In our 35-day test of 40 new mailboxes, every group started near 70% once SPF, DKIM and DMARC were live. The templated dedicated warmer was lowest on day 7, at 66%. Agentic warmup started just above it and was the only method still climbing after that. MailStrike reached 85% by the end of warming and 90% after two weeks of live cold sending. Templated warmup finished at 63%.
The warmup bundled into a sequencer finished at 48%, fourteen points below the 62% of mailboxes that were never warmed. Sequencers are good at sequencing. Warmup is not what they are built for, and the numbers show it. You pay for the sequencer, the domains, the data and the reps whether the mail is read or not, so a warmup that drags placement below doing nothing is the most expensive line on the stack.
Key findings
- Day 7 does not pick a winner. With authentication live, the four groups sat between 66% and 71%. The templated dedicated warmer was the lowest. Agentic warmup was two to three points behind the bundled and unwarmed groups.
- Nothing dropped in the second week. By day 14 every group was flat or up. Agentic had moved from 68% to 77%. The dedicated warmer had moved from 66% to 69%. The other two were still around 72 to 73%.
- Regular warmup stalled, then eased off. A templated dedicated warmer peaked at 69% on day 14, slipped to 66% by day 21, and finished at 63% after two weeks of live sending.
- Sequencer warmup failed under real volume. It reached 72% on day 14, gave back 3 points by day 21, then fell to 48% after live sending, and to 42% for Microsoft inboxes. That is worse than sending from a mailbox that was never warmed.
- Only agentic warmup kept climbing. MailStrike went from 68% on day 7 to 85% by day 21 and 90% after live sending, with Microsoft rising from 83% to 87%.
- Only agentic warmup readied a new domain in time. Median 16 days to full volume. No templated mailbox got there inside the 21-day warming phase.
Overall inbox placement by warmup method, 40 new mailboxes, Google and Microsoft averaged
Why regular email warmup stopped working
For years, warmup was a volume game. Connect a mailbox to a pool, let it trade a few dozen emails a day with other mailboxes, watch the opens and replies climb, and after three or four weeks start sending. That is regular warmup.
Google and Microsoft did not ban it. They learned what it looks like. Pooled mailboxes that open on schedule, reply with a line from a template and behave exactly like the mailbox next to them now read as warmup, and warmup that reads as warmup is discounted. A new mailbox with SPF, DKIM and DMARC already lands near 70%. That early number is setup, not reputation. Regular warmup adds activity on a shared pattern, so placement stays flat and then gives way once cold volume starts. Three things give it away.
The same behavior on every mailbox. Five mailboxes on one domain that open at the same pace and reply at the same rate read as one script running five times.
Content that repeats. Templates, spintax and topic fillers produce mail that looks like other warmup mail, across your mailboxes and across the pool.
A pool on any server. Your reputation is built where your mail is scored. A pool padded with mailboxes on SMTP servers spends your daily warming volume on engagement Google and Microsoft never see.
What the pool is made of decides what warming is worth
This is the part vendors skip. A mailbox sends a set number of warming emails a day, and each one only counts with the provider that hosts the mailbox it lands in. Google and Microsoft score your mail on what their own mailboxes do with it, so a warming send to a mailbox on an SMTP server is a slot of your daily volume spent on engagement neither of them sees.
That is why every mailbox in the MailStrike network is on Google or Microsoft. None of your daily volume is spent anywhere else.
Sequencers are great at sequencing, and bad at warmup
Instantly, Smartlead and lemlist are good tools for what they were built to do: schedule sequences, rotate mailboxes, track replies. Warmup is a bolt-on in all of them, included in the subscription so the pricing page has one more tick.
The bundled version is regular warmup: templated engagement on a shared pool. Smartlead documents spintax and an identifier tag in every warmup email, a fingerprint by design. Lemwarm documents no automated spam rescue, which matters most for a domain already landing in spam.
In our test, bundled warmup matched the field early, slipped by day 21, then fell to 48% after two weeks of live sending. That is 14 points below mailboxes that were never warmed. Most teams never see it, because the sequencer reports “sent” and the drop shows up months later as a reply-rate problem that gets blamed on the copy.
Keep sending from your sequencer. Switch its bundled warmup off on any mailbox a dedicated tool warms. The longer version is in dedicated vs bundled warmup.
Putting “AI” on regular warmup does not fix it
Every warmup tool now says it uses AI. The word covers three different things, and only the last changes what the filters see.
AI-written copy on regular warmup. A model writes the warming email about your industry topic, but the behavior around it (when it is opened, how long it is read, when the reply comes) follows one script per mailbox. Underneath, it is still regular warmup.
An AI assistant next to regular warmup. A chat agent or deliverability advisor sits in the dashboard. Useful, but the warming itself is unchanged.
Agentic warmup. Each mailbox gets its own AI Agent with a persona: its own reply rate, dwell time, active hours, writing style and chance of a typo. The agent writes each thread at send time and behaves like one particular person would, so there is no shared pattern to recognize.
How MailStrike's AI Agents warm a mailbox
Each mailbox gets a dedicated AI Agent. The persona sets reply rate, dwell time, active hours, writing style and typo chance, so no two mailboxes on a domain behave alike. Threads are written fresh at send time from your company description and industry. Activity stays inside working hours. Spam rescue, mark-as-important and link clicks are part of the engagement.
Every mailbox in the network is on Google or Microsoft, so each warming send counts with one of them and none is spent on an SMTP server neither can see.
How we tested AI email warmup
We wanted numbers an outbound team could act on, so we ran our own test rather than quote vendor claims.
- Mailboxes: 40 new mailboxes bought from a mailbox reseller, the way most outbound teams buy them: 20 Google Workspace and 20 Outlook (Microsoft 365). SPF, DKIM and DMARC were published on every domain before day one.
- Groups: four groups of 10 mailboxes, 5 Gmail and 5 Outlook in each. Agentic warmup (MailStrike), templated warmup from a dedicated warmer, templated warmup bundled into a sequencer, and no warmup.
- Days 1 to 21: warming only, on each tool's default ramp.
- Days 22 to 35: two weeks of live cold outbound. Each mailbox sent 18 emails a day to CMOs on a verified list, offering a GEO (generative engine optimization) discovery solution. One offer, one sequence and the same copy in every group, with warming continuing in the background.
- Measurement: a placement test every two days, recording inbox, promotions or spam at Gmail and Google Workspace (reported as Google) and at Outlook.com and Microsoft 365 (reported as Microsoft). Overall figures are the average of the two.
Ten mailboxes per group is a small sample, so read the gaps as direction rather than decimals. We do not name the vendors behind the templated groups: the point is the method, not a scoreboard.
Inbox placement results by warmup method
| Warmup method | Day 7 | Day 14 | Day 21 | Day 35 (after 2 weeks live) |
|---|---|---|---|---|
| Agentic warmup (AI Agent per mailbox) | 68% | 77% | 85% | 90% |
| Templated, dedicated warmer | 66% | 69% | 66% | 63% |
| Templated, bundled into a sequencer | 70% | 72% | 69% | 48% |
| No warmup | 71% | 73% | 71% | 62% |
Day 7 is setup, not method. The dedicated warmer started at 66%, the lowest of the four. Agentic warmup started at 68%, two or three points behind the bundled and unwarmed groups. Nobody fell between day 7 and day 14. The dedicated warmer peaked at 69%, then gave those 3 points back by day 21 and slipped another 3 points once campaigns started, finishing at 63%, level with mailboxes that were never warmed.
Sequencer warmup did worse than stall. It fell 21 points once campaigns started and finished 14 points below no warmup. A warmup that leaves you worse off than none is not a feature. It is a liability bundled into your subscription.
Google vs Microsoft inbox placement
| Warmup method | Google, day 21 | Microsoft, day 21 | Google, day 35 | Microsoft, day 35 |
|---|---|---|---|---|
| Agentic warmup (AI Agent per mailbox) | 87% | 83% | 93% | 87% |
| Templated, dedicated warmer | 69% | 63% | 66% | 60% |
| Templated, bundled into a sequencer | 73% | 65% | 54% | 42% |
| No warmup | 74% | 68% | 66% | 58% |
Microsoft is where sequencer warmup broke. At Google on day 35 it sat 12 points behind the unwarmed mailboxes. At Microsoft it trailed them by 16 points, and fewer than half of its emails reached the Outlook or Microsoft 365 inbox, against 58% for mailboxes that were never warmed. The templated dedicated warmer showed no real split between the providers: it was soft at both and slipped a little at both, finishing at 66% on Google and 60% on Microsoft.
Agentic warmup went the other way. Its Microsoft placement climbed from 83% to 87% during live sending, the only group to move up there. If your list is heavy on enterprise buyers, Microsoft decides your pipeline, and a Gmail-only placement check hides the problem.
New domain warmup: days to full volume
We counted the days until a mailbox first reached 80% inbox placement or better at both Google and Microsoft and stayed there for the rest of the warming phase. That is roughly the point at which a team can send at full volume without guessing.
| Warmup method | Median days to full volume |
|---|---|
| Agentic warmup | 16 |
| Templated, dedicated warmer | Not reached by day 21 |
| Templated, bundled into a sequencer | Not reached by day 21 |
| No warmup | Not reached by day 21 |
Only agentic warmup got a new mailbox to full volume inside the warming phase. A team on regular warmup either waits longer or starts its cold outbound campaigns from a domain the providers do not trust yet.
What poor inbox placement costs an outbound team
Take a team of 10 SDRs, each on five mailboxes sending 18 emails a day, 22 working days a month: 19,800 cold emails a month. At the day-35 placement above, agentic warmup puts about 17,800 of them in front of a person. The warmup bundled into the team's sequencer puts about 9,500 there, fewer than no warmup at all (about 12,300). That is roughly 8,300 emails a month, paid for in domains, data, tooling and SDR hours and read by nobody. The templated dedicated warmer lands about 12,500, level with doing nothing.
Replies followed placement. In the live-sending weeks the agentic group got 9 positive replies per 1,000 sends, against 5 for the templated dedicated warmer, 4 for no warmup and 3 for sequencer warmup. Every group pitched the same GEO discovery offer to CMOs with the same copy, so the only thing that differed was the warmup behind the mailbox.
How to choose AI email warmup software
Every tool on this page says it uses AI. Six questions separate warmup that still works from regular warmup with a new label.
- 1
Does each mailbox behave differently?
This decides the rest. “AI-generated content” on one behavior profile per mailbox is regular warmup: the copy changes, the pattern does not. Look for per-mailbox behavior, threads written at send time, spam rescue (moving warming mail out of spam and marking it not spam) and mark-as-important, which providers record as a positive signal.
- 2
What is the warmup network made of?
Your warming mail is exchanged with other members' mailboxes, and each send only builds reputation with the provider hosting the mailbox it lands in. Ask which providers the pool sits on: sends to SMTP servers spend daily volume Google and Microsoft never see.
- 3
Can you see inbox placement per provider?
Regular warmup hides behind open rates, which Apple Mail Privacy Protection and corporate link scanners inflate. Ask whether the tool runs placement tests that report inbox, promotions and spam separately for Google and Microsoft. Our test shows why: sequencer warmup looked close to no warmup at Google on day 21 and fell well below it at Microsoft once campaigns started.
- 4
Is warmup the product, or a feature?
A sequencer's job is sequencing. A tool whose business is warmup has a reason to keep improving it. A tick on a pricing page does not.
- 5
Does it handle multiple mailbox management?
Outbound teams run dozens of mailboxes. Check that you can set daily volume per mailbox, see each mailbox's placement and reputation trend on one screen, and hear which mailbox is slipping before it drags the domain down.
- 6
Does it tell you when a new domain is ready?
A new domain starts with no reputation, and Google and Microsoft filter it hard until it earns one. The tool should tell you plainly when a mailbox can send at full volume, rather than leaving you to guess.
The best AI email warmup software, compared
| Tool | Warming method | Warming content | Spam rescue | Placement tests | Price from |
|---|---|---|---|---|---|
| MailStrike | Agentic warmup: an AI Agent per mailbox, peer-to-peer engagement, with spam rescue | Written by the LLM per thread, from the thread context and your company description | Yes | Yes | $18 per mailbox a month, billed annually; the per-mailbox price falls as you add mailboxes |
| Instantly | Peer-to-peer templated engagement and spam rescue | Templated | Yes | Yes | $47 a month (Growth), warmup included |
| Smartlead | Peer-to-peer templated engagement (spintax) and spam rescue | Templated, with spintax | Yes | Yes | $39 a month (Base), warmup included |
| MailReach | Peer-to-peer AI-written engagement and spam rescue | AI-written, one behavior per mailbox | Yes | Yes | $19.50 per mailbox a month |
| Warmy | Peer-to-peer industry-templated engagement and spam rescue | Templated to an industry topic | Yes | Yes | Not published; quoted after a demo call |
| TrulyInbox | Peer-to-peer AI engagement on an industry topic, and spam rescue | AI-written on up to five industry topics | Yes | No | Free plan; paid from $22 a month, billed annually |
| Lemwarm | Peer-to-peer templated engagement; no spam rescue documented | Templated; your own templates on the Smart plan | No | Yes | $29 a month standalone ($24 yearly), or free inside lemlist |
Competitor cells come from each vendor's own public material, as recorded on our head-to-head pages. The full field of eleven tools is in the best email warmup tools roundup.
MailStrike: best overall for outbound teams
The one tool here that is not regular warmup underneath. Every mailbox gets a dedicated AI Agent whose persona sets its reply rate, dwell time, active hours, writing style and typo chance, so no two mailboxes on a domain behave alike. Threads are written fresh at send time from your company description and industry, activity stays inside your working hours, and spam rescue, mark-as-important and link clicks are part of the engagement. Each mailbox includes a deliverability test every month, and Flashman, the AI assistant, reads every inbox daily and tells you which mailbox needs attention. It connects over OAuth and runs alongside Instantly, Smartlead, lemlist, Apollo or HubSpot without touching your campaigns. Pricing is per mailbox and the per-mailbox price falls as you add more, so the teams running the most mailboxes pay the least for each one.
Best for: outbound teams running several mailboxes per domain, new domains that need full volume fast, and domains that have slipped into spam. The caveat: it warms in English, and it only connects Google and Microsoft mailboxes.
Instantly and Smartlead: great sequencers, warmup as an add-on
Buy them to sequence. One login and one bill is convenient, but the warmup is regular warmup: templated engagement on a shared pool, and Smartlead documents spintax and an identifier tag in every warmup email, a fingerprint by design. Bundled warmup as a category is what finished at 48% in our test, below no warmup. Keep sending there and switch their warmup off on the mailboxes a dedicated tool warms.
MailReach: AI copy on one behavior per mailbox
A self-serve dedicated warmer with a spam tester in the same plan, and documented spam rescue and mark-as-important. Its warming mail is AI-written, but each mailbox runs one behavior, so a three-mailbox domain repeats one script three times. In our test that pattern is what started at 66%, never left the high 60s, and eased off once live sending started.
Warmy and TrulyInbox: for warming in other languages
The two to look at if you warm in languages other than English: Warmy documents more than 30, TrulyInbox 12. Both write warming mail around an industry topic rather than per thread. Warmy's is templated and its price is quoted after a demo. TrulyInbox has a free plan.
Lemwarm: included with lemlist
The default for lemlist teams, since it comes free with every plan. It is regular warmup on templates, and its material documents no automated spam rescue, which matters most for a domain already landing in spam.
How to test your own inbox placement
You do not need 40 mailboxes to find out whether your warmup still works, but you do need to keep the test clean. Mailboxes on the same domain share one reputation, so two warmup tools on one domain contaminate each other and you cannot tell which one moved the number. Split by domain, not by mailbox.
- 1
Give each warmup provider its own domain.
Every mailbox on a domain uses the same tool, and no domain is warmed by two tools at once.
- 2
Match the domains.
Same age, same mix of Google and Microsoft mailboxes, same daily volume, same list segment and same copy.
- 3
Send your real email to a placement test from one mailbox per domain.
Unchanged, to a test that covers Google and Microsoft.
- 4
Record the split per provider.
Inbox, promotions and spam, for Google and Microsoft separately.
- 5
Repeat weekly.
For a month while you warm, and once more after your first week of live sending.
If your numbers look like the templated rows above (flat through week two, a small slip by day 21, then a real drop once campaigns start, and worse at Microsoft than at Google), your warmup is building activity, not reputation. Do not run the test on a domain that already sends live campaigns, because its reputation is already carrying other traffic. More on reading the results in how to test email deliverability.
Frequently asked questions about AI email warmup software
What is the best AI email warmup software for outbound teams?
MailStrike, on this test. It was the only method that kept raising inbox placement after day 7, the only one to reach full volume inside 21 days, and the only one whose Microsoft placement rose during live sending.
Does regular email warmup still work in 2026?
Not as a way to build placement. In this test a templated dedicated warmer started at 66%, peaked at 69%, and finished at 63%. It never left the band a new, authenticated mailbox already sits in.
Is the warmup built into my sequencer enough?
No. Sequencers are great at sequencing, but their bundled warmup matched the field early, slipped slightly by day 21, then fell to 48% after two weeks of live sending, below mailboxes that were never warmed. If you send through Instantly, Smartlead or lemlist, keep sending there and warm elsewhere.
Are free email warmup trials safe?
Check what the pool is made of before you switch one on. Google and Microsoft score your mail on what their own mailboxes do with it, so every warming send that lands on an SMTP server is a slot of your daily volume spent where neither can see it. MailStrike's network is 100% Google and Microsoft mailboxes.
Is AI email warmup different from regular warmup?
Only if the behavior changes, not just the copy. AI-written mail on one script per mailbox is still regular warmup. Agentic warmup gives each mailbox its own reply rate, dwell time, active hours, writing style and typo chance, and writes each thread at send time.
How long does new domain warmup take for cold outbound?
In this test, 16 days median for agentic warmup to reach 80% or better at both Google and Microsoft and hold it. No templated mailbox got there inside 21 days.
Is Microsoft stricter than Google on email warmup?
Yes, once real volume starts. On day 35 the bundled group was at 54% on Google and 42% on Microsoft. Agentic warmup was at 93% and 87%. A Gmail-only check hides the gap.
How many mailboxes should an outbound team warm per domain?
Enough that one slipping mailbox does not drag the domain, and few enough that each can behave differently. The failure mode in this test was several mailboxes on one domain sharing one script.
Can I use AI warmup software with Instantly, Smartlead or lemlist?
Yes. MailStrike connects over OAuth and runs alongside those sequencers without touching campaigns. Switch their bundled warmup off on the mailboxes a dedicated tool warms.