Gmail ↔ OpenAI integration

An AI agent on OpenAI sorts incoming Gmail, pulls the request details into fixed fields and prepares reply drafts that a person sends

A shared sales mailbox fills up faster than anyone reads it: orders, price requests, supplier invoices, complaints and spam all land in one list. Someone opens each letter, decides what it is, copies the phone and quantity into a CRM or a sheet and writes an answer. The Gmail ↔ OpenAI pair puts an AI agent on that mailbox: Gmail tells it about new mail, the model reads each letter and returns the same set of fields every time, the agent sorts the letter and leaves a reply draft in the same thread. A person reads the draft and presses Send. Below: what moves, which scenarios we cover and what we rely on in both systems.

Request an integration System page: Gmail →System page: OpenAI →

What moves between the systems

Gmail → OpenAIGmail notifies the agent when the mailbox changes, and each new letter with its text and attachments is handed to the model to read, without polling every minute.
OpenAI → GmailThe model returns a fixed set of fields for every letter: type (order, price request, invoice, complaint, spam), company, contact, phone, product, quantity and deadline.
OpenAI → GmailThe letter gets a Gmail label for its type, so the team opens «Orders» first and never sees spam at all.
OpenAI → GmailFor letters that need an answer, the agent leaves a reply draft in the same thread; it stays a draft until a person sends it.
Gmail → OpenAIVoice messages attached as audio files are transcribed, so a spoken order becomes text with the same fields.

What we connect

We agree which mailbox the agent reads, which labels it may set and which letters it never touches, for example the director's personal correspondence.
On the Gmail side the connection asks only for the permissions it needs: read, label and create drafts. It does not ask for permission to send.
On the OpenAI side we use your own developer account, so the usage is billed to you and the data stays under your account's terms.
We write the list of letter types and fields with your sales lead and test them on a hundred real letters from the last month before switching the agent on.
For the first two weeks every draft is read by a person, and we adjust the instructions on the letters the agent sorted wrongly.

Automation scenarios

Orders lost among newsletters

Before a manager scrolls through the mailbox in the morning and finds yesterday's order under a dozen newsletters.

After the order has the «Orders» label and a filled set of fields within minutes of arriving.

Price requests wait for a day

Before a client asks for a price, and the answer goes out the next day because nobody had time to write it.

After a draft with the right price list section is already in the thread, and the manager only checks the numbers and sends it.

Supplier invoices mixed with sales

Before invoices from suppliers sit in the sales mailbox and the accountant hears about them when payment is overdue.

After letters of type «invoice» get their own label with the amount and due date pulled into fields, and the accountant works from that list.

How it works technically

For those who want to check: each point links to the official documentation page it comes from.

Gmail can notify the connected application itself whenever a mailbox changes — Google offers this as the way to learn about new letters without checking the mailbox on a schedule.
Source: developers.google.com ↗ checked 2026-09-27
A draft in Gmail is an unsent letter with the system label DRAFT; when it is sent, the draft disappears and a letter with the label SENT appears — this is how the AI draft waits for the manager.
Source: developers.google.com ↗ checked 2026-09-27
Once a letter is created in Gmail it cannot be changed — the connection stores letters as they are and writes a copy into the CRM.
Source: developers.google.com ↗ checked 2026-09-27
Data sent through the OpenAI platform is not used to train or improve OpenAI models by default, unless the account owner explicitly opts in.
Source: developers.openai.com ↗ checked 2026-09-27
Structured Outputs makes the model always answer in a predefined structure, without missing a required field or inventing a value outside the allowed list.
Source: developers.openai.com ↗ checked 2026-09-27
File transcription: upload a finished recording and get the final text, or receive the text as a stream while the model is processing the file.
Source: developers.openai.com ↗ checked 2026-09-27

Limits

The limits come from both systems. The model can be wrong: it may misread a type or a quantity, so the fields and drafts are treated as proposals, and no letter is sent without a person. OpenAI limits the number of requests per minute by account level, which grows with usage, so a very busy mailbox on a new account is processed with a short queue. Gmail does not allow a letter to be changed once it exists, so the agent never edits incoming mail: it only adds labels and creates drafts next to it.

How to start

Three steps from a first message to work on your project.

01
Leave a request
Describe the task in the form — in text or by voice. Budget and timing are optional.
02
We review and estimate
A developer studies the brief, asks about anything unclear and sends an estimate.
03
We start
Once scope and price are agreed, the work begins.

Leave a request

or write to info@coreviaflow.space

Live project We built and run Gemalli — a B2B export platform with producer pages, a CRM and a deal chat. Read the case →

Questions

Can the agent send a letter to a client by itself?

No. It creates drafts, and the connection does not even have permission to send. A person reads the draft and presses Send in Gmail.

Will our correspondence be used to train OpenAI models?

By default, no: OpenAI states that data sent through its platform is not used for training unless the account owner opts in. The agent also reads only the mailbox and labels we agreed on.

Can the fields go straight into our CRM?

Yes, as a separate step: because the model always returns the same fields, they can be written into a CRM or a sheet. We usually switch that on after the first two weeks, when the sorting is checked.