Use AI to build an email timeline that separates events from plans

Decide what counts as evidence

Before asking for a timeline, set a boundary: use only the messages you provide, and do not fill gaps with assumptions. Ask the assistant to distinguish a completed event from a confirmed future plan, a tentative suggestion, and an unresolved or conflicting claim. A message saying “I can send the draft Friday” is a commitment or plan, not proof that the draft was sent. A proposed meeting time is not a confirmed meeting unless the thread supports that interpretation.

If you have only part of a conversation, say so. Missing replies, calendar changes, or messages in another mailbox can change the account of events. The timeline should reflect the evidence in the supplied mail, not claim to be a complete record.

Ask for dates and message pointers

Give the assistant the relevant thread or messages and request one dated entry per event or plan. Require the date as written or clearly established in the message, and have it flag missing years, ambiguous dates, or time-zone uncertainty rather than silently resolving them. Ask for the sender, message date, subject, and a short locator such as a distinctive phrase from the relevant passage.

A pointer makes verification practical. A date and summary without a way to find the supporting message can be difficult to check in a long thread. If the mail view gives you a subject, sender, and date, use those details to locate the original; do not treat an AI-generated pointer as a link or as proof.

Prompt:
“Create a chronological timeline using only these messages. Label each entry COMPLETED, CONFIRMED PLAN, PROPOSAL, or UNCLEAR. Include the date, what the message actually supports, sender, message date, subject, and a short phrase to help me find the passage. Mark conflicting or incomplete dates; do not infer that a plan happened.”

Keep the timeline’s categories separate

A helpful timeline makes status visible at a glance. A completed event describes something the message says occurred: “The package arrived,” for example. A confirmed plan records an agreed future action or time, while a proposal records an idea awaiting acceptance. Use “unclear” when the wording does not establish which applies.

Preserve the distinction between a promise and its completion. “I’ll send the revised file tomorrow” supports a planned action. A later message saying “Attached is the revised file” may support that it was sent, but check whether the attachment is present and whether it is the file in question. The assistant should not upgrade a plan merely because its date has passed.

Work through a fictional thread

Suppose a thread contains a proposed call, a reply accepting a time, and a later note saying the call needs to move. A careless summary might list the original time as an event. A careful timeline records the proposal, then the acceptance as a confirmed plan, then the change as a new message that affects the plan. Unless a message says the call took place, its completion remains unconfirmed.

Keep the source details close to each entry so you can inspect the sequence. The sample below uses fictional addresses and bracketed fields; it illustrates a format, not evidence about any real exchange.

PROPOSAL — [date in message]: [sender] suggested a call at [time]. Source: “[subject],” [message date]; find “[distinctive phrase].”
CONFIRMED PLAN — [date in message]: [recipient] accepted [time]. Source: “[subject],” [message date]; find “[distinctive phrase].”
CHANGE / UNCLEAR — [date in message]: [sender] asked to move the call; no replacement time is confirmed in the supplied messages. Source: “[subject],” [message date]; find “[distinctive phrase].”
Fictional participants: [person] <alex@example.com> and [person] <sam@example.com>.

Verify the original, not just the summary

For each entry, open the cited message and check the full passage in context. Confirm who sent it, when it was sent, who received it, and whether a reply changes or qualifies the statement. Compare the date in the message with the date the assistant assigned; watch for year omissions, relative wording such as “next Tuesday,” and differences between a message’s sent date and the event date.

Check names, amounts, deadlines, promises, and attachments especially carefully. A short excerpt can omit a condition in the sentence before or after it. If the source message does not support the entry, correct the timeline or label the point uncertain. Keep the original messages available rather than relying on the generated summary alone.

A final check before sharing

Review the timeline from top to bottom and ask: Does every claim have a source? Are completed events clearly distinct from future plans and suggestions? Are contradictions shown rather than smoothed over? Are dates precise only where the messages make them precise? These checks help prevent a neat-looking chronology from overstating what the thread establishes.

AI assistance is plan-dependent, and its output can be wrong. If the feature is available to your account, treat the result as a draft for organizing evidence, not as a definitive record. You can confirm it worked when each entry leads you back to a message that supports its wording and status—and any gap remains visibly unresolved.

• Every entry has a sender, message date, subject, and findable passage.
• Proposals and plans are not presented as completed events.
• Conflicts, missing details, and unsupported claims remain marked uncertain.