AI for researchers: what Claude and ChatGPT can do, and what to add

Claude and ChatGPT both help with literature reviews and analysis, if you pick your sources and check every citation.

Picks: 8; ranked

Claude is the pick if you analyse data: its Claude Science app, in beta on Pro and higher plans, runs analyses on your own machines and ties every figure to the code behind it. ChatGPT is the pick for a sourced report, because deep research lets you choose the sources and approve the plan first. Add connectors for arXiv, Zotero and web search to reach papers and your own library. Before you start, know that Anthropic processes the prompts and answers you send, even though raw datasets stay on machines you choose.

Good for

Academics and graduate students doing literature reviews
Lab scientists analysing genomics, single-cell or chemistry data
Industry analysts writing sourced market or technical reports
Anyone who keeps a Zotero library and wants to search it by asking

Claude

Claude Science runs analyses on your own hardware and records how each result was made.

Plan: Pro ($17 a month billed annually, $20 monthly, as of November 2026) or higher to get Claude Science. Academic and nonprofit scientists can get the discounted Claude Team plan for scientists.

What it can do

Run analyses in the Claude Science app with Python and R kernels that keep your variables in memory
Read papers and query 60+ scientific databases without learning each one
Get figures, tables and notebooks that carry the exact code, environment and conversation that made them
Have a background reviewer flag incorrect citations, untraceable numbers and figures that do not match their code
Run jobs on your laptop, a Slurm cluster over SSH, or your Modal account
Use built-in specialists for genomics, single-cell, proteomics, structural biology and cheminformatics
Save a pipeline as a skill, or connect your lab tools with a connector

Limits

Claude Science is in beta. It works on Pro, Max, Ultra, Team and Enterprise plans.
On Team and Enterprise, an admin must turn it on first. It is on by default on the Team plan for scientists.
Prompts and answers are processed by Anthropic under standard retention. Raw datasets stay on the machines you choose.
Anthropic lists model training as opt-out on Free, Pro, Max and Ultra, so check your privacy settings before you upload unpublished work.
The plan for scientists needs the principal investigator to verify eligibility. For-profit and industry R&D teams use the regular Team or Enterprise plans.

Plugs to turn on

tooluniverse: Gives Claude 2,700+ scientific tools across life science and literature search.
Zotero MCP: Search, read and annotate your Zotero library by asking.
arXiv MCP: Search arXiv, pull full text and follow citation graphs.

ChatGPT

ChatGPT deep research builds sourced reports from the URLs and files you choose.

Plan: Plus for expanded deep research, or Pro for the maximum amount. Free includes limited deep research.

What it can do

Run deep research: pick the URLs and trusted sources, add private files, then review the plan before it starts
Interrupt and redirect a deep research run while it works, and follow how it weighs sources
Download the finished report as a PDF or DOCX, or keep refining it in chat
Start from shared science chats such as brainstorming research questions, checking manuscript references and finding where a paper deposited its data
Look at the Rosalind Academic Plan if you do life sciences research (OpenAI points to it on its pricing page)

Limits

Free ChatGPT has limited deep research. Heavier use needs Plus or Pro.
Business plans do not train on your data by default. On Free, Go, Plus and Pro, chats can be used for training unless you opt out.
Deep research prioritises the sources you add, so a thin source list gives a thin report.

Plugs to turn on

Exa MCP Server: Official web search that signs in with OAuth. Works in ChatGPT and Claude.
Notion MCP: Keep reading notes in Notion and let ChatGPT search them.

Plugs worth turning on

Start with Zotero if your reading lives there, and arXiv or Paper Search for papers. Add a web search server for everything else.

Run a literature review with Claude Science

Claude

1
Get the app and a plan
Download Claude Science from the Claude Science page. You need Pro, Max, Ultra, Team or Enterprise. On Team and Enterprise, ask your admin to turn it on.
2
Connect your library
Add a Zotero or arXiv connector from the plug pages so Claude can search papers you already trust.
Uses Zotero MCP, arXiv MCP

Ask for a structured review

Ask for claims, methods and sample sizes in a table, with a link to each source.

Prompt

Find 15 papers since 2021 on transformer-based protein structure prediction. For each, give the title, year, method, dataset and one limitation. Link every paper, and mark any claim you could not trace to a source.

4
Check what the reviewer flagged
Read the citation and number flags from the background reviewer. Open each paper you will cite and confirm the claim yourself.

Build a sourced report with ChatGPT deep research

ChatGPT

1
Start deep research
Click the plus button in the message box and choose Deep research.
2
Add your sources
Add the URLs you trust, connect sources you have access to, and attach private files. Deep research prioritises these over the open web.

Approve the plan

Describe the question, then read the proposed plan. Narrow the scope or change sources before it starts.

Prompt

Compare the evidence for intermittent fasting and time-restricted eating on HbA1c in adults with type 2 diabetes. Use only the sources I attached plus peer-reviewed trials. Give the sample size and follow-up for each study.

4
Export and verify
Download the report as a PDF or DOCX. Open every source it cites before the report goes into your paper.

Check a manuscript's references

1
Upload the draft
Attach the manuscript to Claude or ChatGPT. Remove anything under embargo or covered by a data agreement first.

Ask for a reference audit

Ask for duplicates, formatting errors and claims that do not match the cited paper.

Prompt

Check every reference in this manuscript. List duplicates, entries missing a DOI or year, and any place where the text claims something the cited paper does not say. Do not guess: say "could not verify" when unsure.

3
Verify the flagged ones
Search the flagged papers with the arXiv or Paper Search connector and read the passage yourself.
Uses Paper Search, arXiv MCP

Verdict

Our advice: use Claude Pro or higher if you analyse data, ChatGPT Plus or Pro if you mostly need sourced reports. Add Zotero and arXiv or Paper Search either way, and open every source before you cite it.

Our pick: Zotero MCP

Our advice: match the assistant to the job, connect your own library first, and treat every generated citation as unverified until you open it.

Best for

Analysing genomics, single-cell or chemistry data with traceable code
Literature reviews from a library you already keep
Reference checks on a draft

Skip it if

Your data agreement forbids sending prompts to a third party
You would cite a paper without opening it

Questions

What is the best AI for researchers?

It depends on the job. Claude Science, on Pro and higher plans, runs analyses and ties each figure to its code. ChatGPT deep research builds a report from the sources you approve.

Is Claude Science free?

No. It is in beta on Pro, Max, Ultra, Team and Enterprise plans. Pro costs $17 a month billed annually or $20 monthly, as of November 2026. Academic and nonprofit scientists can get a discounted Team plan.

Can AI do a literature review?

It can find and summarise papers, but you must check them. Claude Science has a reviewer that flags incorrect citations. ChatGPT deep research lets you choose sources and read the plan first.

Can Claude connect to Zotero?

Yes, through community-built Zotero connectors. Check the install box on that page for your app.

Can Claude or ChatGPT search arXiv and PubMed?

Yes, with a connector. The arXiv MCP searches arXiv and Paper Search covers arXiv, PubMed and bioRxiv.

Are there Claude skills for science?

Yes. See the Claude science skills ranking. Claude Science also lets you save any pipeline as a reusable skill.

Is my unpublished research data private?

Claude Science keeps raw datasets on machines you choose, but Anthropic processes your prompts and answers under standard retention. Anthropic lists model training as opt-out on individual plans, so set it before uploading.

Where does Claude Science run?

On your laptop, a lab Linux box, an HPC login node or a cloud VM. Jobs run on local kernels, a Slurm cluster over SSH, or your Modal account.

Does ChatGPT deep research cite its sources?

It builds the report from the sources you approve and shows how it evaluates them while it works. Open each one yourself before you cite it.

Sources

Docs, issue trackers and discussions this page is built from.

Show all 6

Claude Science (beta) · claude.com
Claude pricing · claude.com
Deep research in ChatGPT · chatgpt.com
ChatGPT plans and pricing · chatgpt.com
Chats for science and medicine · chatgpt.com
ChatGPT data controls · help.openai.com