Pretending AI isn't here is denial; using it without thinking is dishonesty. The honest position sits in the middle, and it's learnable.
The tools change every few months. The question doesn't:
Whose thinking is on the page?
If the thinking is yours, with AI as an assistant, the essay is yours. If it's the model's, with you as the typist, it isn't, and your reader will figure that out.
Three categories sort almost every use. Knowing which box you're in, before you open the model, is the whole discipline.
| Category | What it looks like | What to do |
|---|---|---|
| Permitted | Brainstorming. Grammar checks. Critique of a paragraph you wrote. Explaining a concept from your sources. | Use freely. Declare it. |
| Conditional | Generating an outline. Summarising a source. Sample sentences in a discipline style. | Verify every claim. Declare each use. |
| Forbidden | Generating paragraphs you submit unchanged. Inventing citations. Writing the thesis or the analysis. | Don't. This is misconduct. |
Never the reverse. Ask the model first and you've outsourced the thinking. Write first and the model is your editor, a role it does well.
Where is your personal line between using AI and outsourcing the thinking?
Machine-drafted prose has a signature. Once you can hear it, you can hear it in your own draft, and increasingly, so can the software.
The author looks right. The journal exists. The DOI won't resolve. Treat every model-suggested source as a lead, not a citation, open the database and verify it yourself, or it's fabrication.
End every essay with a one-paragraph AI Statement under the references. It isn't a confession, it's a transparency note.
ChatGPT was used for three purposes. (1) Brainstorming counter-arguments before I drafted that paragraph. (2) Flagging unclear sentences, which I then rewrote myself. (3) Explaining a difficult passage in Jin (2022), verified against the original. No sentences were generated by the model; all citations were checked against the sources.
It doesn't excuse improper use, and it does become part of the record, so an honest statement always beats a discovered understatement.
Sort seven real uses into the three boxes, then test whether the sources a model hands you actually exist.
Permitted, conditional, or forbidden?
| Scenario | Perm. | Cond. | Forbid. |
|---|---|---|---|
| A stronger verb for a sentence I wrote. | ☐ | ☐ | ☐ |
| Three objections to my draft thesis. | ☐ | ☐ | ☐ |
| Write the introduction in my voice. | ☐ | ☐ | ☐ |
| Summarise a long PDF before I read it. | ☐ | ☐ | ☐ |
| Five citations to support my thesis. | ☐ | ☐ | ☐ |
Ask a model for three 2022+ sources on your topic and click each DOI. In most class tests, one in three is fabricated.
Chapter 20, pages 87 to 90, including the three-paragraph A/B/C comparison, the prompt banks, the six tells, and Exercises 20.1 to 20.3 (including a real source-verification test).
Chapter 20 · Next, from page to stage