An AI draft often needs the shape of a situation, not every detail in the original email. A name, account number, medical fact, or private explanation may add exposure without improving the wording. Before asking for help, decide what the draft must communicate, replace identifying details with useful placeholders, and keep any key separate from the prompt. Then check the draft against the original before restoring anything. This process can make a request easier to review, but it cannot guarantee privacy: consider the AI tool’s own data handling before sharing even a redacted message.
The useful bits
- Keep only details the recipient needs to understand the message.
- Use consistent placeholders that preserve roles and relationships.
- Check the draft against the original before restoring real details.
Start with the message’s job
Before editing, write down what you want the email to do: request a new appointment, explain a delay, ask for a decision, or decline an offer. Then identify the facts needed for that purpose. If the reader only needs to know that a delivery is late, its tracking number and the customer’s home address may not help an AI produce a polite, clear request.
Separate useful context from identifying detail. A role, approximate sequence of events, or general description may be enough. For example, “a customer has not received an order” can explain the situation without including their full name, address, or order reference. Keep exact facts only when the wording depends on them, and consider whether a less specific version will work.
Look for details that identify or expose someone
Review the message and any quoted thread, signature, attachment text, or forwarded content you plan to include. Personal details can appear in a sentence, but also in a subject line, email address, signature block, or copied reply. Confidential information may include internal plans, private negotiations, credentials, or details about another person.
Remove passwords and one-time codes rather than replacing them with a prompt-friendly version. Do not provide real passwords or OTPs to an assistant. Also avoid including information simply because it is present in the source. Ask: would the draft change if this detail were absent? If not, leave it out.
- Check names, addresses, phone numbers, account and order identifiers.
- Check quoted messages, signatures, attachments, and recipient details.
- Remove secrets and unrelated personal or business information.
Replace details with useful placeholders
Use short labels that preserve the meaning needed to draft the message. For example, replace a person’s name with [CUSTOMER], a colleague with [MANAGER], a company with [SUPPLIER], and a specific date with [DATE] if the exact date is not necessary. Keep each label consistent throughout the prompt so the assistant can follow who did what.
Avoid replacing every meaningful fact with the same generic label. If two people have different roles, give them different placeholders. If timing matters, you might retain a relative description such as “after the agreed delivery date” rather than a real date. Do not include a separate list mapping placeholders to identities unless there is a clear need; keeping that mapping outside the AI prompt reduces the chance of accidentally sharing it.
Original: Please tell Maria Lopez at maria.lopez@example.com that order 73104 will arrive on 18 June, not 12 June. Her delivery address is 14 Oak Lane. Redacted prompt: Draft a courteous email to [CUSTOMER] explaining that [ORDER] is delayed and is now expected on [NEW DATE], rather than [ORIGINAL DATE]. Apologize, state the revised timing clearly, and invite [CUSTOMER] to reply with questions. Do not invent a reason for the delay.
Give the assistant a bounded task
A clear request can reduce the temptation to paste in extra background. State the audience by role, the outcome you want, the tone, and any facts the draft must preserve. Also say what it must not add: for instance, do not invent a cause, promise compensation, or create a deadline that was not provided.
If you use AI assistance in BITTU Mail, availability depends on the plan and account provisioning. Whatever tool you use, consider its data-handling terms and your organization’s policies before sharing content. Redaction lowers the amount of information in a prompt; it does not make sharing confidential material automatically appropriate.
Check the draft before restoring anything
Treat the generated text as a proposal, not a verified record. Compare it with the original message and confirm that it has not changed the request, reversed who is responsible, or introduced an unsupported explanation. Check names, dates, recipients, promises, and financial information carefully. AI results can be wrong, including when the prompt is clear.
Restore only details the final email needs. Replace [CUSTOMER] with the intended name if appropriate, then verify spelling and the actual recipient separately. If a precise date or amount is necessary, confirm it from the original source rather than relying on the draft. Remove unused placeholders and check the subject, quoted text, and signature before sending.
Use a short final check
A quick review should confirm both privacy and accuracy. If the draft works without a sensitive detail, leave that detail out. If restoring a fact changes the meaning or introduces a commitment, pause and verify it with the appropriate source or person before sending.
- Did I remove unnecessary identifiers, secrets, and quoted private details?
- Are the placeholders consistent, and is any identity key kept outside the prompt?
- Have I checked the draft against the original and restored only necessary, verified facts?
About this guide
Prepared with AI assistance for the BITTU Mail editorial library. Check important details against your original messages and current account settings. Product access depends on your role, plan and provisioning state. Provider interfaces and requirements can change.
View current plans · Security information · Suggest a correction