Starting Your Dissertation This Year? Here's How to Stop AI From Inventing Your References

The problem isn't laziness. It's how general AI models work

Ask a general-purpose AI chatbot for sources on a topic, and it will often give you a beautifully formatted reference (author, year, journal, volume, page numbers) that doesn't exist. This isn't a rare glitch. It's a predictable consequence of how these models generate text: they produce what a plausible citation looks like, not what's actually been published. Universities have been dealing with this for long enough now that "AI-fabricated references" is its own recognised category of academic misconduct case, distinct from plagiarism.

The frustrating part is that these fake references are often the most convincing-looking ones in a reference list: clean formatting, a real-sounding journal name, an author who does actually publish in the field. A supervisor or examiner has to go and check every single one to catch it.

What this means practically, this term

If you're starting a lit review, a dissertation, or even a long essay this year, the safest working assumption is: any reference an AI chatbot gives you needs to be independently verified before it goes anywhere near your reference list. Not spot-checked: verified, every time.

That's a tedious ask when you're also trying to write. Which is why the more useful fix isn't "be more careful with ChatGPT". It's using a tool that structurally can't do this in the first place.

The tool matters more than the caution.

Answers that can only come from real papers

This is the specific problem MyOwnArc is built to solve. Instead of generating answers from its general training, it answers only from the papers actually sitting in your library: PDFs you've uploaded, or papers you've imported from PubMed, OpenAlex, Semantic Scholar, or Crossref. Every claim it gives you carries a citation that opens the exact passage in the source PDF, so you're not trusting the AI; you're checking the source in two clicks, the way a supervisor or examiner would.

Practically, for a student starting research this term, that looks like:

• Building a library as you read, rather than a folder of PDFs you'll "organise later"
• Asking questions scoped to one paper, a folder, or your whole library, and getting an answer with a numbered citation to the actual passage
• Highlighting as you read, then writing notes that cite your library with a simple @ mention, building your APA 7 (or Harvard, Vancouver, IEEE) reference list automatically as you go, instead of assembling it from memory the night before submission
• Running a literature discovery search across PubMed, OpenAlex, Semantic Scholar, and Crossref together, rather than four separate searches you have to de-duplicate by hand
• Working outwards from what your papers already cite. MyOwnArc pools every reference in your library into one list, ranked by how many of your papers cite it and flagged when you already hold a copy, so the paper everyone keeps citing and you haven't read yet is the obvious next one to get

Not a reason to avoid AI in your research, but a reason to be specific about which one

None of this means AI has no place in a dissertation. It means the tool matters more than the caution. A student plan is $5/month with a verified academic email: a fraction of what a reference manager, a PDF reader, an AI research assistant, and a notes app would cost separately, and considerably less than the cost of an examiner flagging a reference that turns out not to exist.

If you're starting serious research this term, it's worth deciding now, before the first draft of your lit review, whether your AI tool can actually show its working, or whether you're planning to check every reference by hand later, under deadline pressure, when it's much harder to catch.