Markdown edition

# Check decisions and owners in one Outlook thread with Claude

> Summarize one authorized Outlook conversation with Claude, then open every cited email before keeping decisions, open questions or owner assignments.

By BIG CHANGE Editorial

Published: 2026-10-19T20:02:34.865Z
Updated: 2026-10-19T20:02:34.865Z
Canonical: https://bigchange.ai/blog/claude-outlook-thread-decisions-owners-citations

![Conceptual charcoal illustration of one reader at a desktop monitor opening a message in an abstract email conversation.](https://bigchange.ai/api/media/file/outlook-cited-email-check-hero-v1.png)
AI-generated conceptual illustration by BIG CHANGE; no actual mailbox, add-in interface, model output or hands-on test is depicted.

Claude for Outlook can summarize a conversation into decisions, open items and owners, with citations that open the source emails. That gives you a starting point for a working note. The useful result comes after you check each cited message, especially where a suggestion could be mistaken for an agreement.

## The big change

- **What changed:** Anthropic's beta Outlook add-in can read a thread and cite individual emails in a summary of decisions, open questions and who owes what.
- **Why it matters:** You can review one long conversation by following each claim back to a message, then keep only the decisions and assignments the emails actually support.
- **What to watch:** A citation points to a message; it does not settle whether a later reply changed the plan. The add-in also requests broad delegated Microsoft Graph permissions, so the wording of a summary request is no limit on its mailbox access.

## Before you open the thread

Choose one real conversation that you own or are authorized to process. Note its subject and date range. Check your organization's rules before using privileged or regulated mail. This guide uses the direct Claude account route, which requires a paid Pro, Max, Team or Enterprise plan. Claude for Outlook is in beta. It supports Outlook on the web, new or classic Outlook on Windows with Microsoft 365, and Outlook on Mac with Microsoft 365. The mailbox must use Exchange Online; the current manual excludes on-premises Exchange, Outlook 2016/2019 perpetual editions, iOS and Android. [Anthropic's Outlook manual](https://claude.com/docs/office-agents/outlook) lists the supported clients and plans.

Install the official [Claude for Outlook add-in on Microsoft AppSource](https://marketplace.microsoft.com/en-us/product/office/WA200010724?tab=Overview) through a route your organization permits. Its **Get it now** button is the individual installation path. An organization can deploy it through **Microsoft 365 Admin Center → Settings → Integrated apps**. Deployment and Microsoft Graph consent are separate steps. Anthropic says a Global Administrator must grant tenant consent; otherwise a user may see **Need admin approval**. The requested delegated permissions include `Mail.ReadWrite`, `Calendars.Read`, `User.Read` and `offline_access`. `Mail.ReadWrite` is broader than reading one conversation. Ask your administrator to assess that grant and the organization's access controls before processing mail. This single-thread task does not require cross-app mode or the Excel, Word and PowerPoint add-ins.

## Make and check a thread summary

1. **Open the conversation.** In a supported Outlook client, open an email in the selected thread. Choose **Claude** in the message ribbon and sign in with the intended paid Claude account. If the button is hidden, open the reading-pane overflow menu, choose **Customize actions**, then check **Claude** under **Apps**. Confirm you are in the intended mailbox and conversation before asking for a summary.
2. **Request a bounded account.** In the Claude pane, use a prompt such as this. It is a proposed instruction for Anthropic's documented thread-summary feature, not a tested result:

> For this Outlook conversation only, list decisions that participants actually agreed, questions that remain open, and each next action with its owner. Cite the specific email for every item. Distinguish proposals from accepted decisions, and mark missing or conflicting evidence as unresolved. Do not draft or send mail, change a calendar item, or take another mailbox action.

3. **Open every citation.** Anthropic says selecting a citation opens its source message in Outlook. For each line of the summary, compare the email's sender, date and wording with Claude's claim. Read later replies for a correction or reversal. Count an owner only when a message assigns or accepts that action; a person mentioned in the thread is not automatically responsible. Check the claimed decision separately from its proposed next step. If an attachment is necessary to support a line, open and inspect it through a permitted route. The add-in's inline attachment support is limited: the manual names `.docx` and `.xlsx` on the open email, and gives different handling for PowerPoint and PDF files.
4. **Keep only verified lines.** In a dated private working note, record the thread subject and date range, the supported decisions, open questions, and owner/action items with their source messages. Save that note only if the mailbox owner and your data policy permit it. If an email does not support a claim, correct the line or leave it unresolved. Before using the note for a decision or correspondence, check the original messages again.

The expected output is a cited draft summary, followed by your checked note. This procedure has not been run against a mailbox for this article, so it makes no claim about completion time, coverage of every message, citation accuracy or a particular result.

## If the evidence or setup fails

If **Claude** is missing, check the supported client, add-in deployment and ribbon setting. **Need admin approval** calls for an administrator's consent review, not a workaround. If the pane fails, gives no citation, or seems to omit part of the conversation, inspect the actual thread and add-in state before retrying. Read missing emails yourself and withhold the affected line until it is supported. When messages conflict, record the conflict and the latest dated evidence; do not let the summary choose a winner without support.

Email and attachments from other people are untrusted input. They can contain instructions intended to steer the add-in. Ignore such instructions as authority for this task and review any proposed mailbox action before accepting it. The prompt above asks for analysis only. Anthropic says the add-in does not send mail or invites on its own, but its other features can prepare unsent drafts and calendar forms, so check the pane before acting.

Usage limits and billing depend on your paid plan; check the [current plan terms](https://claude.com/pricing) and your organization’s limits before running the task. Anthropic lists no fixed per-thread price or accuracy guarantee in the Outlook manual. The open item is read through Office.js, and thread retrieval uses Microsoft Graph. Anthropic says Graph calls and the access token stay in the browser, while mail content read under direct Claude sign-in becomes prompt data sent to `api.anthropic.com`. It says backend inputs and outputs are deleted within 30 days, subject to its stated exceptions. The add-in does not inherit an organization's custom retention setting. Chat history is stored separately in local browser storage and can be cleared in **Settings**; reinstalling the add-in alone does not clear it. Check these data terms and your organization's policy before processing sensitive mail.

For related Office tasks, see our guides to [checking workbook figures in a Word memo](https://bigchange.ai/blog/claude-excel-word-memo-figure-check), [reviewing a selected Word rewrite](https://bigchange.ai/blog/claude-word-selected-passage-tracked-changes-guide) and [updating a PowerPoint chart from Excel](https://bigchange.ai/blog/claude-excel-powerpoint-native-chart-update-guide). Each starts from a different file and produces a different result.

## Sources & further reading

- [Use Claude for Outlook](https://claude.com/docs/office-agents/outlook), Anthropic. Its beta manual documents thread summaries with per-email citations, supported clients and plans, installation, Graph consent, attachment limits, mailbox access, retention and prompt-injection risks. Checked 19 October 2026; the page shows no publication date or fixed add-in version. It does not establish that this particular prompt will produce a complete or accurate note.
- [Claude for Outlook on Microsoft AppSource](https://marketplace.microsoft.com/en-us/product/office/WA200010724?tab=Overview), Microsoft listing linked from Anthropic's installation instructions. Use the listing to identify the official add-in; tenant deployment and consent remain separate administrator decisions.
- [Claude pricing](https://claude.com/pricing), Anthropic. Current plan terms distinguish usage allowances from Enterprise seat and usage billing; check the terms and any organization limits before running a paid task. No per-thread charge is quoted.

## Sources

- [Use Claude for Outlook](https://claude.com/docs/office-agents/outlook) — Official beta manual checked 19 October 2026: supported clients and paid plans, thread summary and citations, installation, Graph permissions, privacy, retention and prompt-injection limits. No stated page date or current numeric add-in version.
- [Claude for Outlook on Microsoft AppSource](https://marketplace.microsoft.com/en-us/product/office/WA200010724?tab=Overview) — Official installation destination linked from Anthropic's manual; tenant deployment and Graph consent are distinct administrator steps.
- [Claude pricing](https://claude.com/pricing) — Official plan terms checked 19 October 2026: Pro, Max and Team have usage limits; Enterprise lists seat price plus usage at API rates with organization spend controls. No fixed Outlook-thread charge is quoted.