Agenticum · Thoughts · 8 min read

Which AI Model for Your Business: ChatGPT, Claude, Gemini or Copilot

Type "Claude vs ChatGPT" into Google and you get benchmark charts. For a company in Dubai or Riyadh the first questions are different: which plan keeps your documents out of model training, where the data sits, and what a seat costs. We checked the four vendors' own pages on 27 October 2026.

Nikolay Zuykov · Agenticum · 27 October 2026

Take an example. A finance team in Dubai has to pick an assistant. Most comparisons rank the models on reasoning tests and end with "it depends on your use case". They skip the questions that matter to this team: its contracts live in SharePoint, two people already pay for ChatGPT Plus from their own cards, and customer passport scans count as personal data under UAE law.

Questions like these decide the answer more than any benchmark. So this guide starts there.

Start from where your documents already live

An assistant is useful in proportion to what it can read. The vendors know this, which is why each one ships a version that sits inside an office suite.

Many companies we talk to end up with two: the suite assistant for everyday mail and documents, and one standalone tool for the people who write, analyse or code all day. That is fine. What is not fine is the third category, which is the personal ChatGPT or Gemini account someone opened on their own, because that one plays by consumer rules.

The data question, plan by plan

Every vendor now says some version of "we don't train on your data". The sentence is true for the business plans and false for the consumer ones, and the difference is written in the small print of each pricing page. Here is what the pages said on 27 October 2026.

ProductConsumer plans (Free, Plus, Pro, Go, personal Gemini, Claude Free, Pro, Max)Business plansWorth knowing
ChatGPT (OpenAI)"Content is used to train our models. Opt-out available." You switch it off in the settings; by default it is on.Business and Enterprise: "No training on your business data by default."Enterprise lists data residency in ten regions, and the UAE is one of them. Business is for teams of 2 to 200.
Claude (Anthropic)Pricing page: model training "Opt-out". The privacy centre describes it as a setting you choose, and incognito chats are never used.Team: "No model training on your content by default." Enterprise: "None by default."Enterprise adds audit logs, SCIM, custom retention and IP allowlisting; 20-seat minimum. Data residency is not listed on the pricing page.
Gemini (Google)Personal accounts keep "Gemini Apps Activity" on by default. A subset of chats is read by human reviewers, and reviewed chats are kept for up to three years even if you delete your history.Workspace: "Workspace does not use customer data for training models without customer's prior permission or instruction." No human review. Interactions stay within your organisation.Google's own note: your Workspace data region settings do not apply to files inside Gemini Notebook.
Copilot (Microsoft)Personal Copilot is a different product with different terms; we did not assess it here."Prompts, responses, and data accessed through Microsoft Graph aren't used to train foundation LLMs." Stored under the same contract as the rest of your Microsoft 365 data.Copilot is covered by Microsoft's data residency commitments. Anthropic models offered inside Copilot are excluded from the EU Data Boundary, so check which models your admin has turned on.

Two consequences follow. First, the "bring your own ChatGPT" habit is the real risk in most offices, because the consumer default is to learn from what people paste in. Second, the business plans are close enough on the training question that it should not be the deciding factor between them. Residency, retention and admin controls are where they still differ.

What a seat costs

Vendors show prices in the currency of the buyer's location, so the table mixes dirhams, dollars and euros exactly as the pages showed them to us. Annual billing where the page offered it; taxes excluded.

PlanPrice per user per month, as shown on 27 October 2026Notes
ChatGPT Business, standard seatAED 73 (AED 92 billed monthly)Shown to a UAE buyer. Premium seat AED 365 with 5x usage.
ChatGPT EnterpriseCustomCredit-based and token-based pricing available.
ChatGPT Plus (personal)AED 80Go is AED 30, Pro from AED 385. Consumer terms apply.
Claude Team, standard seatUSD 20 (USD 25 billed monthly)2 to 150 seats. Premium seat USD 100.
Claude EnterpriseUSD 20 plus usage at API rates20-seat minimum.
Claude Pro (personal)USD 17 (USD 20 billed monthly)Max from USD 100.
Google Workspace Business Standard with GeminiEUR 13.60Google's page quoted us in euros; Starter EUR 6.80 has only the Gmail assistant, Plus EUR 21.10 adds Vault. Check the AED or SAR page for your region.
Google AI Pro (personal)AED 76.99Personal accounts only; AI Plus AED 36.99.
Microsoft 365 Business Standard with CopilotUSD 23.50Business Premium with Copilot USD 32. Shown on Microsoft's Middle East page.
Microsoft 365 Copilot Business add-onFrom USD 18 (was USD 21)On top of an existing Microsoft 365 licence. Advanced features such as Cowork and Autopilot are billed on usage.

Read the table for shape, not for the last dirham. For a team of twenty, the standalone assistants cost roughly the same as the suite add-ons, and all of them are cheaper than one hour of a specialist's time per person per month. The money question is rarely the licence. It is the hours people spend fixing output they trusted too early, which is why the test in the next section matters more than this table.

Local law changes the weight of these answers

Both markets now have a general data protection law with teeth, and both regulate sending personal data abroad.

In practice this means the residency column above is not a nice-to-have for a clinic, a bank, a school or any company that handles Emirates ID or Iqama data. If a vendor cannot tell you where the data is processed and stored, the assistant can still be used, but only for content that contains no personal data. That is a policy question, not a model question, and we are not giving legal advice here; ask counsel about your own data flows.

Test on your own work, not on benchmarks

Model names change every few months. The pricing pages we checked listed GPT-6 Astra, Claude Sonnet, Gemini 4 Argon and a model picker inside Copilot. Any ranking we printed would be stale by spring. A two-week trial on your own tasks will not be.

  1. Pick three recurring tasks with a verifiable result: a weekly client summary, a reply to a standard tender question, a reconciliation in Excel.
  2. Give each assistant the same inputs, from the same person, for ten working days. Use the business plan of each, not the free version.
  3. Score two things only: was the output used as-is, and how many minutes did someone spend correcting it. Keep the failed attempts in the count.
  4. Note what each tool refused or could not reach. An assistant that cannot open your SharePoint folder loses on the first day, whatever its benchmark score.

Cost per correctly finished task, including the correction time, is the number to compare. It usually points to one tool for writing and one for data, and that is a legitimate result.

Our take

For a company of 20 to 200 people in the UAE or Saudi Arabia, we would decide in this order. If the company runs on Microsoft 365 and the add-on budget exists, start with Copilot, because the permission model is already in place. If it runs on Google Workspace, use the Gemini that comes with Business Standard and see how far it goes before buying anything else. If neither suite is central, buy ChatGPT Business or Claude Team for the people who need it, and choose between those two on the test above, not on the headlines.

Whatever you pick, write down that personal accounts are not to be used for company material, switch off training on any consumer account that remains, and put the decision in the calendar for review in six months. The vendors will have changed their plans by then. They changed them twice while we wrote this.


Sources checked on 27 October 2026. Prices are as shown on the vendors' pages to a buyer in the UAE (ChatGPT, Google AI plans), on Microsoft's Middle East page (USD) and on Anthropic's and Google Workspace's pages (USD, EUR). The recommendations are our analysis at Agenticum, not legal advice.