.
• Check whether your institution has an approved tool for research work, and use it if one exists.

If your ethics approval says data will be stored in a particular way, the tool you use is part of how that promise is kept or broken.

Plagiarism risk in both directions

There are two failure modes. The first is submitting generated text as your own. The second is subtler: the tool reproduces phrasing from sources without attribution, and you carry that phrasing into your document believing it is your own.

Guard against the second by writing your notes in your own words and your own sentence structure, and by checking any passage that reads more polished than the rest of your draft. Where you have relied on an author's exact formulation, quote it and cite it.

A workable workflow

1. Do your own reading first, with notes in your own words.
2. Use AI for process tasks: outline checking, consistency checks, clarity revisions of your own text.
3. Verify every factual claim and every citation against the source.
4. Keep a record of which tools you used, for what, and when.
5. Declare in the form your institution requires.
6. Read the final document as your own work, because it is.

PassResearch is built around this position. The corrections workspace and citation tools are designed to keep every claim connected to a source you have actually captured, and the platform's own editorial policy is that generated material is not treated as evidence. See our integrity position for how the platform describes its own use of automation.

Key takeaways

• AI assists with process; sources provide evidence.
• Nothing is cited until the source has been opened and checked.
• Originality means you can explain, defend and account for the work.
• Disclosure requirements vary, and the stricter rule wins.
• Confidential data does not go into a tool without explicit permission.
• You are accountable for the final document, whatever helped produce it.

If you are uncertain about a specific case, the safest action is to ask your supervisor before submitting, and to keep the answer.