ChatGPT or Claude in the office: choosing the tool and preparing your people

There is no need for a league table between ChatGPT and Claude. For office work, both of them write, summarise and read long documents. The difference that matters to a company lies elsewhere. It lies in the contract, in the path the data takes and in the usage rules. That is why you settle these points first, and only then choose the tool. Then you prepare your people, one office at a time. If the company uses both, the same rules apply.

Why not start from the rankings?

Model rankings change every few months. New versions come out, old ones are retired and prices change. A comparison made today says little about what the company will be using a year from now.

Public benchmarks measure generic tasks. They do not show how the tool handles the procurement office’s contracts or the reports exported from the ERP system. And they say nothing about the terms of use, which are the part the company signs.

What lasts are the decisions about which documents can be used, who manages the accounts and who reviews a text before it leaves the company. They hold for any tool.

A typical scenario. In one company, some offices use ChatGPT with personal accounts. Others have tried Claude. Management wants to bring some order and asks for a side-by-side trial. A few weeks later two reports arrive, and each one praises the tool it tested. Neither says what happened to the client contracts that staff had already pasted into their conversations. That was the question that mattered.

What do you decide before the tool?

There are three things to decide, in this order.

• The contract. Which version you use and on what terms. Where the data is processed and how long it is kept. What happens when you leave. The clauses to look at are described in our guide on contracts with an AI vendor.
• The data. Which documents can go into the tool, and which only after they have been anonymised. Client contracts, staff data, unpublished financial figures and technical designs usually call for a specific decision. A simple list is enough, with each type of document and the rule next to it.
• The usage rules. Who uses the tool, for which tasks, who reviews a text before it goes out, and where the instructions that work are recorded. A few pages written for everyone are enough. Usage rules and the register of AI systems are the governance side of AI in a company.

You choose the tool afterwards. The choice often becomes easier, because the versions that do not meet the agreed terms rule themselves out.

Personal or business version: what changes?

Both providers offer personal versions and versions for organisations. The terms are not the same. For work documents, this difference matters more than any ranking.

• Who manages the accounts. In the business version, an administrator manages the accounts. They know who uses the tool and can remove access when someone leaves the company. With a personal account, the conversations stay with the person, even after they have left.
• Model training. For the business versions, both providers state that they do not use customer content to train their models. Anthropic says so in its commercial terms. OpenAI says it does not use data from its business versions for training unless the customer chooses to allow it. In the personal versions, it depends on an account setting. For Claude, since 2025 people on a personal version choose whether to allow it. If they do, the data can be kept for up to five years. For ChatGPT, the setting is under Data Controls.
• Which features are on. Depending on the version, the administrator can switch certain features on or off. They include sharing conversations through a link, connections to email and file stores, and memory from one conversation to the next. Which features stay switched on is a decision to make and to write down.

Terms change over time. It is worth reading them on the day you choose and keeping a dated copy.

Who decides?

It is not for each office to decide on its own, nor for IT alone. The choice affects the contract, the data and everyday work. That is why several people are involved, each with a clear role.

Who	What they decide
Management	Whether to adopt the tool, for which tasks and with which version
IT	Accounts, access, settings and connections to other systems
Data protection (the DPO, where there is one)	Which data can go in, and with what precautions
Legal and procurement	The contract terms and a copy of the terms accepted
Head of each office	The department rules and who reviews a text before it goes out

You also need one person to coordinate the work. They keep the list of approved tools up to date, collect questions from the offices and bring unclear cases to management. It can be someone already on staff, as long as they have the time and the mandate to do it.

If the choice between the two is still open, an outside view helps you read the terms and contracts before you sign. That is the work of our second opinion on AI projects and vendors. In that service we do not offer products of our own.

How do you prepare people, office by office?

A one-size-fits-all course achieves little. Each office works with different documents and faces different risks. The basics are the same, though, and there are three of them.

• Clear requests. A good request states the task, the source document, who the result is for and the form it should take. “Summarise this contract” gets a generic answer. “List the term, renewal and penalties of this contract in a table, with the clause number” gets an answer you can check.
• Checking the answers. The tool writes well even when it is wrong. So every figure is checked against the source, and every quoted clause is found in the document. Names, dates and amounts are reviewed one by one. Who reviews the work is decided beforehand, not afterwards.
• Department rules. Which documents can be used, and with which version of the tool. What the tool is not asked to do. Where the instructions that work are saved, so that colleagues can use them too.

Then the work moves to each office’s real tasks. Here are a few examples.

• Sales. A draft quote is prepared from earlier ones. Prices and conditions are checked against the current price list.
• Procurement. The tool finds the clauses on term, data and exit in a long contract, and shows where to look. The decision stays with whoever signs.
• Finance and administration. Two reports from the ERP system are compared side by side. Each discrepancy is traced back to the source before it is used.
• Technical department. The tool explains a passage from a manual in plain words. First you check that the manual is the current one.
• HR. Internal notices and policies are summarised for colleagues. Decisions about people, starting with candidates, are not left to the tool.
• Executive assistants and management. A meeting summary is drafted, with the actions and who owns them. It is always checked before it is passed on.

It pays to start with a single office that has recurring tasks and documents in good order. The rules written for the first office become the basis for the others. People who already use the tool every day can move on to an advanced level, with saved instructions, long documents and multi-step tasks.

What if the company uses both?

It happens often. One office gets on better with one tool, another with the other. Or two companies in the same group have made different choices. That is not a problem, as long as the rules are the same.

• Business versions for both. The terms of both are read and kept on file. You do not run one on a business plan and the other on a personal one.
• Same permitted data. A document that cannot go into one does not go into the other either.
• Same checks. The same rules apply to checking figures and quotations. Who reviews the work is decided in the same way.
• One list. The approved tools sit in a single register, with version, intended use and owner.

Only the practical steps change, such as where instructions are saved, how a document is attached or how a conversation is shared with a colleague. The course explains the rules once and has people practise on both tools.

In brief

• There is no single winner between ChatGPT and Claude. Rankings change quickly and say little about office work.
• First you settle the contract, the data allowed and the usage rules. Then you choose the tool.
• For work documents the business version is the better choice, with managed accounts and training terms read and kept on file.
• Several people decide, each with their own role. One person coordinates the work.
• Each office learns to write clear requests, check the answers and follow its department rules, working on its own tasks.
• If you use both tools, the same rules apply.

Frequently asked questions

Which of the two writes better?

It depends on the task and the version, and it changes with every update. A trial on the office’s real documents tells you more than any ranking. It should use documents the rules allow, not confidential ones.

Can we let everyone use their own account?

For work documents, it is better not to. The terms are those of the personal version. The company does not know what goes into the tool. When someone leaves, the conversations go with them.

Is switching off training enough?

No, it is only one part. What also matters is how long the data is kept, who can read it, which connections are switched on and how you exit the contract.

Do ChatGPT and Claude need different courses?

No. The rules, the way to write requests and the checks on the answers are the same. A few practical steps differ, and these are practised on each tool.

---

Preparing each office to use the approved tool, with its own documents and rules, is the work of our ChatGPT and Claude courses for office teams. They run from basic to advanced level and take place remotely or in person, with travel anywhere in Italy. If the tool has not been chosen yet, the course starts from the contract and the data.

ChatGPT is a trademark of OpenAI, and Claude is a trademark of Anthropic. We name them only to identify the tools under discussion.

Sources

Consulted on 21 October 2026.

• Anthropic, Commercial Terms of Service
• Anthropic, Updates to Consumer Terms and Privacy Policy, 28 August 2025
• OpenAI, Enterprise privacy at OpenAI
• OpenAI Help Center, Data Controls FAQ