0/10 sections
Wrap-up
Key Takeaways
Pull the threads together: where each tool wins, where it warns, and what to actually do in the week after the workshop.
SECTION OBJECTIVES
- Pick one tool and one configuration surface to invest in this term.
- Agree an AI-use disclosure stance for your next paper.
- Leave with a 10-item action checklist.
You've now seen the three tools through the same lens: persistent roles, grounded sources, document-aware drafting, and reusable specialists. The comparison below is a starting point, not a verdict. Pick on the basis of where your sources, your collaborators, and your institution already are.
Gemini vs ChatGPT vs Claude. For research
| Dimension | Gemini | ChatGPT | Claude |
|---|---|---|---|
| Best at | Open-web deep research, multimodal input, source-grounded reading packs (via NotebookLM) | Configurable specialist agents (Custom GPTs), live folder/Drive workflows, broad ecosystem | Long-context drafting, methods/critique work, careful citation discipline, team workflows |
| Persistent context surface | Gems + NotebookLM notebooks | Projects + Custom GPTs + Custom Instructions | Projects + Skills |
| Source-grounded mode | NotebookLM (strongest of the three) | Projects + Connectors | Projects (large knowledge base, citations) |
| Team / co-author support | Workspace sharing of Gems and notebooks | Shared GPTs and team workspaces | Claude CoWork shared Projects (most explicit) |
| Drafting surface | Canvas | Canvas | Artifacts + Projects |
| Watch-out | Citation accuracy outside Deep Research can drift | Memory across Projects can leak context; audit periodically | Smaller plugin/connector ecosystem than ChatGPT |
Ethics & integrity checklist
- 1Disclose AI assistance in line with your institution's and target journal's policies. And check both before submission.
- 2Never paste student work, unpublished peer-review material, or interview transcripts into a tool whose data-handling terms you haven't read. Or without turning off data usage for training.
- 3Treat every citation an LLM produces as unverified until you have opened the source yourself.
- 4Keep a brief AI-use log per paper (tool, purpose, date). It makes the methods paragraph and any reviewer query trivial to answer.
- 5Discuss tool norms with co-authors and supervisees before, not after, the first draft.
Official data-handling and privacy policies
10 take-home actions
- 1Build one Gem, one Custom GPT, or one Claude Skill this week for a research task you do at least monthly.
- 2Convert one currently-painful folder of PDFs into a NotebookLM notebook or a Claude Project.
- 3Write your ChatGPT Custom Instructions (or the equivalent for your chosen tool). Under 1500 characters.
- 4Pick one paper-in-progress and create a Project for it; pin the charter as message #1.
- 5Run one Deep Research report on a research question you're currently exploring; treat it as a starting brief, not a finished one.
- 6Draft a 1-paragraph AI-use disclosure you'd be comfortable including in a methods section.
- 7Agree shared AI norms with your most frequent co-author or RA.
- 8Trial a source-grounded tool against an open-web tool on the same question; note where each wins.
- 9Use AI to pre-flight one draft before submission: check methods-description completeness, citation consistency, and argument flow.
- 10Subscribe to one signal source for AI-in-research developments (a newsletter, a journal special issue, a colleague). Refresh quarterly.
Further reading
- NotebookLM and Gemini documentation. Google's own guidance is updated frequently and underrated.
- Anthropic's prompt-engineering and Skills documentation. The clearest writing on context engineering in the industry.
- OpenAI's Custom GPT and Projects help centre. Read the data-handling sections, not just the how-to.
- Your institution's research-integrity office. Most have issued AI-use guidance in the last 18 months.
- Recent journal editorials in your sub-field on AI authorship and disclosure norms.