AI at Work

7 mistakes teams make in their first month with Claude

The seven mistakes are: using personal accounts for work, treating Claude as a search box, working without shared projects, switching on every connector at once, having no checking habit, training once and stopping, and leaving finished work in chats instead of the document system. Each has a simple fix: a business plan, proper briefs, shared projects, connectors added for a reason, a routine for checking, short follow-up sessions and an agreed place where work is saved.

1. Using personal accounts for work

People often start on their own free or Pro account. Work on a personal account sits outside the business's control, and on the individual plans each person decides whether their chats may be used to improve Claude. When someone leaves, their chats and projects go with them.

Instead, move work onto Claude Team or Enterprise in the business's name. Anthropic states that business plans do not train models on your content by default. On Team, an Owner can verify the company's email domain and stop new personal accounts being created on it. Claiming existing personal accounts is an Enterprise feature, so ask people to move their own work across.

The plan choice is covered in Claude Team or Claude Enterprise: which does a business need?

2. Treating it as a search box

A one-line question gets a general answer. Teams who stop there decide Claude is no better than a search engine.

Instead, brief it like a new colleague: give it the document, say who the result is for, what good looks like and what it must not do. Ask it to flag gaps rather than fill them.

Ten worked examples are in 10 Claude prompts for office work.

3. Working without shared projects

When everyone uploads the same price list or style guide to their own chats, each person gets slightly different results and nobody knows which file is current.

Instead, set up a shared project for each recurring job, with written instructions and the reference documents it needs. On Team and Enterprise a project can be open to everyone in the organisation or kept private to invited members, who can be given view or edit access. Chats inside a shared project stay private unless the person shares them.

4. Switching on every connector

It is tempting to connect mail, calendar, files and every other service on day one. Each connector widens what Claude can read, and some let it send, edit or delete. Nobody then knows what is on or why.

Instead, start with the document system the business already uses, read-only, for a pilot group. Add others when someone names a real task, keep write actions on Needs approval and keep a short list of what is switched on.

The order we suggest is in Claude connectors and MCP explained.

5. Having no checking habit

Claude's drafts read well, which makes mistakes easy to miss: a figure from the wrong period, a date that was a plan rather than an event, a clause that changed in a schedule nobody opened.

Instead, agree what gets checked for each kind of work. Figures against the source, commitments against what was agreed, and every claim that a document says something against the passage itself. Write it down as part of the shared project's instructions.

A method for document work is in can you trace the answer back to the files?

6. Training once and stopping

A single introductory session produces a burst of use, then most people go back to how they worked before. The questions that matter come up later, on real work.

Instead, follow the first session with short ones built around the next real example of each job. Ask what went wrong, fix the instructions and share what worked with the rest of the team.

7. Leaving work in chats

A good draft that stays in a chat window is not part of the record. Colleagues cannot find it, it is not in the client file and it may not be the version that was sent.

Instead, agree that finished work goes back into SharePoint, Google Drive or your document management system, under your normal naming. Claude is where the draft starts, not where the record lives.

A better first month

Most of these fixes cost nothing but a decision. Make them in the first week and the rest of the month goes on real work rather than workarounds.

A short, practical rollout of this kind is what our Claude training in Ireland is built around.

Common questions

Can we stop staff using personal Claude accounts for work?

Partly. On Team, an Owner can verify the company domain and stop new personal accounts being created on it. Taking over existing personal accounts on the domain is an Enterprise feature. A short written policy covers the rest.

How many connectors should we switch on at the start?

One or two: the document system the business already uses, read-only, for a pilot group. Add more when someone names a real task that needs one.

How much training does a team need in the first month?

A hands-on first session on real work, then short follow-ups around the next real example of each job. Repetition on the team's own tasks does more than a longer first session.

For help putting this into practice, see Claude implementation and training for Irish businesses.

Sources and product references

Checked 16 November 2026. Features and access can change, so check what your own licence includes.

  1. Anthropic: Claude plans
  2. Anthropic: claim and migrate accounts on your domain
  3. Anthropic: manage project visibility and sharing
  4. Anthropic: use connectors to extend Claude's capabilities
  5. Anthropic: model training on individual plans

Editorial guidance informed by Acuity's work with teams. Examples are fictional and client materials stay private. To record who owns an AI use and when it is reviewed, see AI Register.