How to Use ChatGPT for Essay Writing Without Hallucinating Sources

Every student has done this. You open ChatGPT, type "write me an essay about the ethics of obedience in psychology," and get back something that sounds confident, has a coherent argument, and cites three studies you've never heard of.

You check them. One doesn't exist. One is misattributed. One is real but says the opposite of what ChatGPT claimed.

This isn't a bug. It's how language models work. And there's a straightforward fix — but almost no one does it.

Why ChatGPT Makes Up Citations

ChatGPT is a language model, not a search engine or a database. When you ask it to write about a topic with no sources provided, it generates text that sounds like an academic essay — including citations that sound like real papers. It's predicting what text should look like, not retrieving actual research.

The model has been trained on a vast amount of academic text, so it knows the pattern: claim → citation → supporting detail. It fills that pattern convincingly. But "convincing" and "accurate" are different things.

This isn't a failure of the AI. It's a failure of how the question was asked.

The Fix: Give It Your Sources Before You Ask

The simplest thing you can do is give ChatGPT your actual research before asking it to write anything.

When you paste in your sources — titles, URLs, your own notes on why each one matters — the AI has real material to work with. It stops filling the citation pattern with invented references and starts using the ones you provided.

The output is different in a real way:

• Arguments are grounded in sources you've actually read
• Every citation links to something you can verify
• The framing matches your own notes on what each source says
• Counterarguments are drawn from your research, not generated from training data

This isn't about tricking the AI. It's about using it the way it works best: as a drafting and reasoning tool, not as a research tool.

How to Do It, Step by Step

Step 1: Save sources as you find them

As you research your essay, save every useful source — articles, papers, videos, textbook pages — with a short note explaining why it matters and which argument it supports. Don't rely on browser history or open tabs.

Step 2: Organise by paper or argument

Group your sources. If you're writing about obedience in psychology, you might have one group for experimental evidence, one for ethical critiques, one for real-world applications. This structure becomes useful when you prompt the AI.

Step 3: Copy your research context

Before opening ChatGPT, copy all your saved sources as plain text. Each entry should include the source name, the URL if you have it, and your note. The whole thing might be 500–1,000 words — well within what any AI can process.

Step 4: Paste context first, then prompt

Open ChatGPT (or Claude, or Gemini — the same approach works everywhere). Paste your research context at the top. Then add your prompt beneath it.

Step 5: Ask for something specific

Don't ask for a finished essay. Ask for one specific thing at a time.

Five Prompts That Work Once You've Pasted Your Sources

Outline your argument:

"Using only the sources in my research context above, outline a 5-paragraph argument for [your thesis]. Cite each source by name."

Draft the introduction:

"Write an introduction paragraph for my essay on [topic] using the research context above. Under 150 words. End with my thesis."

Find what's missing:

"Review the sources I've provided. What important perspectives or counterarguments are absent? What should I find before I write?"

Generate a study guide:

"Turn the sources above into a concise study guide for [topic]. Key terms, main claims, one sentence per concept."

Check your thesis:

"My thesis is [X]. Based on the sources above, is this well-supported? What's the strongest objection a reader might raise?"

What the Output Actually Looks Like

Without sources, a prompt like "write about the ethics of Milgram's obedience experiments" produces something like:

Milgram's landmark 1963 study on obedience found that 65% of participants were willing to administer what they believed to be dangerous electric shocks (Milgram, 1963; Blass, 2004; Perry, 2012).

Two of those citations may be accurate. One might be invented or misattributed. You don't know without checking.

With your sources pasted first, the same prompt produces:

Based on your note that the Milgram (1963) study showed 65% compliance and your source from the APA ethics guidelines linked above, there's a clear tension between the study's scientific value and the deception involved in the methodology.

It's using what you actually found. The citation is real because you gave it a real source.

The Habit That Makes the Difference

The students who get the most out of AI for writing aren't the ones who ask it to write their essays. They're the ones who treat it as a collaborator that needs to be briefed — who bring their actual research into the conversation and use the AI to help them think and draft, not to generate from nothing.

That habit starts with where you save your sources. If they're scattered across tabs, bookmarks, and Google Docs, copying them in is too much friction and it won't happen consistently.

If they're organised in one place — already structured by argument or class — it takes one click and thirty seconds. Then the AI actually has something useful to work with.

---

Vaulterly is a free research vault for students. Save your sources, add notes, organise by class or paper, and copy your whole vault as context for any AI tool — in one click.