WORKFLOWS / INBOX AND COMMUNICATION
Triage the inbox and prepare replies.
Work through incoming email with a clear next action for each message. AI helps identify what needs attention, find the information needed to reply, and prepare drafts. You control what gets sent and what leaves the inbox.
WHAT THIS WORKFLOW PRODUCES
An inbox ready for action.
Keep the work in the mailbox. Use a short review list when helpful, with links back to the original threads. A separate executive dashboard is optional. The aim is to stop rereading messages without moving them forward; an empty inbox is useful only if important requests remain visible somewhere appropriate.
INBOX REVIEW · TUESDAY 3 SEPTEMBER
LAST CHECKED 10:04 AM
THE OPERATING METHOD
Understand the workflow without AI.
This is the method itself, runnable with a mailbox and a short list. AI is introduced only after the method makes sense on its own.
Read the thread
Check the latest request and whether someone has already answered it.
YOUChoose the action
Reply, delegate, ask for a decision, or file.
YOUFind the answer
Check the relevant document, owner, or earlier agreement.
YOUReview the response
Confirm facts, recipients, tone, and authority to commit.
YOUFinish the handling
Send or delegate, then file completed correspondence and keep unresolved work visible.
YOUAn email asking for a decision is still unresolved when it has been summarized or moved into a folder.
WHERE THE FACTS LIVE
Use the systems where the work already lives.
These are examples rather than a required tool set. Choose connections your organization approves and check what each can read or change. Permission to open an executive's mailbox in one application does not automatically transfer to another. Add business context only when replies regularly need it and a suitable connection is available. Named tools link to their pages in the library.
INFORMATION NEEDED TO REPLY
OPTIONAL BUSINESS CONTEXT
THREE WAYS TO RUN IT
Choose how much machine.
The levels describe how much work AI takes on. A connection can appear at Level 1 when that is the practical starting point. Level 3 adds a bounded agent: it chooses among approved information-gathering steps and prepares the next action. Every level ends with a person deciding what gets sent. The tools named are examples. Use what your organization already runs, provided the connection is supported and approved.
LEVEL 1
Sort the next actions.
Connect the mailbox to an approved AI assistant and ask it to review a defined group of messages. It identifies requests, deadlines, and decisions. You handle the messages in your usual email application.
Use this version when you want help getting through the inbox without setting up a recurring automation.
WHAT THIS VERSION REDUCES
Rereading threads and working out which messages need a response.
WHAT A PERSON STILL DOES
Corrects priority, replies, delegates, and files.
MAIN LIMITATION
AI only sees the messages and context available through the connection. It can mistake a polite mention for a request or miss an agreement made elsewhere.
CONNECTION NOTE
Claude can read and draft in Gmail, and its Microsoft connection can open a shared or delegated Outlook mailbox you are authorized to use. What it sees depends on the account you connect and your organization's settings.
Build this version, step by step 5 steps, one action each. Prompts carry a Copy button.
Connect the AI assistant to the right mailbox.
EXAMPLE
Claude with Gmail, or Claude's Microsoft 365 connection for an Outlook mailbox you are authorized to open. Start with read-only access where the tool offers it; otherwise make sure every change needs your approval.
Tell it which messages to look at.
THE SCOPE
Business messages that arrived in the inbox since the last review, including the ones already marked as read. A read message is not a handled message. Say what to leave out, such as newsletters.
Ask for a list of next actions.
Review these inbox threads. Group them into Reply, Delegate, Decision needed, and File. For each, link the original message, state the request in one sentence, and quote any explicit deadline. Check the latest replies before calling something unanswered. Flag missing context. Do not send, move, delete, or unsubscribe.
Check the list against the inbox.
THE CHECK
- The important threads are opened and the suggested action makes sense.
- A few messages marked for filing are spot-checked.
- The assistant has said whether it hit a search limit or could not read something, rather than claiming it covered everything.
Handle the messages in the mailbox.
THE HANDLING
Reply or delegate from the mailbox itself. A request stays visible in the inbox until it has an owner and a next step.
LEVEL 2
Prepare replies with the right context.
Let AI do the research behind the reply. For a selected thread, it checks approved documents or records, separates confirmed facts from missing information, and prepares a draft for review.
Use this version when writing the response is easy, but finding a reliable answer takes time.
WHAT THIS VERSION REDUCES
Searching for supporting information and starting replies from scratch.
WHAT A PERSON STILL DOES
Resolves conflicting information, approves commitments, checks recipients, and sends.
MAIN LIMITATION
A current-looking document can still be wrong. AI must surface conflicting dates or terms instead of quietly choosing one.
ATTACHMENTS ARE NOT READ
Claude's Gmail connection sees that an attachment exists and what it is called, but not what is inside it. Upload the file separately or reach it through a file connection. Do not say an attached report has been read when it has not.
Build this version, step by step 5 steps, one action each. Prompts carry a Copy button.
Connect a small set of sources that answer the usual questions.
EXAMPLE
Gmail plus the approved Google Drive folder that holds the documents customers and suppliers usually ask about. Outlook and SharePoint work the same way through Claude's Microsoft connection.
Write down the reply rules once.
THE SAVED RULES
The tone, the signature, who owns which kinds of request, and which commitments need approval before they are promised. Old replies show the style; they do not prove that an old price or policy still applies.
Ask the AI to prepare one reply.
Prepare a reply to this thread using the approved sources. Show me the evidence separately from the customer-facing draft. Flag conflicting information and anything requiring a new commitment. If you cannot establish a fact, leave it for my decision rather than inventing an answer. Do not send.
Keep the working notes out of the email.
WHAT GOES WHERE
Source links, doubts, and internal comments stay in the assistant's response to you. The draft in the mailbox holds only the email the customer will read.
Review in the mailbox right before sending.
THE CHECK
- The thread has no newer reply that changes the answer.
- The draft sits in the right conversation, with the right recipients and attachments.
- If the connection made a new message instead of a reply draft, the reviewed text is copied into Reply in the mailbox.
CAPABILITIES CHECKED AGAINST OFFICIAL DOCUMENTATION
Claude's Gmail connection documents attachment metadata only; attachment contents are not read. Source, checked September 5, 2026.
LEVEL 3
Run a recurring inbox review.
Run the review when eligible messages arrive or at agreed times. The agent decides whether it needs a document, an earlier reply, or a human answer. It prepares a draft inside the conversation and applies only the handling rules you have explicitly approved.
Use this version when similar messages arrive regularly and someone can own the review and upkeep.
WHAT THIS VERSION REDUCES
Repeated inbox scanning, routine lookups, and draft preparation.
WHAT A PERSON STILL DOES
Sends replies, makes sensitive decisions, and approves changes to handling rules.
MAIN LIMITATION
A recurring agent can repeat a bad rule at scale. Start with drafts and suggested labels before permitting any automatic filing.
WHY N8N HERE
n8n's Gmail node can create a draft with a thread ID, so the draft appears inside the conversation, where the assistant will look for it. n8n Cloud is a paid subscription, and the self-hosted edition is free. Make can run the same shape, but its standard Gmail module creates a loose draft rather than a reply in the thread. For Outlook, use n8n's Microsoft Outlook node and its draft and folder actions.
BEFORE EXPANDING
Compare review time, missed important messages, and drafts needing substantial correction. Do not judge the workflow by the number of messages it moves.
Build this version, step by step 8 steps, one action each. Prompts carry a Copy button.
Choose how the workflow starts, and what it ignores.
TWO WAYS TO START
Either the Gmail trigger watches the inbox, or one label, and runs once per new message, or a schedule runs it at 10 AM and 4 PM on working days. Tell it what to skip: newsletters, system mail, anything marked personal.
Write a short handling guide and give it to the agent.
WHAT THE GUIDE HOLDS
Who the priority contacts are, who owns which topics, how drafts should sound, and which topics the agent must leave alone. Keep one copy, paste it into the AI Agent node's system message, and update it there. Do not feed the whole email archive into every run.
Give the agent one place to look things up.
THE LIMITS
The AI Agent node gets the current thread and one tool: search the approved Google Drive folder. The folder and the account are fixed in the credential and the tool settings, because a sentence in the prompt is not an access restriction.
Have it write the draft inside the conversation.
THE DRAFT STEP
Use the Gmail node's Draft: Create action and pass in the Thread ID from the trigger, so the draft sits under the message it answers. In the same run, add a label such as "AI draft ready". This first version gets no tools for sending, deleting, or unsubscribing.
Keep a record so it never drafts the same message twice.
THE RECORD
One row per message ID in a Google Sheet or in n8n's built-in Data Table, holding the last version processed, the draft ID, and the review state. The agent checks the row before it drafts. If a person has edited the draft, it flags the new information instead of overwriting their work.
Make the exceptions show up as review items.
WHAT BECOMES A REVIEW ITEM
A missing attachment, conflicting instructions, a source it could not open, or a decision only the executive can make. A failed lookup produces the label "Review incomplete", never a confident draft. Nothing written in an email can give the agent new tools, new recipients, or new rules.
Test it before you trust it.
THE TEST SAMPLE
- A routine reply.
- A sensitive request.
- A thread that was already answered.
- An email that contains a misleading instruction.
Nothing should send, the draft should sit in the right thread, and running it twice should not create a second draft.
Change the rules on purpose, never by accident.
HOW THE RULES CHANGE
Look at the recurring corrections together, then change the written handling guide once someone approves the change. The guide is maintained context; the agent does not learn from private correspondence on its own.
CAPABILITIES CHECKED AGAINST OFFICIAL DOCUMENTATION
n8n's Gmail node documents Draft: Create with a Thread ID field, so a draft is attached to an existing conversation. Source, checked September 6, 2026.
n8n documents a tool-using AI Agent node. Source, checked September 6, 2026.
n8n's Microsoft Outlook node documents draft and folder operations for an Outlook mailbox. Source, checked September 6, 2026.
n8n Cloud is a paid subscription; the self-hosted community edition is free. Source, checked September 6, 2026.
Make's Gmail integration lists Create a Draft among its actions; the module has no thread field. Source, checked September 6, 2026.
WORKS EVEN BETTER WITH
Agree on who can send which replies. Put unresolved executive decisions in the daily brief, and keep a clear review point for AI drafts.
Build a daily executive brief → | Working Together playbook → | What does an AI executive assistant take off your plate? →