Starter
For your first email program.
$16USD / month
$192 billed yearly.
20,000 emails / month
Choose Starter- 5 brands · 3 seats
- 5 custom sending domains
- 500 email verifications/month
- $1.50 / 1,000 extra emails
What warmup is for
Warmup is the process of building sending volume gradually on a new domain or IP so that receiving providers can form a reputation for it. A brand new sending identity has no history, and in email an absence of history is not neutral. It is suspicious.
The logic from the receiver's side is simple. A domain that has never sent anything suddenly emitting a hundred thousand messages looks exactly like a compromised or disposable domain. The only way to distinguish yourself is to arrive slowly, to recipients who visibly want the mail.
A receiver can answer an unproven sender in two ways: block it, which ends the story, or throttle it, which paces the sender while it keeps going. First sends to major providers may land in spam and settle as the domain warms. Volume steps up in ramps and holds at the allowance, never past it.
This page is the overview for warmup within email deliverability. The step-by-step version is how to warm up an email domain, and email warm up and email warming cover the same process from the inbox side. Email warm up tools compares the warmup services, and IP warming covers a new dedicated IP.
What changes when an AI agent warms up a new sending domain
An agent working over MCP can import 40,000 contacts and build a campaign to all of them in the same session, on a domain verified an hour earlier. Asked to send today, it will try. Nothing in the import gives the domain a reputation. Receivers judge the sending domain by what it has sent, and a fresh import has sent nothing.
A well-briefed agent reverses the order. It checks the account's readiness and sender identity first, then picks the most engaged contacts for a small first wave, then steps up no more than double per send. It treats two signals as a hold: a Gmail 421 deferral and a complaint rate that starts to rise. George Hartley, Co-founder of Nitrosend, sets the rule for a deferral: to pause for 15 minutes, send one test, and only if it delivers, hold below the volume that triggered the deferral for 24 hours before raising again. Fast retries make it worse.
The same plan works whether the agent is Claude, ChatGPT, Cursor, Codex or Gemini CLI. Email for Claude shows the setup for one of them.
How the daily allowance works, and why
Nitrosend holds every account to a daily allowance set by its plan and sender standing. Calendar age and payment alone cannot raise it. Standing rises on evidence such as a verified domain, a list that verifies clean, or clean sending history, and the full allowance table by sender standing sits with the step-by-step guide.
On the Free plan, recipients per rolling 24 hours run from 100 to 5,000 by sender standing. A new account starts at 100, and the allowance rises to 500 at Standard and 5,000 at Trusted. Nitrosend checks the allowance before a message is queued, so no account sends past its plan and standing, and a Probation account cannot put a bulk send on a cold domain. The check protects the sender as much as the recipient. A single bad first send burns the domain, and on a shared pool it hurts every other sender in it.
Warmup applies to application mail as well. A new domain sending receipts is still a new domain. Transactional email covers that case, and a transactional email API or the email API sends under the same allowance as a campaign.
When a send is larger than the allowance left for the day, Nitrosend paces it and says when the rest can go, instead of letting it all out at once.
An example ramp for a 40,000-contact list
One illustration, engaged recipients first, each step conditional on the last| Stage | Send size | Who receives it |
|---|---|---|
| Wave 1 | 2,000 | The most engaged contacts, people who opened or clicked recently |
| Next sends, first weeks | Up to 10,000 | Engaged contacts, widening slowly, each send no more than double the last |
| Later sends | Up to 20,000 | A broader slice of the list, while bounces and complaints stay low |
| Full list | 40,000 | Only if every earlier step held |
Hartley grows a ramp only when the last send had hard bounces under 2%, complaints under 0.1% and clicks holding. A rising complaint rate means hold, not continue.
Engaged first, always
The sequence matters more than the numbers. Early positive signal is what the receiver is measuring, so the first sends go to the contacts most likely to open, click and reply, and the audience widens in batches. Opens, replies and a total absence of complaints from a small audience are worth far more than volume.
Warmup means real sending to people who want the mail. Warming networks and inbox-exchange services that trade synthetic opens between mailboxes are a shortcut Hartley rules out. The early signal a receiver reads has to come from people who asked for the mail. Email warm up tools compares what those services do and where they fall short.
Before and during a ramp, verify placement with a test send rather than assuming the ramp is working, and watch blacklist status in case a listing is suppressing everything downstream.
Dedicated IP: a second reputation to build
A dedicated IP is often a poor fit. Unless volume is high and steady, the IP does not receive enough traffic to establish a stable reputation, so it stays permanently unproven and placement is worse than it would be in a well-run shared pool.
A new dedicated IP also requires its own careful warmup, exactly like a custom sending domain, so choosing one adds a second reputation to build rather than replacing the first. IP warming covers that ramp. Get authentication right, warm the domain, keep bounces low, and revisit the IP question only if volume genuinely justifies it.
Warmup applies to application mail as well, which surprises people: a new domain sending receipts is still a new domain. Transactional email covers that case, and a transactional email API or the Before. Left unguided, an agent builds one campaign to all 40,000 contacts for today, on a domain with no sending history. Receivers see a stranger arrive with 40,000 messages, and most of the send is held back by the allowance anyway.
After. With the Nitrosend MCP tools, the agent plans before it sends:
nitro_get_statusreads account context and readiness for that sender identity, so the plan starts from where the account stands.nitro_manage_domainsverifies the new sending domain and runs a diagnose on its records.nitro_define_segmentpreviews the most engaged contacts from the import, without saving the segment, to size a first wave of about 2,000.nitro_compose_campaigncreates a broadcast draft for that 2,000-contact first wave. Creating the draft delivers nothing.nitro_review_deliverychecks the first-wave draft before anyone sees it, with validation, spam analysis, preflight and a preview of what the engaged cohort will receive. It cannot send.nitro_send_test_messagesends one real test of the draft to the team, so the wave is seen in a real inbox. It never sends to an audience.- A person reads that review and approves the first wave with
nitro_control_delivery. The agent does not approve its own campaign. - After the wave lands,
nitro_get_insightspulls the analytics for it, and the agent proposes the next wave, no more than double, only if bounces and complaints stayed low.
The difference: 2,000 engaged recipients on the first day instead of 40,000 strangers, and every next step waits on the numbers from the last one.
Where an agent's warmup goes wrong
- The whole import goes out on day one. An agent that treats a freshly imported list as an audience sends to all of it, and a domain with no history meets receivers at full volume.
- Volume more than doubles after one good send. A clean first wave reads as permission to jump, and a five-fold step resets the trust the first wave earned.
- A warming network stands in for real sends. Asked to speed things up, an agent signs the domain up to an inbox-exchange service, and the domain's early signal comes from mailboxes no real recipient owns.
- A 421 deferral gets fast-retried. A retry loop resends deferred mail within seconds, which tells Gmail the sender ignores its signals. The right move is a pause and one test.
- The ramp keeps climbing while complaints rise. An agent following a fixed schedule raises volume on the calendar, not on the results, and the complaint rate is what decides the next step.
Go deeper
More on deliverability
Related reading
References
The primary sources behind the rules on this page. Provider policy and the underlying standards, not vendor marketing.
Common questions
How long does email warmup take?
Two to four weeks for a domain going to normal production volume. The pace is set by results, not the calendar. Each send is no more than double the one before, and the next step waits until bounces and complaints from the last send stay low.
Do I need a dedicated IP?
Usually not, unless your volume is high and steady. Below that a dedicated IP gets too little traffic to build a stable reputation, and a well-run shared pool places mail better. A new IP also needs its own ramp of two to four weeks, so it adds a second reputation to build rather than replacing the first.
What happens if I send too much too soon?
Mail gets slowed, sent to spam, or blocked. On the Nitrosend Free plan the allowance runs from 100 to 5,000 recipients per rolling 24 hours by sender standing. A new account starts at the low end, a verified domain and a clean list can raise it early, and clean sending raises it further.
Can an AI agent run the warmup?
Yes, if it plans the ramp before it sends. Over MCP an agent can verify the domain, preview the engaged contacts, size a first wave of about 2,000, draft it, send one test and read the results afterwards. A person approves each wave with nitro_control_delivery, and the agent never approves its own send.