Claude vs Gemini for New Zealand document work.

How do Claude and Gemini compare for document work?

Claude is a standalone assistant that reads whatever you upload or connect. Gemini is built into Google Workspace, with Gemini Notebook (formerly NotebookLM) as its dedicated tool for answering questions across a set of sources. The table compares what matters for long documents.

Where is each one stronger?

Claude suits the heavy reading jobs: a 150-page tender pack, two drafts of a lease, a year of board papers. Every current model takes up to a million tokens, and Claude Docs and Slides turn the result into a finished document. Its connectors reach Google Drive and SharePoint alike, which helps firms whose documents sit in both. See our Claude guide.

Gemini suits teams whose documents already sit in Drive. Gemini Notebook keeps a defined set of sources together and answers from them, and Google added a secure cloud computer in July 2026 so it can run code for data analysis on higher tiers. For a Workspace Business Standard customer it adds no cost. See our Gemini Notebook guide.

How do you test Claude and Gemini on the same documents?

Use real documents with personal information removed, and give both tools identical files and questions. Our lesson on summarising documents with citations covers the checking habit.

1. Compare two lease drafts. Give both tools two versions of the same commercial lease and ask for every changed clause, with clause numbers and old and new wording. Compare with a redline from your word processor: count missed changes and changes that never happened.
2. Brief from 30 board papers. Load a year of board papers and ask ten questions you can answer yourself, two of which the papers do not cover. Score correct answers, accurate references to the right paper, and whether each tool says “not in these documents”.
3. Extract 20 supplier invoices. Ask for a table of supplier, invoice number, date, amount excluding GST, GST and total. Check every row: the GST in a GST-inclusive total should be 3/23 of it, and scanned invoices are where errors cluster.

For the second task, our engineering guide to evaluating long context models explains why the absence questions matter most.

What do Claude and Gemini cost over a year?

Worked example for a six-person team, at prices shown on 19 October 2026:

• Claude Team, Standard seats, annual billing: 6 × $40 = $240 a month, or $2,880 a year, stated as GST-inclusive.
• Gemini, firm already on Business Standard: no extra licence cost.
• Gemini, firm on Business Starter: upgrading six people to Standard costs 6 × (NZ$21 − NZ$10.50) = NZ$63 a month, or NZ$756 a year, with GST and billing term to confirm.

Later costs:

• Limits, not invoices. Heavy document users hit Claude’s five-hour and weekly limits first; a Premium seat costs five times Standard. On Gemini, notebook source caps and Workspace AI quotas are the ceiling.
• Retention choices. Claude Fable’s default 30-day retention may clash with a stricter internal policy, which pushes you to Opus 5.5 or Sonnet 5.5 for some work.
• Switching. Notebooks and Projects do not move between vendors, but your source documents and question sets do. Keep the test set; it is your evidence when you reconsider.

Which should you choose?

Choose Claude if

• Your core work is long documents you read, compare and rewrite, across any file store.
• Documents sit in both Google Drive and SharePoint.
• You want finished outputs, such as a board paper or slide deck, from the same tool.
• Your tests show it misses fewer changes and admits more gaps.

Choose Gemini if

• Your documents live in Google Drive and you are on Business Standard or above.
• You want a defined set of sources per matter or project, kept together in Gemini Notebook.
• You prefer not to add a second AI vendor to your privacy assessment.
• You may later need an enterprise notebook whose data stays inside your own Google Cloud project.

Where do both fall short?

• Superseded versions. Give either tool three drafts of a policy and it may answer from the wrong one. Remove old versions or name the current one in the prompt.
• Scanned pages. Photographed or faxed pages produce misread figures. For volume work, see OCR versus vision models.
• Legal meaning. Both can find and compare clauses. Neither should decide what a clause means under New Zealand law; see our contract clause review workflow.
• Offshore processing. Neither processes documents in New Zealand. Under the Privacy Act 2020 you remain responsible for keeping personal information in those documents secure (IPP 5), so strip what the task does not need.
• Scale. Past a few hundred documents, loading everything into one conversation gets slow and expensive. That is a retrieval problem; see RAG versus long context and our AI document processing service.

Written by the Levy Street team in Wellington. Last updated 19 October 2026.