Inboxes for ChatGPT

Most account-based cold email with ChatGPT starts as a habit: paste the company, paste the person, ask for an email, send it from the mailbox you already have. Then ChatGPT gets connected to that mailbox and the habit becomes a workflow: it researches, writes, schedules and sends on its own. It works, which is the problem, because the mailbox it got connected to is on the company domain, the one your invoices use, and the pace was never set by anyone.

The emails are not the issue. One researched email per person is better outreach than any template, and it deserves to leave from somewhere built to carry it: separate domains, mailboxes with sender history, a daily cap each. Our mailboxes are ordinary Google Workspace, Microsoft 365 and Azure accounts with you as admin, so ChatGPT connects to them exactly as it connected to the company one. This page is about moving the workflow onto them and holding it to a pace. We sell the mailboxes, not ChatGPT, and this is not a review of it.

The honest shape: the buying happens in the portal, the sending happens from mailboxes you connected, and ChatGPT is the writer, the scheduler and, through Actions, the checker. The free API page lists every check a GPT can call.

• Separate domains, not the company one. The workflow this page describes usually starts connected to the company account. Move it before it scales. Our prewarmed orders come with the domain included, forwarded to your site, authenticated.
• Allow the connection as admin. Google Workspace and Microsoft 365 both let an admin control which outside apps an account may authorise, and you are the admin of every mailbox we deliver. If a connection is refused, that setting is where to look.
• Authentication confirmed, by the GPT if you like. SPF, DKIM, DMARC and forwarding are configured before handover; the authentication checker confirms it, and a GPT with Actions can run it for every domain in the fleet.
• Addresses verified. ChatGPT can research a person it cannot verify an address for, and a bounce is the loudest signal a new mailbox sends. Verify before the first send; the waterfall calculator finds the cheapest verifier order.
• A suppression list in the instructions. Replies, unsubscribes, customers. The suppression checker does it by hand; the workflow needs the same check before every send.

Written as what you are doing rather than as a click path, because ChatGPT’s settings move and a page full of yesterday’s button labels makes everything around it look stale.

1. 1Size the fleet before you write a single email. A list of 400 accounts at the paces below is about 27 Google mailboxes for one email each per day, not 3. The calculator gives the number; the Cold Email AI GPT will give you the same one.
2. 2Order in the portal. Prewarmed for sending from the day they land, fresh on your own domains for a third of the price at volume. Delivered within 48 to 72 hours of payment, authenticated, with you as admin of every mailbox.
3. 3Connect each mailbox through ChatGPT’s Google Workspace or Microsoft 365 connection. Sign in as the mailbox, authorise, repeat per mailbox. Azure tenants are Microsoft 365 mailboxes and connect the same way. The credentials come to you at delivery; keep them with you and never paste them into a chat.
4. 4Give ChatGPT the cap as a standing instruction, and schedule the runs. The figures below, per mailbox per day, spread across the working day, rotating across the connected mailboxes. A scheduled daily run that reads replies first, researches the next accounts, writes, sends within the cap, and stops.
5. 5Give the GPT the checks, not the credentials. Actions on the free API let it verify a domain before a send. It never needs a mailbox password for that, and the mailbox connection is made by you, so never give it one.

Personalisation improves the reply rate. It does nothing for the rate at which a receiving server will accept mail from one new mailbox, which is the number above. When the drafts outrun the fleet, the drafts wait or the fleet grows.

Prewarmed inboxes arrive with sender history; fresh inboxes are warmed by us for 14 to 21 days before handover. Our warmup then keeps running at 30 emails per inbox per day underneath ChatGPT’s sending for as long as you hold the inboxes. There is nothing in the agent to switch off, and nothing to add.

Replies falling on one mailbox while the rest hold, bounces climbing on one domain: take that mailbox out of the rotation, run the blacklist checker and the bounce analyzer on its bounces to tell the list from the domain, and message us on WhatsApp at +1 904 736 2539. Replacement inboxes are part of the package when deliverability drops, and a few spare mailboxes connected are what make a flagged one an inconvenience rather than a stopped workflow.

Prewarmed inboxes include a free prewarmed .COM or .CO domain and send from the day they land: $8.00 per inbox per month on Google and Microsoft, $7.00 on a two-month commit, $6.00 quarterly. Azure is billed per tenant, $69 monthly, $60 on two months, $50 quarterly, with up to 100 inboxes on one domain, which at capacity is $0.69, $0.60 and $0.50 per inbox. Fresh inboxes on your own branded domains, warmed for 14 to 21 days first, run $3.50 to $2.00 per inbox per month on Google, $3.00 to $2.00 on Microsoft 365, and $50 flat for an Azure tenant. Every price is at /price.md, which a GPT can read as plain text, and how it works covers everything between paying and the first send.

Can ChatGPT send cold email from WarmInboxes mailboxes?

Yes. Every mailbox we deliver is an official Google Workspace, Microsoft 365 or Azure account with you as admin, and ChatGPT connects to Google Workspace and Microsoft 365 accounts to read, draft, schedule and send mail. Connect the mailbox the way you would connect any Workspace or 365 account, and ChatGPT sends from it, through Google or Microsoft, not through anything of ours.

What is the Cold Email AI GPT?

A custom GPT we publish for cold email research and copy, free to use, at warminboxes.com/AI. It knows the pacing, the infrastructure and the deliverability rules on this site, and it will tell you the same daily limits this page does. It does not place orders and does not hold any account of yours.

Can a GPT run the deliverability checks?

Yes, through Actions. Every free tool on this site is also a keyless API, described at warminboxes.com/openapi.json, and a GPT given that file can check SPF, DKIM, DMARC, blacklists, MX and the rest before a send. The free MCP server exposes the same tools to clients that speak MCP.

Can ChatGPT order inboxes for me?

Order in the portal. Our keyed MCP server is documented for the clients listed on the developers page; we have not tested ChatGPT’s MCP connector against it, and we would rather say so than describe a path we have not run. An order takes two minutes in the portal, and the mailboxes arrive within 48 to 72 hours of payment.

How many emails a day can ChatGPT send per mailbox?

About fifteen a day on Google Workspace, eight on Microsoft 365, two on a full Azure tenant of 100, the same as for a person or a sequencer. ChatGPT will happily draft three hundred in an afternoon; give it the cap as a standing instruction, have it schedule sends across the day, and size the fleet so the instruction is easy to keep.

Does a highly personalised email still need a prewarmed mailbox?

Yes. A receiving server judges the domain and the mailbox before it reads a word. Prewarmed mailboxes arrive with sender history; fresh ones are warmed for 14 to 21 days first, and our warmup keeps running underneath at 30 emails per inbox per day for as long as you hold them.