What still gets replies in 2026: 8 cold email campaigns, 950 leads, and the 5 moves that mattered

📊The scoreboard: eight campaigns, counted per lead

Eight campaigns from my own sending accounts, each about a week, each testing a different brief against a fresh segment. The sequencer's dashboard also reports opens and clicks, and I'm not leading with either: Apple Mail Privacy Protection inflates opens, and corporate security scanners click every link before a human sees the email.[13] Replies are the only column in this table a machine can't fake, so that's the column, divided by leads, because that is the denominator you can reproduce from your own numbers.

Where that sits against the public datasets that publish a method:

Read it both ways. By Instantly's yardstick, these campaigns are average. By Smartlead's, which counts replies per contact the way the table above does, they are top decile, and the SDR Managers campaign beats Instantly's top quartile outright. Both readings are true. The denominator is the difference, which is why the benchmark tool at the bottom asks for yours.

The number I track after replies is how many of them turn into meetings. Gong and 30MPC's 85 million emails put the top 10% of reps at 29.6% of replies booked; the comparison group sits at 15.3%.[4] Hold your own funnel against that, not against a screenshot.

One honest caveat: these are my accounts, my lists and my niche, and the sample is campaigns rather than a controlled study. Treat it as an existence proof that the system works, not as a forecast for your numbers.

🧪What changed between 1.4% and 8.6%

The briefs differed on segment, angle and ask, so the table can't isolate one variable and I won't pretend it does. But five things fall straight out of it, and the rest of this page is built from them.

🎯Move 1: pick 100 people, not 1,000

The infrastructure section below says a mailbox can safely send twenty to forty emails a day. That number is doing you a favor. It means you are structurally prevented from spraying, so the list has to be the first draft of the message.

The biggest jump on the scoreboard had nothing to do with writing. SDR Campaign V3.5 sent to SDRs got 1.4% of leads to reply. The same brief version sent to SDR managers, a list of about the same size, got 8.6%. The seat changed, and with it whether the email was about a problem the reader owns. Size matters too, separately. Belkins' 16.5 million emails show campaigns under 50 recipients replying at 5.8%, against 2.1% for campaigns over 500.[3] Smartlead's 850 million sends put the median sender at a reply from 0.74% of contacts and the top decile at one in 38.[2]

That is what a segment is: one persona, one nameable operational break, and a list short enough that every email can be about that break.

🔎Move 2: let the machine read before you write

Most teams that bought AI for outbound pointed it at the writing. Instantly's benchmark, built on a year of sends across 700,000 businesses, puts the average reply rate at 3.43%. The top quartile gets 5.5%. The top decile clears 10.7%.[1] The spread between average and top is not talent. It's whether anything in the email is true of the reader and nobody else.

Buyers noticed before the vendors did. In Gartner's survey of 645 B2B buyers, 69% said they prefer to validate AI-generated insights with a sales rep rather than take them at face value.[8] Your prospect is not anti-AI. They are anti-being-processed.

What "personalization" degraded into

The current default is personalization theater: a merge tag, a scraped job title, and a compliment about a funding round that closed nineteen months ago.

Every clause in that email is true of ten thousand companies. It contains no information. A human being who spent four minutes on Meridian Capital's website could not have written it, because a human being would have found something to say.

Reading by hand is the expensive part. A 200-contact, seven-step sequence with real research behind each touch runs to roughly 58 hours, about $2,000 in fully loaded rep time; we broke that math down line by line in the real cost of manual prospecting. That is the step worth automating, and it is the one most tools skipped.

🎁Move 3: no ask in email one

Everyone repeats "don't sell in the first email" and almost nobody defines it. So here is the operational version: the first email contains no ask. Not a demo. Not a call. Not "do you have 15 minutes." Not "is this a priority for you." Not even "worth a chat?", which is still an ask wearing a casual outfit.

How a founder scaled to $2.5M using cold email

What replaces the ask is something that is only useful to this specific lead. A number about their market. A gap you noticed in their funnel. A comparison against the two competitors they actually lose to. The point isn't generosity for its own sake. Giving something specific is the only credible proof that you did the work.

The same prospect, two ways

The second email names the fund, the size, the incumbent portal, the specific operational break, and when it will happen. None of that comes from a template.

SimpleSend wrote that entire email, and the highlights mark the signals it found on its own. It read Meridian's site, found the fund close and the Juniper Square portal they route investor updates through, and built the message around them. I never opened their website. The portal is the whole reason this email lands, because it is the detail that proves a human looked, and it came out of the research step rather than a merge field.

That split is the point. SimpleSend finds what is true about a prospect at a scale I can't match by hand. I decide what it means and what to offer. Neither half works alone.

The close matters as much as the opener. Gong Labs' analysis of 304,174 cold emails found that an interest close, asking whether the problem is worth a conversation, booked meetings at 30%, against 15% for a specific-time ask and 13% for an open-ended one. Their phrase for it: you are selling the conversation, not the meeting.[5]

✍️Move 4: a structure that earns the reply

The list gets you the right reader and the research gets you something to say. Structure gets you the reply. This breakdown from a top-performing enterprise SDR is the clearest one I've found:

Cold email structure from a top-performing enterprise SDR

The skeleton I run, in order:

1. Subject line. Lowercase, two to five words, specific to them, zero hype. It should read like an internal email, because internal emails get opened. your Fund III portal, not Transform Your Investor Experience!
2. Open with the signal. First sentence names the specific, verifiable thing about them that caused you to write. Never "hope you're well." Never "I came across your profile." The signal is the opener.
3. The bridge. One sentence connecting that signal to a consequence they'd recognize. This is where you demonstrate you understand their world, not just their website.
4. The value. The thing you're giving. Concrete, ungated, theirs whether or not they reply.
5. The close. Interest-based, not calendar-based. "Might be nothing" beats "Are you free Thursday?" because it gives them a way to respond that isn't a commitment.

Follow-ups are most of the yield

How much of the yield depends on whose data you read, and the honest range is wide. Instantly's 2025 sends put 58% of replies on step one and 42% on the follow-ups.[1] Gong and 30MPC's 85 million emails put 70% of replies after the first email.[4] Smartlead found sequences of three to five steps get 2.3 times the replies of a single email,[2] and Belkins found steps two through six account for 58.6% of replies, with over half of email-sourced meetings coming from step three or later.[3] Woodpecker's older dataset had reply rate going from 9% with no follow-up to 27% with at least one.[12]

The catch is that a follow-up has to carry a new signal. "Bumping this" and "just floating this to the top of your inbox" are not follow-ups, they're nags. Each touch needs its own reason to exist, which, again, is a research problem before it's a writing problem. The scoreboard agrees: the campaigns that averaged four to five sends per lead did not out-reply the ones that averaged three.

📡Move 5: the infrastructure floor

You can write the best email of your career and never find out, because it landed in spam. Google's and Yahoo's bulk-sender rules apply to anyone sending 5,000 or more messages a day to their users, and Gmail has been ramping up enforcement since November 2025, from spam-foldering to temporary and permanent rejections.[9][10][11]

The non-negotiables

• SPF and DKIM passing and DMARC published on every sending domain. Google accepts a DMARC policy of none to start; aligned is where you want to end up.[9]
• One-click unsubscribe via the List-Unsubscribe and List-Unsubscribe-Post headers (RFC 8058) on marketing and subscribed mail, with requests honored within two days.[9][11]
• Spam complaints under 0.10%, and never reaching 0.30%. The higher number is the cliff, not the target.[9]

How the filters actually decide, at altitude

You don't need to reverse-engineer Gmail. You need the four layers, in order, because they tell you where your leverage is:

1. Authentication. Is this sender who they claim to be? Binary. Fail and nothing else matters.
2. Reputation. How has mail from this domain and IP performed historically? Slow to build, fast to burn.
3. Engagement. Do recipients open, reply, move it out of spam? Or delete-without-reading and mark as junk?
4. Content. Links, images, spam-trigger patterns, and increasingly whether the body reads as machine-generated.

The current meta

The providers publish the rules above and nothing about how many mailboxes to run or how much to send from each. What follows is where experienced operators converge, and their own write-ups disagree at the edges, so treat these as ranges.[14][15]

Notice what the volume ceiling implies. At twenty to forty sends per mailbox per day, you are structurally prevented from spraying. The infrastructure is telling you the same thing the reply data is: the constraint is quality per email, so that's where the work has to go.

⚡How the eight campaigns were produced

Everything above is correct and almost nobody does it, for one reason: doing it by hand for 200 contacts is 58 hours of work. Every campaign on the scoreboard came out of this loop instead, with SimpleSend connected to Claude over MCP so the research runs as a tool call rather than a browser tab.

That's the inversion. The pitch isn't "AI writes my emails". AI does the reading that makes a real email possible, and I stay in the seat where the judgment happens.

The whole workflow, step by step from one Claude chat, with the prompts to paste, is on the method page.

📋Steal these: the brief, the sequence, the checklist

Three things you can paste into a doc today. The brief is the same shape I give SimpleSend; it works just as well as a template for a human, it just takes 58 hours to fill by hand.

THE BRIEF (one per segment, under 250 words)

Who: [one persona, one segment, 100 people or fewer]
The sentence that is true of everyone on this list and nobody off it:
[...]

What I sell, in one line: [...]
The operational break this segment hits, and when it hits: [...]
Proof, one customer, one number: [company like theirs] had [same problem].
We did [intervention]. Result: [number, with units].

Signals to look for (up to five, all verifiable on their site or in the news):
1. Incumbent vendor or tool visible on their site
2. Recent raise, close, launch or senior hire
3. Open role that implies the break
4. [segment-specific]
5. [segment-specific]

What I'm giving in email one (ungated, theirs whether or not they reply): [...]

Voice, two sentences of how I actually write: [...]

Constraints: under 120 words. Lowercase subject, two to five words, specific
to them. No "hope you're well". No calendar ask in step one. Every follow-up
carries a new signal.

THE SEQUENCE (four steps, twelve days)

Step 1, day 0: the signal
  Subject: [lowercase, their thing, two to five words]
  Line 1: the verifiable thing about them that made you write.
  Line 2: the bridge. What that signal usually means for someone in their seat.
  Line 3: the gift. Concrete, ungated, theirs either way.
  Close: interest, not calendar. "Might be nothing, but the timing looked right."

Step 2, day 3: a new signal
  A second thing you noticed, and the one insight you'd give a friend in
  their role. No link. No ask.

Step 3, day 7: proof, then a soft ask
  One customer who looked like them, one number. Then lower the cost of yes:
  "Want me to send the [checklist / runbook / teardown] anyway?"

Step 4, day 12: the break-up
  One line. "Closing the loop in case the timing changes. The [asset] is
  yours either way."

PRE-SEND CHECKLIST

Infrastructure
[ ] SPF and DKIM pass and DMARC is published, on every sending domain
[ ] Secondary domains only; the primary brand domain never sends cold
[ ] 2 to 3 mailboxes per domain, warmed 14 to 21 days before real volume
[ ] 20 to 40 sends per mailbox per day; 50 is the ceiling, not the target
[ ] One-click unsubscribe headers where the rules require them, honored
    within 2 days
[ ] List verified before sending; bounces are a reputation event
[ ] Spam complaint rate tracked: under 0.10%, never near 0.30%

Copy
[ ] The segment sentence is written (true of everyone on the list, false
    of everyone off it)
[ ] Email one has no ask
[ ] The first line is a verifiable signal, not a greeting
[ ] Under 120 words, one idea, zero or one question
[ ] Subject is lowercase, two to five words, specific to them
[ ] Every follow-up carries a new signal
[ ] Would you still send this if they could never buy from you? If not, cut it.

🧮Benchmark yourself

Every benchmark on this page divides by something different: leads, sends, unique opens. Most dashboards pick the flattering one. Put your last campaign in and see where it sits on each scale, and what the same list would cost to research by hand.

Tiers use Instantly's 3.43% average, 5.5% top quartile and 10.7% top decile, and Smartlead's 0.74% median and one-in-38 top decile, per contact.[1][2] Manual cost uses 2.5 minutes per email and $35 per loaded rep hour.

Run this on your own list

Everything above is the system. The fastest way to judge it is to point it at a prospect you already know and read what comes back.

1. [1]Instantly — Cold Email Benchmark Report 2026 (a year of sends across 700k+ businesses)
2. [2]Smartlead — The State of Cold Email 2026 (850M+ sends, senders with 500+ sends)
3. [3]Belkins — B2B cold email response rates, 2026 study (16.5M emails, replies ÷ sent)
4. [4]Gong Labs × 30MPC — The Ultimate Cold Email Data Report (85M+ cold emails)
5. [5]Gong Labs — The surprising cold email CTA that books more meetings (304,174 emails)
6. [6]Lavender — The Cold Email Benchmark Report (231,818 emails, ~50k inboxes)
7. [7]Lavender — The best length for a cold email (28.3M sales emails)
8. [8]Gartner — Survey of 645 B2B buyers: 69% prefer to validate AI-generated insights with sales reps (May 2026)
9. [9]Google — Email sender guidelines (bulk sender requirements)
10. [10]Google — Email sender guidelines FAQ (enforcement timeline)
11. [11]Yahoo — Sender best practices (bulk sender requirements)
12. [12]Woodpecker — Follow-up statistics (older platform dataset; direction, not a 2026 figure)
13. [13]Mailjet — Email bot clicks: how security scanners inflate engagement metrics
14. [14]Scaled Mail — Cold email infrastructure setup guide (operator guide; opinion, not provider policy)
15. [15]Litemail — How many domains and inboxes for cold email in 2026 (operator guide; opinion, not provider policy)

Benchmark figures move quarter to quarter, vary by industry, list quality and offer, and depend on what each dataset divides by. Treat them as direction, not promises. The campaign results at the top are my own accounts, over my own lists, in one niche. Yours will differ.