Summarize a Long Email Chain by Unresolved Decisions with ChatGPT

A long email chain can contain useful context while making it difficult to see what still needs a decision. Earlier proposals are quoted repeatedly, answers address only part of a question, and a later message may change one condition without resolving the overall issue. A chronological summary can reproduce that complexity rather than reduce it.

This guide creates an original decision-focused digest with ChatGPT. It uses fictional examples and current documentation checked on October 6, 2026. The output is a reviewable summary of supplied records; it does not send messages, infer acceptance from silence, or authorize the decisions it identifies.

Define the thread and information cutoff

Identify the discussion, desired reader, and cutoff date. State whether the digest should cover one thread or several related threads. Combining unrelated conversations can create apparent agreement between people who were discussing different subjects.

Collect the approved message export with sender, date, subject, and a stable message ID. Include relevant attachments or mark them unavailable. A message approving “the attached version” cannot be interpreted reliably if the attachment is missing.

The OpenAI projects reference describes organizing related sources and chats. You can use a bounded packet for this digest, but confirm that the supplied messages represent the intended scope. Do not claim to summarize an inbox that has not been provided or connected.

Separate original messages from quoted history

Identify repeated quoted blocks and link them to their original message where possible. Retain the original record, but avoid counting every quote as a new statement. Otherwise one early proposal may appear more widely supported simply because it was quoted in many replies.

Preserve additions made inside a quoted block. A reviewer may answer questions line by line, so deleting all quoted text can remove actual new responses. Mark uncertain authorship when formatting does not make the distinction clear.

Check date and timezone interpretation before arranging the sequence. A reply's timestamp establishes when the message was sent, not necessarily when a referenced action occurred. Keep those distinctions visible when they affect the latest status.

Extract decision topics and their questions

Ask ChatGPT to identify concrete choices, approvals, and unresolved questions. Group messages by decision topic rather than subject line alone. One thread can contain several decisions with different owners and dependencies.

For each topic, record the proposed options, constraints, current evidence, and actual decision authority if supplied. Separate a question asking for information from a request for approval. A factual answer may resolve the former without settling the latter.

Use a stable decision ID. This allows the digest to track the same issue as the discussion develops and prevents a renamed proposal from becoming a duplicate decision entry.

Build the decision register

The register should show why an item remains open and what would close it. A label such as “pending” is insufficient when readers do not know whether the gap is evidence, an owner response, or a final approval.

Record decisions already closed separately so the digest does not repeatedly reopen them. Preserve the closure evidence and scope; approval of one version may not settle a later revised proposal.

Ask for a digest with closure evidence

Supply the cleaned message packet, decision schema, and cutoff. Require references for the latest status and for any claim that a decision has been resolved. Do not let silence or positive tone substitute for documented approval.

Create a decision-focused digest of the supplied email records.
Group by unresolved decision, not by chronological message summary.
Distinguish new text from repeated quoted history and preserve inline replies.
For each decision cite the latest supporting messages and attachment versions.
Do not infer agreement from silence, courtesy, or incomplete responses.
Separate closed decisions with their closure evidence from open questions.
Return the register, concise digest, missing-source list, and messages not classified.

Review unclassified messages. Some may be irrelevant housekeeping, but others may contain a condition or late revision the model overlooked. Give them a recorded disposition rather than discard them automatically.

Examine a fictional partial answer

Suppose one message asks whether a workshop should move to Friday and whether the revised activity pack is approved. A later reply says Friday works. That response may address scheduling, but it does not establish approval of the activity pack.

The digest should show the two questions separately or preserve their distinct status in one topic. The activity-pack decision remains open unless another source supports closure. A summary saying “the workshop changes were approved” would overstate the evidence.

If a later message approves version R-2 while the latest attachment is R-3, record the version mismatch. Ask whether approval extends to the new revision instead of assuming that approval carries forward unchanged.

Write the reader-facing digest

Put the unresolved decision, latest supported state, missing condition, and next step in each entry. Keep background brief and link to source messages when the intended reader can access them. Use neutral attribution so competing views remain accurately represented.

Avoid reproducing every argument. Retain reasons that materially affect the choice, and place detailed history in the register. The digest's purpose is to help the reader act on open issues without searching the entire chain again.

Mark unknown owners and dates explicitly. Do not choose the most frequent sender as the decision owner or invent a deadline from the meeting's apparent urgency. Those may be useful clarification questions, but they are not established facts.

Review and maintain the decision record

Check each entry against the underlying messages and relevant attachments. Confirm that withdrawn proposals are not presented as current and that partial answers remain partial. Review the digest with the responsible project owner before using it as an operational record.

If the chronology is disputed, use the project timeline reconstruction workflow for that supporting analysis. If the digest must compare several proposed options, the decision memo guide addresses the resulting decision paper.

When new messages arrive, update the affected decision IDs and record the new cutoff. Preserve the earlier digest version so readers can understand which information was available at the time. Close an item only when its recorded question and conditions are actually resolved by the evidence.

Frequently asked questions