Yellow Zone, to use with extreme caution These uses are not forbidden but carry significant risks. AI output here should be treated as a rough draft at best, never as final product.

1. “Write My Introduction”

The risk: AI can fabricate references, cite papers that don’t exist, miss key literature in your specific niche, and misrepresent the state of the field.

Safe practice:

I am writing the introduction to a paper on [specific narrow topic]. Suggest a structural outline: what topics should I cover, in what order, and what logical flow connects them? Do not write any content — only suggest the structure.

Then you write the introduction yourself using real, verified references.

Elsevier: “AI output can only serve as inspiration in the manuscript preparation process, and any manuscript needs to represent the authors’ authentic and original contribution.”

---

2. Interpreting Your Own Results

The risk: AI does not understand your data. It cannot see your raw numbers, experimental conditions, or quality controls. It will confidently generate plausible-sounding interpretations that may be completely wrong.

What happened in practice: A study found that AI-generated summaries of scientific papers were only 80-90% accurate. That means 1 in 5 to 1 in 10 statements could be factually wrong.

Safe practice: Write your own interpretation. If you want AI input, ask it to suggest alternative interpretations to consider — but evaluate every one against your actual data.

---

3. Generating Conclusions

The risk: Conclusions require scientific judgment, contextual knowledge of your field, and accountability for the claims you make. AI has none of these.

Safe practice: Ask AI to review your draft conclusions and point out:

• Logical leaps not supported by your data
• Overstated claims
• Missing caveats or limitations

This turns AI into a critical reviewer rather than an author.

---

4. Statistical Reporting

The risk: AI frequently mishandles statistics — reporting wrong test names, inventing p-values, confusing statistical significance with effect size.

Safe practice: Never ask AI to interpret statistics or choose statistical tests. You may ask it to format your correctly-calculated statistical results into a standard reporting format:

Format these statistical results into a sentence following APA style: t(24) = 3.42, p = 0.002, Cohen’s d = 0.68

---