Key takeaways
- The choice depends mostly on where your documents are. Copilot works inside Word, Excel, Outlook and Teams; Gemini inside Gmail, Docs and Sheets. ChatGPT and Claude are separate tools that connect to both.
- Entry prices are close, as of November 2026: $20 per user per month on annual billing for ChatGPT Business and Claude Team, $30 for Microsoft 365 Copilot in large companies. Usage billed on top differs more than the seat price.
- For long contracts, the only four-tool comparison found, by a company that sells procurement AI courses, puts Claude first and Gemini second. It publishes no scores: read it as one view, not a test.
- No assistant is reliable on spreadsheets without a check. Microsoft’s own Excel agent completed 57.2% of a public benchmark’s tasks, and the best model on a harder 2026 benchmark scored 34.89%.
- EU data residency depends on the plan. OpenAI offers it on ChatGPT Enterprise, Google on Workspace Enterprise Plus, Microsoft through its EU Data Boundary, except for Anthropic models; Anthropic documents it for Claude only through cloud platforms.
Which AI is best for procurement?
The best AI for procurement is usually the assistant your company already licenses, in the suite where your contracts and spreadsheets are. Microsoft 365 companies start with Copilot, Google Workspace companies with Gemini. ChatGPT and Claude make sense as a second tool, for long documents or where the main suite has no AI licence.
This page compares four general AI assistants: ChatGPT, Microsoft 365 Copilot, Claude and Gemini. It is built from what the vendors publish (plans, prices, data terms) and from published reviews and benchmarks, each named with who wrote it. It is not a hands-on test: no tool was scored for this page.
Two names cause confusion. Here, Copilot means Microsoft 365 Copilot and the free Copilot Chat, not the Copilot in Dynamics 365 or the US government’s Procurement Co-Pilot tool. And if you are looking for procurement software with AI built in, such as a sourcing or spend analysis suite, that is a different choice: eProcurement Toolkit’s guide How do you choose AI procurement software? covers it.
Which assistant fits which procurement task?
For each everyday procurement task, the table gives the assistant the published sources point to, who says so, and the check the task needs. The verdicts on tasks come mostly from one source, Procurement Tactics, a company that sells AI courses for procurement teams and publishes no scores. Where the evidence is that thin, the table says so.
| Task | What the sources point to | Who says it, and the limits | What to check |
|---|---|---|---|
| Reading long contracts (summary, risk clauses) | Claude first, Gemini a close second | Procurement Tactics’ comparison. On the vendors’ pages, Claude Team lists a context window of up to 1M tokens, depending on the model; ChatGPT Business lists 256K tokens for its reasoning model, about 320 pages of input. No independent test of the four products on contracts was found | Every clause and value against the signed text: in a 2025 benchmark, the best legal AI tools extracted contract data about as well as lawyers, 73% to 75% against 71% |
| Spend analysis and classification in a spreadsheet | The assistant inside your suite: Copilot in Excel for Microsoft 365, Gemini in Sheets for Google Workspace | Procurement Tactics says the link with the spreadsheet decides. Suplari, a spend analysis vendor, warns that a general assistant does not know your supplier master, so two spellings of one supplier count as two suppliers | Supplier names merged first; totals rebuilt with formulas; a sample of lines checked against your taxonomy |
| Drafting an RFP or tender section | Any of the four | Procurement Tactics says that for short drafting and everyday research, ChatGPT and Claude are close enough that habit and workflow fit should decide. Microsoft publishes a step-by-step Copilot scenario for a supplier RFP in Word | Requirements specific to your company; standards and figures the draft names; no confidential data outside a company workspace |
| Supplier research | An assistant with web search: ChatGPT, Claude and Copilot Chat list it on their plan pages | JAGGAER, a procurement software vendor, writes that Claude sometimes invents plausible supplier details, such as ISO certifications. The same risk applies to any assistant | Open every link; check certificates with the issuer; confirm the company exists and has the size you need |
| Negotiation preparation | Claude for the quality of one answer, custom GPTs in ChatGPT for the same method across a team | Procurement Tactics only | No real target price or fallback position outside a company workspace; made-up figures for role-play |
Long contracts are where sources most often name one tool, Claude; for spreadsheets the suite you work in decides; for drafting and research the four are close, and every output needs a check.
For how to use one assistant on these tasks day to day, with example prompts and data rules, see ChatGPT for procurement. The prompts work the same way in Copilot, Claude or Gemini.
Can you trust an AI assistant with procurement numbers?
No AI assistant can be trusted with procurement numbers without a check. Totals, price comparisons, savings and supplier concentration are where the published evidence is weakest, and an assistant gives a wrong figure in the same confident tone as a right one. The evidence comes from vendors and researchers alike.
- Microsoft’s own Excel agent. Microsoft reported in 2025 that Agent Mode in Excel completed 57.2% of the 912 tasks of SpreadsheetBench, a public benchmark. Microsoft adds that the benchmark does not test pivot tables, charts or formatting.
- A harder benchmark. On SpreadsheetBench 2, presented at NeurIPS 2026 with business spreadsheet workflows from start to end, the best model reached 34.89% overall task accuracy.
- Changed numbers. Apple researchers found that the performance of all the models they tested fell when only the numbers in a problem changed.
- Spend data. Suplari, which sells spend analysis software, writes that a general assistant does not know your supplier master, so every concentration analysis is wrong by default when one supplier appears under several names. Procurement Tactics says Copilot in Excel will sometimes put a supplier in the wrong category or miscalculate a concentration figure.
- Reading figures. JAGGAER writes that Claude can give confident wrong answers when asked simply to read the numbers in a spreadsheet.
Suplari and JAGGAER sell procurement software that competes with general assistants, so read their warnings as vendor arguments. The mechanism they describe is easy to test on your own data, which is the point.
- Ask the assistant to compute in the spreadsheet or in code, with the formulas visible, never as a total written in the chat.
- Merge supplier names (one spelling per supplier) before any total, ranking or concentration figure.
- Rebuild the key totals yourself from the source file and compare.
- For classification, check a sample, for example 50 lines, against your taxonomy and count the errors.
- Before a figure goes to a supplier or a committee, a named person checks it and signs it off.
Which assistant fits your company setup?
Your company setup narrows the choice before any task does. The suite your documents live in, the data terms your IT and legal teams require, and whether EU data residency is mandatory each remove options. The table starts from the setup and gives the first assistant to look at.
| Your setup | Look first at | Why | Watch out for |
|---|---|---|---|
| Microsoft 365, documents in SharePoint, OneDrive and Outlook | Copilot Chat (included), then Microsoft 365 Copilot licences for buyers who work in Excel and Outlook | It works inside the files and mails the buyer already has; Microsoft publishes a procurement scenario across Excel, Outlook, Teams and Word | Copilot sees everything the user’s permissions allow, which Procurement Tactics notes is often more than users realise: clean SharePoint sharing first. Agents are billed on usage |
| Google Workspace, documents in Drive, Docs and Sheets | Gemini in Workspace Business Standard or above | Gemini works in Gmail, Docs, Sheets and Drive; Business Starter has it in Gmail, not in Docs, Sheets or Drive | EU data regions for Gemini features need Enterprise Plus or the Assured Controls add-ons |
| Long contracts are a large part of the work | Claude Team or Enterprise, next to your main suite | Anthropic lists a context window of up to 1M tokens on Team, depending on the model; the one four-tool comparison found puts Claude first on contracts | EU data residency for Claude’s own apps is not documented by Anthropic; Enterprise usage is billed on top of the seat |
| EU data residency is mandatory for a standalone assistant | ChatGPT Enterprise | OpenAI lists data residency in ten regions, including the EU, on Enterprise | ChatGPT Business has no data residency; Enterprise pricing is on request |
| No company AI licence yet | A company workspace of any of the four, before any use on supplier data | Business plans do not train on your data by default and come with a data processing agreement | Personal accounts may use your content for training: no supplier prices, bids or contracts there |
Start from the suite your documents are in, then check data residency and admin controls with IT; add a second assistant only for a task the first one does badly.
What do the business plans cost, and what do they do with your data?
The business plans of the four assistants cost about the same per user: $18 to $30 per month on annual billing, as of November 2026. They differ more on what is billed on top, on admin controls and on where data can stay. Prices below are the vendors’ public list prices, read on 7 November 2026; each links to its pricing page.
| Plan | List price per user per month (as of November 2026) | Useful for procurement work | Training on your data | EU data residency |
|---|---|---|---|---|
| ChatGPT Business | $20 billed annually, $25 billed monthly (Standard); 2 to 200 users | File uploads, data analysis, Excel, Word, PowerPoint and Google Sheets extensions, connectors to Microsoft 365 and Google Workspace, deep research | Not by default | No |
| ChatGPT Enterprise | Custom pricing | Larger context window, SCIM, role-based access, custom data retention | Not by default | Yes, in ten regions including the EU |
| Microsoft 365 Copilot Chat | Included with an eligible Microsoft 365 subscription | Web-grounded chat, Copilot in some Microsoft 365 apps; agents billed on usage, with an Azure subscription | Microsoft lists enterprise data protection | Microsoft’s EU Data Boundary, except models provided by Anthropic |
| Microsoft 365 Copilot (large companies) | $30 paid yearly, on top of an eligible Microsoft 365 licence | Copilot in Word, Excel, PowerPoint, Outlook and Teams, using your business data; Researcher and Analyst agents | Microsoft says prompts, responses and Microsoft 365 data are not used to train its foundation models | Same as Copilot Chat |
| Microsoft 365 Copilot Business | From $21 paid yearly; $18 in the first year for offers taken between 1 July and 31 December 2026; $25.20 monthly | Same apps, for Microsoft 365 Business plans | Same as Microsoft 365 Copilot | Same as Copilot Chat |
| Claude Team | $20 billed annually, $25 billed monthly (Standard); 2 to 150 seats | Projects, files created with code execution, connectors, Claude for Microsoft 365, web search, context window up to 1M tokens depending on the model | None by default | Not on the pricing page; Anthropic documents EU processing only through AWS Bedrock and Google Vertex AI |
| Claude Enterprise | $20 per seat billed annually, 20 seats minimum, plus usage at API rates | SSO, SCIM, audit logs, Compliance API, custom data retention | None by default | Same as Claude Team |
| Google Workspace Business Standard (Gemini included) | €13.60 with a one-year commitment (the page showed euros; Plus €21.10, Starter €6.80) | Gemini in Gmail, Docs, Sheets and Drive, the Gemini app | Google’s pricing page says your company data is not used to train models when you chat with Gemini | Only on Enterprise Plus or with Assured Controls add-ons |
The seat prices are close; the real budget questions are usage billed on top (Copilot agents, Claude Enterprise) and the plan level your IT team needs for SCIM, audit logs and data residency.
These are the vendors’ list prices, and they change often: Microsoft’s first-year offer on Copilot Business ends on 31 December 2026. Check each pricing page on the day you decide. Usage billed on top is the hardest part to budget: Microsoft bills agents by use, and Anthropic bills Claude Enterprise usage at API rates on top of the seat.
On data, the line that matters is a company workspace against a personal account, more than one brand against another. On their business plans, OpenAI and Anthropic say they do not train on your data by default, and Microsoft says the same for Microsoft 365 Copilot. On personal plans, your content may be used for training. A company workspace does not lift your supplier NDAs or your own data rules.
What do published reviews and analysts say?
Published reviews comparing ChatGPT, Copilot, Claude and Gemini for procurement are few, and none is independent. No open comparison by Gartner, Forrester, IDC, The Hackett Group, Spend Matters or CIPS was found in November 2026. The sources that exist come from course sellers and from software vendors.
| Source | Who wrote it | What it says | How to read it |
|---|---|---|---|
| Procurement Tactics, AI for procurement comparison (2026, no date on the page) | A procurement training company that sells AI courses | Five tasks: Claude first on contracts (Copilot for standard contracts), supplier continuity plans and one-off negotiation preparation; Copilot or Gemini first on spend classification, depending on your suite; Claude or Gemini for category strategy content | The only four-tool comparison found. No scores, no tool versions, no named testers; the page calls its own assessment directional |
| JAGGAER blog on Claude (July 2026) | A source-to-pay software vendor | Claude reads long contracts well; it cannot run transactions or approvals, sometimes invents supplier details, and misreads spreadsheet numbers | A competitor of general assistants. Its limits match other sources; its own figures cannot be checked |
| Suplari blog on ChatGPT and spend (June 2026) | A spend analysis software vendor | General assistants do not know your supplier master and do not show their work | A vendor argument, with a mechanism you can check on your own data |
| Benchmarks of AI models (Vals AI and others) | Independent evaluators | Rankings of models on finance, legal and spreadsheet tasks | They rank models through their APIs, not the ChatGPT, Copilot, Claude or Gemini products, and they change every month |
There is no independent product test for procurement: use these sources for where to look, and your own documents for the decision.
Gartner published a paid report on AI use cases in procurement in October 2026; its content was not available to read for this page.
Which procurement teams have published results with these tools?
Few procurement teams have published results with a named general assistant, and the ones that exist come from vendors or resellers. None was found for ChatGPT Enterprise in procurement. Read the cases below as vendor claims: useful to see what teams try, not as proof of a result you would get.
- Anthropic, with Claude and Zip (vendor claim). In a customer story published by Zip, a procure-to-pay vendor, Anthropic’s procurement team says its procure-to-pay volume more than doubled while procurement headcount stayed flat, with 95% policy adoption in the first year. The team uses Claude for vendor negotiations, savings calculations, policies and procure-to-pay metrics. Both companies sell the tools in the story.
- Softcat, with Microsoft Copilot (reseller story). Softcat’s internal procurement team used Copilot to analyse 11 supplier responses, recorded transcripts and internal discussions stored in OneDrive when choosing a travel management system. The story gives no figures. Softcat resells Microsoft products.
- A US federal agency, with Microsoft 365 Copilot (Microsoft blog). A contracting officer’s justification document took under 30 minutes instead of half a day, in one case. The larger figure in the same post, up to 3,600 hours a year, is an estimate for as many as 800 contracts a year.
- Microsoft’s procurement scenario (demonstration). Microsoft publishes a five-step Copilot scenario: spend gaps in Excel, negotiation preparation in Outlook, supplier scoring from Teams, a supplier table in Copilot Chat and contract terms in Word. Microsoft says it is for demonstration only; it has no results.
How do you choose and roll out an assistant in a procurement team?
Choose by trying the assistant your company licenses on two real tasks, in the company workspace, and counting the errors. Published reviews cannot tell you how a tool handles your contracts, your spend file or your suppliers’ language. Your own documents can, in a few weeks.
- List the AI licences your company already has, and the personal accounts your buyers use.
- Pick two tasks with one owner each: for example contract summaries and spend classification.
- Run them in the licensed assistant, with the checks above, for two to four weeks.
- Write down every error found at review. That list shows where the tool saves time and where it costs time.
- Add a second assistant only for a task the first one keeps getting wrong, such as very long contracts.
The rules that go around the tool, data classes, access rights, human sign-off and the EU AI Act’s AI literacy duty, are on the AI procurement strategy page.
I run AI training for procurement teams: in-house, up to 12 people, from €2,500 for a half day or from €4,500 for a full day, on the assistant your company licenses. The team practises RFP drafting, contract review with a check against the text and supplier research with sources, and in the full day spend classification and negotiation preparation. It leaves with a one-page charter on which data goes into which tool, and an attendance record for the AI literacy duty of the EU AI Act.
Frequently asked questions
Short answers to the questions people ask, each one sourced above.
Which AI is best for procurement?
No single AI assistant is best for every procurement task. The assistant that works inside your suite usually fits best: Microsoft 365 Copilot in a Microsoft 365 company, Gemini in Google Workspace. The one published four-tool comparison puts Claude first for long contracts. Whatever you choose, check totals and contract values against the source.
Is Claude good for procurement contracts?
Claude is often named for long contracts: Anthropic lists a context window of up to 1M tokens on its Team plan, depending on the model, and the one four-tool comparison found, by a procurement course seller, ranks it first. No independent test confirms it. Check every clause and value it gives in the signed text.
What is Copilot for procurement?
In most searches, Copilot for procurement means Microsoft 365 Copilot used by buyers in Excel, Outlook, Teams and Word, or the Copilot Chat included with Microsoft 365. It is not the Copilot in Dynamics 365 or the US government’s Procurement Co-Pilot. As of November 2026, Microsoft 365 Copilot costs $30 per user per month, paid yearly.
Is ChatGPT or Copilot better for procurement?
It depends on where your documents are. Copilot works inside Microsoft 365 and sees what the user can already see, which helps with Excel and Outlook work but makes clean access rights a must. ChatGPT is a separate tool that connects to Microsoft 365 and Google Workspace. On their business plans, both say they do not train on your data by default.
Is Gemini good for procurement?
Gemini fits procurement teams that work in Google Workspace: from Business Standard upwards, it works in Gmail, Docs, Sheets and Drive. Published procurement reviews of Gemini are thin; the one four-tool comparison puts it second on contracts and first on spend classification for Google Workspace teams. EU data regions need Enterprise Plus or an Assured Controls add-on.
Can an AI assistant replace procurement software?
No. A general assistant drafts, summarises and analyses, but it cannot run approvals, purchase orders or payments, and it does not know your supplier master. Procurement suites now build AI in. For choosing procurement software with AI, eProcurement Toolkit has a neutral guide to AI procurement software.
Project already started? Check where it stands.
The health check scores your platform and your programme in three weeks.
