Claude is an AI assistant made by Anthropic. You type (or paste, or upload) what you need in plain English, and it drafts, summarizes, rewrites and works through problems with you. It is not a search engine, a bookkeeping system or an employee – it's a fast, capable first-draft partner that still needs a person to check its work.
This guide covers what that means in practice for a 5–50 person firm, using a fictional example: Northfield Bookkeeping, eight people, one office manager, a lot of client email.
What Claude is good at
Claude does its best work on tasks that are mostly reading and writing:
- Drafting. Client emails, proposals, engagement letters, job ads, policy documents. Northfield's office manager pastes in a messy request and gets a clean, polite reply to edit.
- Summarizing. Long email threads, meeting notes, a 40-page contract you need the gist of before a call.
- Rewriting. Making something shorter, friendlier, more formal, or easier for a client to follow.
- Sorting and structuring. Turning a pile of notes into a checklist, a table or a set of action items.
- Thinking out loud. Working through a pricing change or a tricky client conversation before you have it.
You can use it at claude.ai in a browser, in the desktop app for Mac or Windows, or in the iPhone and Android apps. Claude can also create documents, slide decks and spreadsheets inside a conversation, on every plan including Free.
What Claude isn't
Knowing the limits is what keeps you out of trouble.
It isn't always right. Claude's own help center is plain about this: it "can write things that might look correct but are very mistaken." It can produce a confident figure, quote or rule that turns out to be wrong. Anthropic's advice is not to treat it as a single source of truth and to check anything high-stakes. For a small firm, that means tax figures, legal wording and anything a client will act on gets checked by someone who knows.
It isn't your system of record. Claude doesn't replace your accounting software, CRM or file storage. It can read from some of those tools if you connect them, but the real data still lives there.
It isn't a decision-maker. It drafts. A person approves. Make that the rule from day one.
It doesn't automatically know your business. Out of the box, Claude doesn't know your clients, your prices or how you like emails to sound. You get much better results when you give it that context – which is what Projects are for (more below).
How it compares to ChatGPT
Claude and ChatGPT are the same kind of tool: general-purpose AI assistants you talk to in plain English. For everyday office work – drafting, summarizing, rewriting – both are capable, and the gap between them matters less than most comparisons suggest.
For a small team, the questions that matter more are:
- Which plan are you on, and what happens to your data? (See the next section.)
- Can you share setups across the team? In Claude, a Project holds instructions and reference files so every chat inside it starts with the same context. Projects are on every plan, but only Team and Enterprise members can share them with colleagues.
- Does it fit the way you already work? The best test is practical: take three real tasks from last week, remove anything sensitive, and run them through each tool. Pick the one whose drafts you'd send with the fewest edits.
What happens to what you type
This depends on the plan, so it's worth knowing before anyone pastes in client work.
- Free, Pro and Max (individual plans). Chats can be used to improve Anthropic's models if you allow it with the Model Improvement setting in your privacy settings. Incognito chats aren't used for training even with that setting on. Conversations flagged for safety review can also be used, as can any chat you rate with the thumbs up or down buttons.
- Team and Enterprise (business plans). By default, Anthropic doesn't use inputs or outputs from its commercial products to train models. The exception is feedback: if someone rates a chat or reports a bug, that conversation can be used. Owners can switch off chat rating for the whole organization in the admin settings.
The practical takeaway: if your team uses Claude for client work, put them on a business plan rather than personal accounts, and check the privacy settings once yourself.
What never to paste in
Even on a business plan, set a simple rule. At Northfield, it's this: if you wouldn't put it in an email to the wrong person, don't put it in Claude. In practice that means leaving out:
- Account numbers, card numbers and passwords.
- Government ID numbers, such as Social Security or tax ID numbers.
- Full client financial records when a summary or made-up example would do.
- Health, legal or HR details about clients or staff.
- Anything a client contract says must stay in specific systems.
Most tasks work just as well with names and numbers swapped out. "Draft a reminder to a client who's missed two document deadlines" needs no real client at all.
Where to start
Pick one repetitive writing job, set up a Project for it, and agree the review rule with your team before anyone uses it with clients. That alone covers most of what small firms get out of Claude in the first month.
Want this set up for your team, with the right plan, privacy settings and a one-page use policy? That's what the Claude Setup covers.