Which AI assistant is best for work? Five compared on terms, controls and seat prices

Prices and features checked: 2026-11-09. Prices in the currency the vendor shows, checked on the date shown; we do not convert.

How we compared them

Five assistants, one set of criteria, published before the verdicts. We did not run a benchmark; we read each vendor's pricing page and data-use documentation on 2026-11-09 and scored what they state, because for work the terms matter at least as much as the model. Where a page refused our request or rendered without numbers, the cell says so rather than guessing. Three of the five (ChatGPT, Claude, Perplexity) pay us nothing; Perplexity is in our catalog but we have no active programme with it; Google and Microsoft have no programme we use. Affiliate status did not affect any verdict.

The criteria, in the order we weigh them for work:

1. Data handling. Are your prompts used to train models by default? How do you opt out? Is there a business plan where training is excluded without a setting?
2. EU availability and contracting entity. Which legal entity you contract with in the EEA, and any stated data-residency position.
3. Seat price with date and currency, at the plan a small team needs. We report the currency the vendor served us and do not convert.
4. Admin controls on the team plan: central billing, SSO, retention settings.
5. Where it lives. Inside the suite you already work in, or a destination app you switch to.
6. Who should not buy it. Every product has a buyer it will disappoint.

The research backdrop is covered in its own section, because it changes the advice: assistants help most on routine writing and for less-experienced people, and can mislead on tasks just outside their competence.^{[1] [2]} B

Fit by task type

This is a judgement from the vendors' own documentation of what each product does and where it runs, not a measured test. Use it to shortlist, then try the free tier on your own work.

Task type	Strong fit	Also works	Note
Drafting and editing documents	Claude, ChatGPT	Gemini, Copilot	The randomised evidence for routine writing is the strongest in the field [1]
Working inside Word, Excel, Outlook, Teams	Microsoft 365 Copilot	—	Copilot is the only one that lives inside those apps as part of the licence
Working inside Gmail, Docs, Sheets	Gemini (Workspace)	—	Admin controls Gemini Apps Activity on work accounts [5]
Research with sources to check	Perplexity	ChatGPT, Gemini	Perplexity answers with citations and sells more of them on Pro; you still have to read them [22]
Long documents and careful summaries	Claude	ChatGPT, Gemini	A destination app rather than an overlay
Analysis of data you paste or upload	ChatGPT, Claude	Gemini, Copilot	Correctness drops outside the model's competence; check the numbers [2]
Coding help	Claude, ChatGPT	Gemini, Copilot	Covered in depth in AI coding agents

Data handling and training opt-outs

This is the table the other comparisons skip. Every cell is from a vendor page read on 2026-11-09; where we could not read a page we say so.

Product	Consumer plan trains on your chats by default?	How to opt out	Business plan excluded from training?	Retention (as stated)
ChatGPT	Not verified today. OpenAI's data-use help article, privacy policy and enterprise-privacy page all returned HTTP 403 to us	Not verified; check the data-controls setting in your own account before use	Not verified today	Not verified today
Claude	Used for training only if you allow "Model Improvement" in Privacy Settings, if a chat is flagged for safety review, or on explicit opt-in [3]	Privacy Settings, "Model Improvement" [3]	Yes: "By default, we will not use your inputs or outputs from our commercial products" to train [4]	Deleted chats removed from back-end storage within 30 days; feedback kept 5 years; policy-violation data up to 2 years [4]
Gemini	Yes when "Keep Activity" is on, which is the default for adults; activity is used "including training generative AI models" [5]	Gemini Apps Activity, turn off "Keep Activity"; audio and Live recordings are off by default [5]	On work and school accounts the activity setting "is controlled by your Google Workspace administrator"; an explicit training exclusion was not in the text we read [5]	Auto-delete defaults to 18 months (options 3 or 36 months); with Keep Activity off, chats kept 72 hours; human-reviewed chats up to 3 years [5]
Microsoft 365 Copilot	Not a consumer product. "Prompts, responses, and data accessed through Microsoft Graph aren't used to train foundation LLMs" [6]	No user setting needed; admins set retention through Purview [6]	Yes, this is the commercial service [6]	Set by your admin in Purview; users can delete their own Copilot activity history [6]
Perplexity	Not stated as a default. The privacy notice lists "improve or create services and products, including our AI models" as a use of data [7]	The self-serve data page names no training toggle; threads and files can be deleted [7] [8]	The consumer notice does not apply to Enterprise and API offerings, where Perplexity acts as a processor [7]	Files kept 30 days by default; deleted accounts erased after 30 days; no general period stated [8]

Two practical conclusions. For anyone handling client or employee data, the business plans are the right unit of comparison, because the training question disappears without relying on a setting. And if a vendor's documentation cannot be read today, as OpenAI's could not, that is itself information: you cannot verify the terms you are agreeing to from outside an account.

EU availability and contracting entity

Product	EEA contracting entity	Stated EU data position
Claude	Anthropic Ireland, Limited for customers in the EEA, Switzerland and the UK [20]	No residency statement in the terms we read
Gemini	Google Ireland Limited provides Gemini Apps in the EEA and Switzerland [5]	No residency statement in the text we read; EU/UK legal-basis section present
Microsoft 365 Copilot	Microsoft's commercial terms (not re-read today)	"For European Union (EU) users, we have additional safeguards to comply with the EU Data Boundary"; Anthropic-hosted models in Copilot "are currently excluded from the EU Data Boundary"; web queries sent to Bing are not covered [6] [21]
Perplexity	US entity; VeraSafe named as EU and UK Article 27 representative [7]	Transfers "including but not limited to, the United States" [7]
ChatGPT	Not verified today (pages returned 403)	Not verified today

All five are available to EU users. The difference is where the data goes and under whose terms, and only Microsoft states a boundary, with exceptions.

Team plans and admin controls

• Claude Team covers 2 to 150 seats with central billing and administration and SSO; Enterprise adds spend limits, SSO, SCIM, audit logs and role-based access, with a 20-seat minimum.^[10] Owners can disable members' feedback submission under Organization settings.^[4]
• Microsoft 365 Copilot is managed in the Microsoft 365 admin centre, with Enterprise Data Protection, Purview retention and cost management.^[9]
• Gemini for Workspace puts the activity setting under the Workspace administrator.^[5] Gemini Enterprise plans list user management and IT admin controls.^[11]
• Perplexity Enterprise lists user management, SSO plus SCIM and audit logs; prices were not on the page.^[12]
• ChatGPT Business and Enterprise exist and are "priced per user per month" from two users, per the pricing page FAQ; the amounts did not render for us.^[13]

Seat prices, as served on 2026-11-09

Product	Plan a small team needs	Price as served	Currency	Notes
ChatGPT	Business	see pricing: page rendered plan names without amounts; press reporting puts consumer plans at roughly USD 5 to 200 a month [13] [23] B	—	Business "priced per user per month", from 2 users
Claude	Team Standard seat	$20 per seat per month billed annually, $25 monthly; Pro for individuals $17 annual / $20 monthly [10]	USD, tax not included	Enterprise US$20 per seat, 20-seat minimum, plus usage
Gemini	Gemini Enterprise Business	from US$21 per licence per month; Standard/Plus from US$30 [11]	USD for Enterprise; consumer tiers were served to us in CZK (Google AI Pro 549.99 CZK a month)	We do not convert currencies
Microsoft 365 Copilot	Copilot Business add-on	$18.00 per user per month paid yearly (originally $21.00), $25.20 monthly; Business Standard with Copilot $23.50 yearly; Business Premium with Copilot $32.00 yearly [9]	USD	Microsoft's launch price for the enterprise add-on was USD 30 in 2023 [15]
Perplexity	Enterprise Pro	see pricing: the Pro page rendered no text and the pricing page returned 404 [12]	—	Free tier exists

We set out to publish EUR seat prices alongside USD. None of the vendors served us EUR today: four served USD and Google served Czech koruna for its consumer tiers, so EUR prices remain an open item we will add when a vendor shows them. VAT is added at checkout in the EU.

What the evidence says about assistants at work

Any single productivity number is wrong; the effect depends on the task and the person.

• Routine writing: 453 professionals finished mid-level writing tasks about 40% faster with about 18% higher rated quality, and weaker writers gained most, in an incentivised online experiment.^[1] A
• Analysis: 758 consultants were 12.2% more productive and 25.1% faster on tasks inside the model's competence, but 19 percentage points less likely to be correct on a task just outside it.^[2] B
• Customer support: 5,172 agents resolved about 14–15% more issues per hour with suggested replies; the gain was near zero for the most experienced.^[16] A
• Human plus AI on decisions: across 106 experiments, the combination performed worse than the better single agent on average, with gains in content creation.^[17] A

Usage is already broad. In Slack's spring 2025 Workforce Index of 5,156 desk workers in six countries, 60% used AI at work, up from 36% five months earlier, with 42% using it at least weekly.^[18] A In Germany, a Bitkom survey in early 2026 found almost half of employed people using AI at work, 8% daily and 18% at least weekly, while 48% did not use it at all.^[19] B The question for most teams is no longer whether to use an assistant but under which terms.

Which to pick, by role

• You live in Microsoft 365. Copilot, because it is the only assistant inside your files and meetings under commercial data terms, and the add-on price has fallen since launch. Do not buy if your work happens mostly outside Office; you would pay for an integration you do not use.
• You live in Google Workspace. Gemini, with the administrator controlling activity settings. Do not buy the consumer Google AI plans for work: Keep Activity is on by default and the chats are training data unless you switch it off.^[5]
• You write and edit long documents, or handle client material. Claude Team: training excluded by default on commercial plans, published seat prices, 30-day deletion for removed chats.^{[4] [10]} Do not buy if you want an assistant that appears inside every text field; it is a destination app.
• You research and need to check sources. Perplexity, on the free tier first. Do not buy the Enterprise tier without a written answer on pricing and processing terms, since neither was readable on the public pages today.
• You want the widest general assistant and the biggest user base. ChatGPT; more than 800 million people use it weekly, by its chief executive's account.^[14] B Do not buy a consumer plan for work until you have read the data-controls terms inside your account, because we could not verify them from outside today.

For the personal side of the same decision, see AI personal assistants; for how an assistant fits into a complete system, see how to build your personal AI operating system.

Where demand is heading

US searches for "ai assistant for work" average about 720 a month, small, but grew about 255% a year over three years with a keyword difficulty of 4 (DataForSEO, retrieved 2026-11-09); Google Trends share is moving from "ai assistant" toward "ai agents". In our Future Demand Radar, "personal AI operating systems" scores 53.2 out of 100 in the baseline scenario (band 46.6–59.4), classed "validate in 12 months", and "team productivity and AI collaboration" 51.3 (band 48.6–54.1). Our baseline scenario expects demand for personal-AI terms to reach about 1.6 times today's level by 2028 (band 1.3–1.9). Forecasts are model outputs with stated uncertainty, not facts.

What changed this month

First edition. From the next check, this section lists price and terms changes since the previous one. Two items to watch now: Microsoft is discounting the Copilot Business add-on from $21 to $18 per user per month on annual terms, and OpenAI's public data-use pages were unreadable to automated requests, so the ChatGPT rows above will be updated when they can be read.

What to do next

Decide the terms first, then the model. If your team handles client or staff data, shortlist only business plans with training excluded by default, then run the Tool Finder: it asks about your suite, team size, budget and data constraints and returns a stack with the same dated prices. If you want to try before deciding, Perplexity, Claude, ChatGPT and Gemini all have free tiers that show the product without a card.

Frequently asked questions

Which AI assistant is best for work?

It depends on the suite you live in and the terms you need. Copilot if your work is in Microsoft 365, Gemini for Google Workspace, Claude Team for long documents and client material under published terms with training excluded by default, Perplexity for research with sources, ChatGPT for the widest general assistant once you have read its data controls inside your account. The criteria and tables above are the reasoning.

Do AI assistants use my work chats to train their models?

On consumer plans it varies: Gemini does while "Keep Activity" is on, which is the default for adults;^[5] Claude only if you allow "Model Improvement";^[3] Perplexity lists model improvement among its uses without stating a default;^[7] OpenAI's pages could not be read by us today. On the business plans we could read, Claude for Work and Microsoft 365 Copilot state that your data is not used for training by default.^{[4] [6]}

How much does an AI assistant cost per seat for a small team?

As served on 2026-11-09: Claude Team $20 per seat per month billed annually or $25 monthly; Microsoft 365 Copilot Business add-on $18 per user per month paid yearly (from $21) or $25.20 monthly; Gemini Enterprise Business from US$21 per licence; ChatGPT Business and Perplexity Enterprise did not show amounts on their public pages.^{[9] [10] [11]} No vendor served us EUR today; VAT is added at checkout in the EU.

Does an AI assistant really make a team more productive?

For routine writing, a randomised experiment found about 40% time savings with higher rated quality.^[1] A For analysis, consultants gained inside the model's competence and were 19 percentage points less likely to be correct just outside it.^[2] B On decision tasks, human-plus-AI combinations under-performed the better single agent across 106 experiments.^[17] A The task decides, not the brand.

Are these assistants available in the EU and where does the data go?

All five are available. Claude contracts through Anthropic Ireland and Gemini through Google Ireland in the EEA;^{[20] [5]} Microsoft states EU Data Boundary safeguards for EU users with named exceptions (Anthropic-hosted models in Copilot and Bing web queries);^{[6] [21]} Perplexity transfers data to the United States with VeraSafe as EU representative.^[7] OpenAI's pages were unreadable to us today.

Sources and evidence

1. Experimental evidence on the productivity effects of generative artificial intelligence — Noy, Shakked; Zhang, Whitney (2023), RCT A doi.org/10.1126/science.adh2586 (retrieved 2026-11-09)
2. Navigating the Jagged Technological Frontier: Field Experimental Evidence of the Effects of Artificial Intelligence on Knowledge Worker Productivity and Quality — Dell'Acqua, Fabrizio; McFowland III, Edward; Mollick, Ethan R.; Lifshitz-Assaf, Hila; Kellogg, Katherine; Rajendran, Saran; Krayer, Lisa; Candelon, François (2023), RCT B doi.org/10.2139/ssrn.4573321 (retrieved 2026-11-09)
3. Is my data used for model training? (consumer products) — Anthropic Privacy Center (2026), vendor page A privacy.claude.com/en/articles/10023580-is-my-data-used-for-model-trai (retrieved 2026-11-09)
4. Claude for Work and API: model training default and data retention (privacy.claude.com articles 7996868 and 7996866) — Anthropic Privacy Center (2026), vendor page A privacy.claude.com/en/articles/7996868 (retrieved 2026-11-09)
5. Gemini Apps Privacy Hub and Gemini Apps Activity (support.google.com/gemini/answer/13594961 and 13278892) — Google (2026), vendor page A support.google.com/gemini/answer/13594961 (retrieved 2026-11-09)
6. Data, Privacy, and Security for Microsoft 365 Copilot — Microsoft Learn (2026), vendor page A learn.microsoft.com/en-us/copilot/microsoft-365/microsoft-365-copilot- (retrieved 2026-11-09)
7. Perplexity Privacy Notice — Perplexity AI (2026), vendor page A www.perplexity.ai/hub/legal/privacy-policy (retrieved 2026-11-09)
8. Self-Serve Data Options (deletion and retention of threads, files and accounts) — Perplexity AI (2026), vendor page A www.perplexity.ai/hub/legal/self-serve-data-options (retrieved 2026-11-09)
9. Microsoft 365 Copilot pricing (business plans) — Microsoft (2026), vendor page A www.microsoft.com/en-us/microsoft-365-copilot/pricing (retrieved 2026-11-09)
10. Claude pricing — Anthropic (2026), vendor page A claude.com/pricing (retrieved 2026-11-09)
11. Gemini subscriptions and Google AI plans (served in Czech with CZK prices) — Google (2026), vendor page A gemini.google/subscriptions/ (retrieved 2026-11-09)
12. Perplexity Enterprise (features; pricing not shown; Pro page rendered no text on 2026-11-09) — Perplexity AI (2026), vendor page A www.perplexity.ai/enterprise (retrieved 2026-11-09)
13. ChatGPT pricing (plan names rendered; amounts did not render on 2026-11-09; openai.com returned 403) — OpenAI (2026), vendor page C chatgpt.com/pricing (retrieved 2026-11-09)
14. Sam Altman says ChatGPT has hit 800M weekly active users — TechCrunch (2025), report B techcrunch.com/2025/10/06/sam-altman-says-chatgpt-has-hit-800m-weekly- (retrieved 2026-11-09)
15. Furthering our AI ambitions: Announcing Bing Chat Enterprise and Microsoft 365 Copilot pricing — Microsoft Official Blog (2023), vendor page A blogs.microsoft.com/blog/2023/07/18/furthering-our-ai-ambitions-announ (retrieved 2026-11-09)
16. Generative AI at Work — Brynjolfsson, Erik; Li, Danielle; Raymond, Lindsey (2025), observational A doi.org/10.1093/qje/qjae044 (retrieved 2026-11-09)
17. When combinations of humans and AI are useful: A systematic review and meta-analysis — Vaccaro, Michelle; Almaatouq, Abdullah; Malone, Thomas (2024), meta-analysis A doi.org/10.1038/s41562-024-02024-1 (retrieved 2026-11-09)
18. Workforce Index spring 2025: The New AI Advantage — Slack Workforce Lab (Salesforce) (2025), report A slack.com/blog/news/the-new-ai-advantage (retrieved 2026-11-09)
19. Ein Drittel nutzt KI mindestens einmal pro Woche (Presseinformation) — Bitkom e.V. (2026), report B www.bitkom.org/Presse/Presseinformation/Ein-Drittel-nutzt-KI-mindesten (retrieved 2026-11-09)
20. Commercial Terms of Service (training clause, contracting entity) — Anthropic (2025), vendor page A www.anthropic.com/legal/commercial-terms (retrieved 2026-11-09)
21. Privacy and protections in Microsoft 365 Copilot Chat — Microsoft Learn (2026), vendor page A learn.microsoft.com/en-us/copilot/privacy-and-protections (retrieved 2026-11-09)
22. What is Perplexity Pro? (help center; no price stated) — Perplexity AI (2026), vendor page A www.perplexity.ai/help-center/en/articles/10352901-what-is-perplexity- (retrieved 2026-11-09)
23. OpenAI sees at least 220 mln users paying for ChatGPT by 2030 (The Information via Reuters) — Investing.com / Reuters (2025), report B ng.investing.com/news/stock-market-news/openai-sees-at-least-220-mln-u (retrieved 2026-11-09)