The 13 research prompts that went viral on X last week. #4 saved my discussion section.

A thread by @ChrisLaubAI compiled the most shared Claude prompts in research communities . I tested them. These two are absolute gold.

1. The Contradictions Finder
"List all internal contradictions, unresolved tensions, or claims that don't fully follow from the evidence in this draft."
It catches the logical jumps your brain autocorrects but a reviewer won't .

2. The Assumption Stress Test
"List every assumption this argument relies on. Now tell me which ones are most fragile and why."
This one hurt. I realized my entire results section assumed my sample was representative. It wasn't .

3. The "What Would Break This?"
"Describe realistic failure modes for this approach. Not edge cases."
Great for forecasting sections .

Caveat: Don't paste proprietary data into public LLMs. Use institutional licenses or local models. Protect your IP.

Paolina, #2 hit me right in the impostor syndrome. "My entire results section assumed my sample was representative. It wasn't." I felt that in my SOUL. I defended my thesis last year and during my defense, one committee member gently asked about my sampling strategy and I realized I'd just... assumed. Luckily I had data to partially address it, but the panic was real.

These prompts are basically what good advisors do—ask the hard questions you're avoiding. The "Contradictions Finder" is especially brutal because we all have those moments where our brain just... fills in the gaps. The prompt doesn't let you do that.

Also, the IP caveat is SO important. I know people who've pasted entire unpublished manuscripts into public ChatGPT and then panicked about data privacy later. Most universities now have secure research computing options with LLM access. Worth checking before you upload your next draft.

Thanks for sharing these. Definitely saving for my next paper.