Cold email best practices, tested on 1,116,744 emails
By Ryan Doyle, co-founder of Sales.co. Published October 23, 2026, updated October 23, 2026.
We tested 15 of Lavender's cold email rules on 1,116,744 first emails Sales.co sent between Oct 6, 2025 and Oct 9, 2026. 2 held up, 3 flipped, 9 made no clear difference, and 1 had too few emails to call. The biggest win: offering to send something instead of asking for a call got 51% more human replies (0.76% vs 0.50%) and 2.6x the interested replies.
What this page adds: The cold email rules widely shared as Lavender's advice trace back to its analysis of 28.3 million emails. We sent 1,116,744 first emails for 57 sending workspaces, kept the text of each one, and know which recipients replied. So each rule can be checked against real replies, and each result is checked again with our largest sender removed.
The scorecard · What held · What flipped · What we found that the rules miss · Length · Method · FAQ
The scorecard: 15 rules, 2 held, 3 flipped
Each row compares what the rule recommends with what it warns against. "Held" means the recommended version got at least 10% more human replies per email, both overall and with the largest sender removed. "Flipped" means the warned-against version won by the same test.
| Lavender's rule | Human replies per email | Difference | Without largest sender | Interested replies | Verdict |
|---|---|---|---|---|---|
| Write 25-50 words | 26-50 words: 0.76% 101-125 words: 0.66% | +14% | -4% | +38% | No clear difference |
| Write at a 3rd to 5th grade reading level | grade 4-5: 0.77% grade 10-11: 0.66% | +16% | -29% | +24% | Mixed |
| Use 1-2 sentence paragraphs | 1 sentence per paragraph: 0.68% 3 per paragraph: 0.69% | -2% | -2% | +3% | No clear difference |
| Focus on their problem, not you | more "you" than "I/we": 0.64% more "I/we": 0.74% | -14% | -18% | -23% | Flipped |
| Use a relevant customer story | names a customer result: 0.70% no customer story: 0.69% | +1% | +2% | -8% | No clear difference |
| Don't ask for the meeting | offer to send something: 0.76% ask for a call: 0.50% | +51% | +42% | +160% | Held |
| Use hedge words (might, seems, could) | hedge words: 0.71% none: 0.68% | +5% | +14% | -10% | No clear difference |
| Drop the "did you know" stats | no stat: 0.71% a % or Nx stat: 0.63% | +13% | +20% | +38% | Held |
| Make one ask | one question: 0.69% two or more: 0.73% | -5% | +7% | +5% | No clear difference |
| Keep subject lines to 1-3 words | 1-2 word subject: 0.50% 5-7 words: 0.79% | -37% | -34% | -56% | Flipped |
| No first names in the subject | no first name: 0.56% first name in subject: 0.88% | -37% | -30% | -39% | Flipped |
| No questions in the subject | no question mark: 0.68% question subject: 0.70% | -4% | +6% | -16% | No clear difference |
| No numbers in the subject | no number: 0.68% number in subject: 0.66% | +3% | -4% | -18% | No clear difference |
| Cut "I hope this finds you well" | no pleasantry: 0.69% "hope you're well": 0.74% | -7% | 0% | -26% | No clear difference |
| Leave out links | no link: 0.69% a link: 0.64% | +9% | -1% | +336% | Too few emails |
Body rows: 1,116,744 recipient threads. Subject rows: 901,002 recipient threads (October 6, 2025 to October 8, 2026). "Difference" is the recommended version's rate over the warned-against version's rate. Lavender's rules 2, 5-7, 11, 15 and 20 are not tested: our data has no read time, no reliable label for observations or research quality, and no per-mailbox daily volume.
What held up
Don't ask for the meeting. This was the largest gap in the data. Offering to send something (a short video, an example, a list) got 0.76% human replies per email. Asking for a call got 0.50%. On interested replies the gap was 2.6x: 0.24% vs 0.09%. A question with no meeting in it sat between the two at 0.67%.
Drop the stats. Emails with a percentage or an "Nx" figure got 0.63% human replies per email against 0.71% without one, and 28% fewer interested replies.
Not tested properly: leave out links. Only 2,201 of our first emails fell on one side of it, too few to call.
What flipped
"Focus on them, not you." Emails that said "I", "we" and "our" more often than "you" got 0.74% human replies per email. Emails that said "you" more got 0.64%. With the largest sender removed: 0.69% vs 0.57%. Saying plainly who you are and what you do seems to work better than an email built around the word "you".
Short subjects. One- and two-word subject lines got 0.50% human replies per email, the lowest of any length. Five to seven words got 0.79%. Without the largest sender: 0.50% vs 0.76%.
First names in the subject. Subjects with the recipient's first name got 0.88% human replies per email against 0.56% without it.
No clear difference: write 25-50 words (+14% overall, -4% without the largest sender); write at a 3rd to 5th grade reading level (+16% overall, -29% without the largest sender); use 1-2 sentence paragraphs (-2% overall, -2% without the largest sender); use a relevant customer story (+1% overall, +2% without the largest sender); use hedge words (might, seems, could) (+5% overall, +14% without the largest sender); make one ask (-5% overall, +7% without the largest sender); no questions in the subject (-4% overall, +6% without the largest sender); no numbers in the subject (+3% overall, -4% without the largest sender); cut "i hope this finds you well" (-7% overall, 0% without the largest sender).
What we found that the rules miss
These held overall and with the largest sender removed.
| Finding | Human replies per email | Difference | Without largest sender | Interested replies |
|---|---|---|---|---|
| Greet them by first name | "Hi Sarah,": 0.80% a greeting without their name: 0.54% | +47% | +40% | +94% |
| Keep it to one or two paragraphs | 1-2 paragraphs: 0.83% 7 or more: 0.55% | +52% | +26% | +114% |
| Add a P.S. | a P.S. line: 0.88% no P.S.: 0.68% | +31% | +23% | +50% |
| Drop the exclamation marks | no exclamation mark: 0.70% one or more: 0.54% | +28% | +17% | +105% |
| Ask something | one question: 0.69% no question at all: 0.53% | +31% | +21% | +74% |
| Put their first name in the subject | first name in subject: 0.88% no first name: 0.56% | +58% | +42% | +65% |
| Capitalize the subject's first letter | "Quick question": 0.71% "quick question": 0.53% | +33% | +15% | +69% |
How the email is built matters more than its parts. Our writer labels each email with its structure. The best was "Blunt" at 0.92% human and 0.37% interested replies per email (38,102 emails). The worst was "Observation → Stakes → Offer → CTA" at 0.58% and 0.25% (16,276 emails). Without the largest sender: 0.63% vs 0.47%.
How long should a cold email be?
Lavender's first rule says 25-50 words gets 65% more replies than 125 words. In our data the shortest band did best overall, but the gap was small and it reversed with the largest sender removed. Short emails did get more interested replies, so length is worth trimming, but it is not where the replies come from.
| Body length | Emails | Human reply rate | Interested rate | Human rate, largest sender removed |
|---|---|---|---|---|
| 26-50 words | 81,656 | 0.76% | 0.24% | 0.66% |
| 51-75 words | 394,996 | 0.71% | 0.22% | 0.62% |
| 76-100 words | 412,039 | 0.68% | 0.18% | 0.61% |
| 101-125 words | 163,133 | 0.66% | 0.17% | 0.68% |
| 126-150 words | 54,443 | not shown | not shown | 0.67% |
| 151+ words | 8,982 | not shown | not shown | 0.76% |
Words in the body after the greeting, signature removed. Fewer than 2,000 emails, or one sender above 45% of a band: not shown.
The strongest objection
This is observational data. Different senders write different emails to different lists, so a trait can look good because the senders who use it have better lists. Two checks limit that: each result is shown again with the largest sender removed, and no band is shown when one sender made more than 45% of it. Treat a "held" result as a strong bet for your own split test, not as proof.
What to do with this
Greet them by name, say who you are in a sentence, make one specific offer to send something useful, and end on that question. Skip the stats, the exclamation marks and the meeting request. Across all 1,116,744 first emails the human reply rate was 0.69% and the interested rate 0.20%. For when to send, see the best time to send cold email; to check a subject line, use the subject line grader.
If you would rather have someone write and send it, Sales.co runs done-for-you cold email from $2,000 a month for up to 10,000 emails, month-to-month. See pricing.
Method
- Source: the text of every cold email Sales.co sent from October 6, 2025 to October 9, 2026 (1,202,138 first emails scanned), and the replies the sending tool labeled.
- Unit: recipient thread: one recipient address in one sending workspace, described by the first email that address got. 1,116,744 recipient threads from 57 sending workspaces.
- A human reply is the recipient replied at least once with a reply that is not an auto-reply, out-of-office, bounce, warm-up, test or spam (sending tool's labels). An interested reply means at least one of those replies was labeled 'possibly interested'.
- Each email body was turned into traits (word count, Flesch-Kincaid grade, paragraphs, questions, the ask in the last paragraph, links, stats, pleasantries, "you" vs "I/we") by a script, and the text was dropped. No email, subject line, address or client is published.
- Subject line rows use 901,002 recipient threads and the subject of the first email each recipient got.
- Left out: any band with fewer than 2,000 emails, or where one sending workspace made more than 45% of them. The largest sender made 16% of all first emails.
- Refresh: each time the data is recomputed, this page is regenerated from it.
Cite this: Sales.co, "Cold email best practices, tested on 1,116,744 emails", https://sales.co/research/cold-email-best-practices, data through October 9, 2026.
Sources
- Sales.co sending database: 1,116,744 first emails and 7,719 human replies, October 6, 2025 to October 9, 2026, as of October 23, 2026.
- Lavender's 28.3 million email analysis, as reported by Social Media Today, and the 20-rule summary of Lavender's advice circulating on X, read October 23, 2026.
FAQ
What gets cold emails the most replies?
In 1,116,744 first emails Sales.co sent, the biggest single difference was the ask. Emails that offered to send something (a video, a list, an example) got 0.76% human replies per email, against 0.50% for emails that asked for a call or meeting, and 2.6x the interested replies.
How long should a cold email be?
Length mattered less than most advice says. 26-50 word emails got 0.76% human replies per email and 101-125 word emails got 0.66%, but with the largest sender removed the order reversed (0.66% vs 0.68%). Shorter emails did get more interested replies: 0.24% vs 0.17%.
Should you ask for a meeting in a cold email?
No. Asking for a call got 0.50% human replies and 0.09% interested replies per email. Offering to send something got 0.76% and 0.24%. The pattern held with the largest sender removed (0.69% vs 0.48%).
Should you put the first name in a cold email subject line?
In this data, yes. Subjects that contained the recipient's first name got 0.88% human replies per email against 0.56% without it (901,002 recipient threads), and the gap held without the largest sender.
Do statistics help a cold email?
No. Emails with a percentage or "Nx" figure in the body got 0.63% human replies and 0.15% interested replies per email, against 0.71% and 0.21% without one.
Does "I hope this finds you well" hurt reply rates?
Not in this data. Emails with a pleasantry like "hope you're well" got 0.74% human replies per email against 0.69% without one, and with the largest sender removed the two were level (0.63% vs 0.63%).