Ways I use AI in research

Here are some ways I’ve found AI to be useful in research.

• Catching typos, redundancies, missing information in complete drafts

• Preempting reviewer comments (likely more and more useful as reviewers use AI on their end)

• Qualitative classification of text data (with human reliability sample)

• Improving formatting and theme of ggplot code (but not the initial points, lines, etc)

• Supplementing literature search - but not reading/summarizing. I like pasting in my intro with ‘CITE OTHERS’ next to my current citations and sending a deep research agent to find articles. It makes a lot of errors, but also finds articles I had not found with search engines.

• Additional bug-check of analysis code. often the ‘bugs’ found are not relevant (often only relevant in the case of code reuse for different datasets), but occasionally it finds gold.

• Creating looping harnesses for existing analysis, e.g., if I have a known functioning analysis and want to apply to additional datasets, I’ll generate (and verify) the code to iterate through, and then paste and integrate my analysis.

• Finding words on the tip of my tongue. I often (since childhood, not new since AI) struggle to find a word I’m thinking of.

I use mostly Claude Opus 4.6 currently, alongside GPT 5.5 Pro. For text analysis, I have used Sonnet to reduce cost.

The underlying principle is that I don’t trust it much, and I usually prefer my own choices even if the AI does something better. So, I never use it for first passes, only when the output is verifiable, and often as a second set of eyes instead of as a collaborator.

Today’s models still make mistakes, but they definitely are useful. As they keep getting better, I imagine my use cases will continue to expand, but I imagine I will still prefer my own voice and choices over the model’s.