Separate facts from interpretations when asking AI about an email

Set the boundary before asking

Give the assistant a narrow task: identify what the email explicitly says, then list possible interpretations in a separate section. Tell it not to present an interpretation as something the sender confirmed. If it cannot support an interpretation with wording from the message, it should say so rather than fill the gap.

For example, ask: “Separate explicit statements from possible interpretations. For each interpretation, quote or identify the wording that led to it. Mark uncertainty, and do not add facts that are not in the email.” This instruction is useful whether you paste the email into an AI assistant or use plan-dependent AI assistance in BITTU Mail, where available. AI output can be wrong, so the original message remains the reference.

Keep the source message in view

Use the complete relevant message, not a sentence pulled out of context. A subject line, earlier reply, or sentence later in the email may change how a phrase should be read. Preserve important details such as names, dates, recipients, requests, and conditions. If you remove sensitive or irrelevant information before sharing text with an assistant, make sure the edit does not change the meaning.

Do not provide real passwords or one-time passcodes to an assistant. Once you have an answer, return to the original email and compare the assistant’s wording with what is actually there. A confident tone in the output is not evidence that an inference is correct.

Use a two-column reading

A simple structure makes the distinction easier to inspect. The first column is limited to statements the sender directly makes. The second contains interpretations, each labelled as a possibility rather than a fact. A third note can identify what information is missing or what question would resolve the uncertainty.

Consider this fictional message: “We have not received the revised draft. Could you send it before the planning meeting? We can discuss the timeline then.” The explicit statements are that the sender has not received the draft and asks for it before the meeting. The sentence does not directly say that the current timeline has changed. The sender may want to discuss timing, but the reason and intended outcome are not stated.

Prompt: Separate direct statements from possible interpretations. Identify the words supporting each interpretation, mark uncertainty, and list any missing information.

Email: “We have not received the revised draft. Could you send it before the planning meeting? We can discuss the timeline then.”

Explicit: The sender says the revised draft has not arrived and asks for it before the planning meeting.
Possible interpretation: The sender may want to review the timeline, but the message does not say that a deadline has changed.
Open question: Which timeline item does the sender want to discuss?

Audit every interpretation against the email

Read each interpretation and locate the words that support it. Ask yourself whether the email states the idea directly or whether the assistant has connected separate details. If the support is only a vague phrase, treat the interpretation as uncertain. If you cannot find supporting wording, remove it from your understanding or ask the assistant to explain the gap.

Check names, dates, recipients, promises, and financial details especially carefully. For example, “I’ll try to send it Friday” is not a firm promise to send it Friday. A summary that changes “try” into “will” has turned a qualification into a commitment. Verify quoted wording in the original rather than relying on the assistant’s paraphrase.

• Does the email say this directly, or is it a guess?
• Which exact phrase supports the interpretation?
• Have any qualifiers, conditions, or dates been changed?
• What remains unknown, and could a question clarify it?

Write a reply that does not assume intent

Base your reply on confirmed statements. Acknowledge what the sender explicitly asked for, then ask a neutral question about anything unresolved. Avoid writing as though an inferred motive, decision, or deadline is settled.

For the example above, a cautious reply could be: “I’ll send the revised draft before the planning meeting. Which part of the timeline would you like to discuss?” This confirms the request without assuming that the sender has changed a date or rejected the draft. Before sending, check that your reply does not add a promise or commitment you did not intend to make.

To: [recipient]@example.com
Subject: Re: Revised draft

I’ll send the revised draft before the planning meeting. Which part of the timeline would you like to discuss?

[Your name]

Confirm the final reading

Before using an AI-generated interpretation, compare it with the original one last time. Confirm that explicit statements are accurate, uncertainty is still visible, and the proposed reply responds to the actual request. If the message is ambiguous on a point that matters, ask the sender directly rather than choosing the most convenient interpretation.

The goal is not to make AI decide what someone meant. It is to use it to organize possibilities, then keep the difference between evidence and inference clear. If you cannot point to the words that support a claim, do not treat that claim as confirmed.