Claude

How to use Claude for business without leaking client data

Updated May 2026

Every business owner who starts using Claude hits the same wall around week two: you want to use it for real work — proposals, client emails, internal SOPs — but you're not sure what you can safely paste in. Client names. Contract terms. Financial figures. Internal processes. The tool is sitting right there, and you're second-guessing every prompt.

That hesitation is correct. But the answer isn't to stop using Claude. The answer is to understand exactly what the risks are, set the tool up properly, and build a simple set of rules for what goes in and what stays out.

This article covers the privacy settings that matter, the substitution technique that lets you use Claude on sensitive work without exposing anything real, and five prompts you can run today for actual business tasks.

What actually happens to your prompts

When you type something into Claude, that input goes to Anthropic's servers. By default, Anthropic may use conversations to improve its models — unless you turn that off. The first thing to do before using Claude for any client work is check this setting.

Go to claude.ai/settings. Look for the privacy section. There will be an option about whether your conversations can be used to train the model. Turn it off. This does not make Claude more private in every sense — your conversations still pass through Anthropic's infrastructure — but it means they are not fed into training pipelines.

For most solo operators and small teams, this setting change, combined with the substitution technique below, is enough to use Claude for client-adjacent work safely. If you are in a regulated industry — healthcare, legal, financial services — you need to look at Claude's Team or Enterprise plans, which offer stronger data handling commitments including zero training on your data by default and BAAs where relevant.

The substitution technique

This is the single most useful thing in this article. You do not need to paste real client names, real figures, or real contract language into Claude to get useful output. You replace the sensitive details with placeholders, run the prompt, then fill in the real information yourself in the output.

Here is how it works in practice. Say you need to write a proposal follow-up email after a meeting with a client. Instead of:

"Write a follow-up email to Sarah Chen at Meridian Capital after our meeting yesterday about the Q3 audit scope..."

You write:

"Write a follow-up email to [CLIENT NAME] at [COMPANY] after a meeting about [PROJECT TYPE]..."

The output is identical in quality. You paste the email into your email client and swap in the real details. Claude never sees the sensitive information. This takes about five extra seconds per prompt and eliminates most of the privacy risk for ordinary business tasks.

Five prompts for real business work

Each of these uses the substitution technique. Run them as written, then fill in the specifics yourself.

This first prompt handles client proposals. Use it after an initial conversation when you need to move fast.

I need to write a project proposal for a [TYPE OF PROJECT] for a client in [INDUSTRY].
The scope is [2–3 sentence description of the work, using general terms].
The timeline is [X weeks/months].
Deliverables include [general list without specific names or figures].

Write a professional proposal structure with: an executive summary, scope of work section,
timeline, and a "next steps" section. Do not include pricing — I will add that separately.
Tone: direct, professional, no filler.

After Claude responds, review the structure and replace any generic references with your real client's language. Add pricing manually. Never paste your actual rate sheet or fee schedule into Claude.

This second prompt is for drafting SOPs — standard operating procedures — for internal processes. It keeps your actual workflow private but gets you a usable document structure fast.

Write a standard operating procedure for [PROCESS NAME] at a [TYPE OF BUSINESS].
The process involves: [3–5 steps described in general terms].
Who does it: [role title, not a real person's name].
How often: [frequency].
Format: numbered steps with a brief explanation for each. Include a "common errors" section at the end.

This third prompt handles difficult client emails — the ones you delay writing because you're not sure how to say the thing you need to say.

Write a professional email from a [YOUR ROLE] to a [CLIENT ROLE] about [SITUATION — describe
in general terms without names or confidential details].
The key points to communicate are:
1. [Point one]
2. [Point two]
3. [Point three]
Tone: [direct/warm/firm — your call]. Keep it under 200 words.

The fourth prompt is for meeting preparation. You have a meeting coming up and want to think through it in advance.

I have a meeting with [CLIENT ROLE — e.g. "a prospective enterprise client"] about [TOPIC —
e.g. "renewing a service contract"].
Help me prepare by:
1. Listing the 5 most important questions I should ask
2. Identifying 3 likely objections and a response to each
3. Naming 2 outcomes I should aim to close before the meeting ends
No names, no company details needed. Just the strategy.

The fifth prompt is for internal hiring — writing a job description when you need to bring someone in without revealing your org structure or compensation details.

Write a job description for a [ROLE TITLE] at a [TYPE OF BUSINESS].
Key responsibilities: [list 4–5, described generally].
Skills required: [list 4–5].
Do not include salary range, company name, or location — I will add those manually.
Tone: direct, no corporate fluff. This should sound like a real operator wrote it.

What never goes into Claude — the short list

Even with the training setting off and the substitution technique in place, there are things that should not go into Claude's standard chat interface under any circumstances.

• Real client names combined with financial figures or sensitive project details
• Full contracts, NDAs, or legal documents — even with names redacted
• Employee personal information (salaries, performance reviews, personal details)
• Passwords, API keys, or system credentials of any kind
• Patient information if you work with healthcare clients
• Tax returns, full financial statements, or banking details

If you need Claude to work with documents of this type, you need a different setup — specifically Claude's Team plan with data handling agreements in place, or a workflow where the document is heavily redacted before it ever goes near an AI tool.

Claude for Teams: when to upgrade

The standard consumer version of Claude (claude.ai) is adequate for most solo operators and small business owners who use the substitution technique consistently. But if any of these apply to you, it is worth looking at Claude's Team or Enterprise plans:

• You have employees or contractors prompting Claude with client information
• You work in a regulated industry with documented data-handling requirements
• You need audit trails of who ran what prompts
• You want contractual commitments about data storage and retention

These plans offer zero-retention options, stronger legal commitments, and admin controls. The business case is simpler than it sounds: if you get audited or face a client dispute, you want to be able to say you used the enterprise-grade tool with the appropriate data agreements — not the consumer chatbot.

Building it into your workflow

Privacy habits with AI tools don't stick as rules. They stick as workflow steps. The easiest way to make this permanent is to keep a simple "safe substitution" prompt template saved in a document or notes app. Before you type anything into Claude, you spend 30 seconds replacing real names, figures, and identifiers with placeholders. Then you run the prompt. Then you fill in the real details in the output.

After two weeks, this is automatic. The tool becomes genuinely useful for client-facing work, and you're not second-guessing every prompt.

The goal is not zero risk. The goal is appropriate risk management for your context — solo operator, small team, or regulated industry. The framework above covers the first two. If you're in the third, the upgrade path is clear.

Want the full framework? Claude for Business Owners covers privacy-first setup, 20 ready-to-run business prompts, and the workflows for proposals, hiring, client emails, and SOPs — with a dedicated chapter on data safety. Also worth reading: Claude for the Curious if you want a full picture of what the tool can and can't do before building it into your workflow. Or start free: get AI for the Curious — a plain-English tour of all the major AI tools with prompts you can try today.