Give an AI assistant only the email context it needs

Start with the question, not the thread

Write down the specific help you want: for example, “Summarize the requested next steps” or “Draft a polite reply asking for a later deadline.” That gives you a way to judge which parts of the message matter. If the task is to explain a delivery date, the sender’s unrelated discussion of an earlier project may not be needed.

Next, select the smallest useful excerpt. Include the request and any relevant dates, conditions, or response you are answering. Add a short neutral description if the excerpt would otherwise be unclear, such as “This is a customer asking when a replacement will ship.”

Choose an excerpt or a redacted description

A direct excerpt can preserve exact wording when tone or phrasing matters. Copy only the relevant lines, and remove quoted history that does not affect the question. A redacted description is often enough when names, contact details, or account identifiers are not important to the answer.

Use clear placeholders consistently. For instance, replace a person’s name with [customer] and a date with [requested date] if the exact date is not necessary. Do not replace a key condition with a vague label: if the sender requires a reply before a particular deadline, keep the deadline or mark it as [reply deadline].

Task: Draft a brief reply asking whether the delivery can arrive later.
Relevant email: “Hi [customer], we can send the replacement by [delivery date]. Please confirm the shipping address before dispatch.”
Context: The recipient needs more time to confirm the address.
Request to assistant: Suggest a courteous reply asking whether dispatch can wait until [later date]. Do not promise a new delivery date.

Remove secrets and unrelated details

Never provide real passwords or one-time codes to an AI assistant. Also leave out recovery codes, payment credentials, full identity documents, and private account information unless the task genuinely requires a non-sensitive fact. For ordinary writing help, it usually does not.

Look for details that arrived earlier in the thread or inside a forwarded message: phone numbers, home addresses, account references, internal notes, and attachment contents. Delete what is irrelevant before sharing. If an assistant needs to know that verification is required, say “the account needs verification” rather than pasting the code or secret used to verify it.

Keep the prompt bounded

Tell the assistant what to do and what not to assume. Ask for a summary, a list of questions to clarify, or a draft that uses only the supplied facts. If a date, price, or commitment is missing, ask it to flag the gap rather than fill it in.

For example: “Draft a neutral reply based only on this excerpt. Do not invent a shipping date. If the message does not say whether the request is approved, leave that unresolved.” Clear limits reduce the chance that a polished-sounding draft will introduce a promise or detail that was never in the email.

Review the result against the original

Treat an AI-generated summary or draft as a starting point, not as confirmation that the message has been understood correctly. Check names, dates, recipients, requested actions, promises, and any financial information against the original email. AI results can be wrong, including when the supplied excerpt is accurate.

Before sending, make sure the reply answers the sender’s actual question and does not disclose details you removed from the prompt. If important context was omitted, add only that specific context and ask again; do not automatically paste the whole conversation.

A quick check before sharing

You can confirm the approach worked by checking that the assistant could address the stated task without needing secrets or unrelated history, and that its answer matches the facts you chose to retain. If it asks for more, provide only the missing relevant detail. BITTU Mail offers plan-dependent AI assistance, but availability can depend on the account’s plan and provisioning; regardless of the tool, review generated content before using it.

• Can I describe the task in one sentence?
• Have I kept only the excerpt or facts needed for that task?
• Have I removed passwords, one-time codes, and unrelated private details?
• Have I marked unknown facts instead of inviting a guess?
• Will I compare the response with the original before relying on it?