Gmail sending limits for cold email: the published cap and the safe rate

The question is usually put as "how many cold emails can I send from a Gmail account a day". The honest answer has two halves, and running them together is how domains get burned. Google publishes a daily ceiling, and that ceiling is real. The rate at which cold mail keeps reaching the inbox is a different and much smaller number, and Google does not publish it because it is not Google that decides it.

What Google actually permits

These are the platform figures. They are published, and you can treat them as facts about what the platform will accept.

Two details matter more than the headline. The window is rolling rather than calendar-based, so hitting the cap at 4pm does not reset at midnight. And free Gmail cannot be authenticated on a domain you own, which rules it out for cold email before volume is even a consideration: the mail leaves from gmail.com, SPF and DKIM belong to Google rather than to you, and a DMARC policy on your domain is irrelevant because your domain is not in the From address.

So the practical question is about Workspace mailboxes on a domain you control, and the practical ceiling there is 2,000 messages to 3,000 external recipients per user per day.

Why the published ceiling is not a target

A sending limit is a platform protecting itself. Google enforces 2,000 because past that point a compromised account becomes an expensive problem for Google. It is an abuse control. It says nothing about whether the 300th message of the day will be read.

The number that governs cold email is set on the receiving side, by filters at the providers your prospects use, and those filters are not counting against a published threshold. They are building a behavioural profile of the sending address and the domain behind it, and asking whether it resembles a person using email.

• Volume consistency. A mailbox that sends 15 a day for a month and then 900 in an afternoon has announced itself.
• Reply rate. Real correspondence produces replies. A mailbox that has never received one looks like broadcast.
• Engagement. Opens, replies, messages dragged out of spam, senders added to contacts.
• Complaint rate. The fastest way to lose a domain. Above roughly 0.3% and placement collapses.
• Bounce rate. High bounces say the list was never verified, which correlates strongly with spam.
• Content variance. Five hundred byte-identical messages is a signature no volume limit protects you from.

None of those improve by sending more and several get worse. That is why the safe rate sits so far below the cap. The constraint is not throughput, it is pattern recognition, and the pattern you want is an ordinary person who writes a handful of emails a day.

What we size a Google inbox at, and why

Fifteen campaign sends per inbox per day, plus about 20 warm-up sends alongside them. That is under one percent of the 2,000 Google would accept, and the gap is deliberate rather than timid.

Fifteen is a number a person could plausibly have written. It leaves room for a follow-up wave without the day spiking. It stops any single mailbox carrying enough of the campaign that throttling it costs the whole day. And the headroom between what we send and what the platform allows absorbs a bad week, rather than a bad week pushing you into refusals.

The warm-up sends are counted separately on purpose. They run in your own sequencer, they are traffic between mailboxes you or your tooling control, and their job is engagement rather than pipeline. They still consume platform budget, which is a further reason not to plan campaign volume near the ceiling: warm-up is not free, it is simply mail you are not measuring for replies. Our Google inboxes are provisioned against those two figures, so the sizing you order is the sizing that was planned for.

Keep the two numbers separate. 2,000 is what Google permits. 15 is what we size for. Anyone quoting the first as a cold email rate is quoting a fact about Google and presenting it as advice about deliverability.

What happens when you exceed it

There are two separate failures and they arrive in the wrong order. Hitting the platform cap is loud and recoverable. Damaging reputation is quiet and is not.

The loud failure: a temporary block

Cross the platform limit and Google stops accepting mail from the account. The sequencer sees an SMTP rejection rather than a bounce, and the text names a rate limit.

550-5.4.5 Daily user sending limit exceeded.
421-4.7.0 Try again later, closing connection.
550-5.7.1 Our system has detected an unusual rate of unsolicited
550-5.7.1 mail originating from your account.

A 4.x.x response is a deferral: the sequencer retries and usually succeeds. A 550 is a refusal, and the account is typically blocked for up to 24 hours. Occasionally Google suspends the account outright and an administrator has to restore it. None of that is subtle, and none of it is permanent.

The quiet failure: reputation

The damaging version never produces an error. The mail is accepted, the sequencer records it as delivered, and the reply rate falls to nothing because the mail is being filed as spam. Delivered and inboxed are different states, and only one of them is visible from the sending side.

The sequence is consistent. Placement moves from the inbox to Promotions, then to spam, then receivers begin deferring, and the sequencer retries and eventually records those as sent. By the time anything hard-bounces the domain has been in trouble for a week.

That asymmetry is the whole argument for planning below the cap. A rate-limit block costs you a day. Reputation damage outlasts the block by weeks, does not clear when you reduce volume, and attaches to the domain rather than to the mailbox, so replacing the mailbox fixes nothing. The reliable repair is a new domain and another warm-up, which is expensive enough that the sensible rate is one you never have to discover by crossing.

• Reply rate falling while delivery stays near 100% is the first useful signal, because it moves before anything is rejected.
• A gap between opens and replies that widens over days rather than jumping suggests placement rather than copy.
• Deferrals concentrated on one receiving provider point at reputation with that provider rather than at the list.
• Bounces above roughly 5% mean list quality is the problem and volume is not the thing to change.

Seed testing is the only way to see placement directly. Put a few addresses you control, across Gmail, Outlook and one corporate tenant, into the send, and look at where the mail lands rather than at what the dashboard claims. If placement has already slipped, work through the deliverability checklist before changing volume.

Plan volume from mailbox count, not from one mailbox

Two setups producing 45 sends a day look identical in a report and are not remotely the same to a receiver.

The arithmetic of outbound is mailbox count multiplied by a small safe number, never a few mailboxes multiplied by an aggressive one. Work backwards from the outcome you want.

1. Decide how many conversations a month you want.
2. Divide by your positive reply rate to get the sends required.
3. Divide by working days to get the daily volume.
4. Divide by 15, the per-inbox rate, to get the inbox count.
5. Divide the inbox count by the inboxes you are willing to put on one domain, to get the domain count.
6. Add headroom, because some inboxes will always be in warm-up or in rotation.

Worked through: twenty conversations a month, a 5% reply rate, a third of replies positive, needs roughly 1,200 sends a month. That is about 60 a working day, which is four Google inboxes at 15. Follow-ups come out of the same daily budget, so a three-message sequence absorbs fewer new contacts per day than the send count suggests.

Step five is where Google differs sharply from Microsoft, because it decides how many domains you register rather than how many mailboxes you add. How many Google inboxes to put on one domain covers that side of it, and the Microsoft version of this calculation runs on a much lower per-inbox number and a much higher mailbox count.

How warm-up changes the number over time

A new mailbox has no sending history, and no history is treated with suspicion. Fifteen is a warmed steady state, not a day-one figure.

• Week one: a few sends a day, ideally to addresses that will reply. Warm-up traffic only, no campaign.
• Week two: warm-up continues at around 20 a day and the first real prospect sends begin, well below target.
• Weeks three and four: ramp campaign volume toward 15 while watching placement rather than the send count.
• Ongoing: keep warm-up running alongside live sending, so engagement does not stop the day the campaign starts.

Warm-up runs in your own sequencer on Google exactly as it does on Microsoft, which is why the 20 warm-up sends are quoted as a sizing figure rather than as a service. What warm-up buys is history, and history is the only thing that makes 15 a day unremarkable rather than novel. The mechanics are in how to warm up a domain.

It does not keep climbing afterwards. After four weeks the mailbox has a reputation, and the reason to hold at 15 rather than walk up toward the platform cap is everything above: the ceiling is Google protecting Google, and the receiving side is measuring something else entirely.

The signals that mean reduce volume now

Cut back immediately if reply rate falls sharply with no change to copy or list, if seed sends start landing in Junk, if bounce rate rises above 3%, or if any complaint rate above 0.3% appears. A day of reduced sending is cheap. A domain is not.