Cold Email Benchmarks and Myths in 2025-2026

What the reply-rate numbers actually mean, and which viral stats do not survive their own source

Research date: August 8, 2026
Scope: US and international B2B cold outreach. Recruiting and B2C promotional datasets are labeled as such and never blended into a B2B sales figure.
Evidence standard: platform transparency reports with a stated method and sample size first (Belkins, Instantly, Pin); official platform and regulator documents for deliverability and law (Google, Apple, FTC statute, GDPR, UK ICO); vendor guides with no disclosed sample third; content-farm statistics with no auditable source are not accepted as proof.

Executive summary

Cold email is not dead, and it is not a 26% growth hack. Both stories are laundered. The honest version is duller and more useful: reply rate is almost entirely an artifact of which denominator a vendor chose, most viral tactic-multipliers trace to a single vendor or a 2013 study about a different kind of email, and the two things that are actually pinned down by primary sources are boring infrastructure rules and a law that most people describe backwards.

What holds up:

• Reply rate is denominator theater until you read the method. Belkins measured 0.45% replies per total send across 7,530,489 emails sent in 2025, and said so plainly, noting that its own prior studies reported a much higher number only because they divided by openers instead of by sends. Instantly measured 3.43% average replies per send across billions of platform interactions in the same year. These are not in conflict about reality. They are different fractions. Neither is "the" cold email reply rate. (Belkins, June 26, 2026; Instantly, January 12, 2026)
• "Personalized subject lines get 26% more opens" is a 2013 promotional-email finding wearing a 2026 cold-email costume. The 26% traces to Experian's 2013 Email Marketing Study, which measured unique open rate on opt-in promotional mailings, not cold B2B, and used the open-rate metric that Apple broke in 2021. Personalization does lift cold replies in transparent data, but through the reply, not the borrowed opens figure. (Experian, 2013, via Experian plc newsroom, 2014; Belkins subject-line study, 2025; Pin, 2026)
• "Shorter emails always win" is contradicted by the two most transparent length datasets. Instantly found campaigns under 80 words performed best. Pin found 150-199 word bodies peaked reply rate in recruiting. Both are large and both are honest. They disagree, which means the universal rule does not exist. (Instantly, January 12, 2026; Pin, 2026)
• "Tuesday 10am" is folklore stapled to a real, smaller effect. Belkins' 7.5M-email data does show mornings winning: 8am-noon at 0.54% versus late evening at 0.40%, with Wednesday and Thursday as the best days. That is a real time-of-day association. It is not Tuesday, it is not 10am, and the "30-45% more replies from timing alone" claim has no auditable source. (Belkins, June 26, 2026)
• Open rate is genuinely broken as a precise metric, and practitioners now act like it. Apple Mail Privacy Protection preloads the tracking pixel and reports Apple Mail messages as opened whether or not the recipient did anything. Belkins disabled open tracking entirely for its 2025 study. "Unmeasurable" slightly overstates it, but "do not run your campaign on open rate" is correct. (Mailchimp Apple MPP FAQ; Belkins, June 26, 2026)
• Cold email is not illegal, and the people who say GDPR or CAN-SPAM killed it usually have the law backwards. CAN-SPAM never required prior consent. GDPR Recital 47 explicitly allows direct marketing to rest on legitimate interest, and UK PECR permits emailing corporate bodies without prior consent. There are real rules, and there are real penalties for breaking them, but the rules regulate cold email, they do not ban it. Legal specifics belong with a qualified professional, not a research report. (15 U.S. Code 7704; GDPR Recital 47; UK ICO, PECR electronic mail marketing)

The practical conclusion: stop shopping for the reply-rate number and the tactic-multiplier. Read the denominator, segment your own data, and treat every "+X%" tactic stat as a hypothesis to test on your list, not a platform fact. The two claims that are actually verified are the least exciting ones: fix deliverability infrastructure to a published standard, and know which recipients you are legally allowed to email.

Claim-by-claim verification

Provenance flags

• The "26% more opens" figure is a 2013 promotional-email statistic. It traces to Experian Marketing Services' 2013 Email Marketing Study, which reported personalized subject lines delivering 26% higher unique open rates on opt-in promotional mailings. It is not a cold B2B finding, it is thirteen years old, and it uses the open-rate metric Apple MPP broke in 2021. Campaign Monitor and hundreds of downstream posts repeat it without either caveat. Publish personalization's reply effect, not the borrowed opens number. (Campaign Monitor; Experian plc newsroom, 2014)
• The "30-45% more replies just from changing send time" claim is source laundering. It appears across cold-email vendor blogs with no dataset, no sample, and no method. The only transparent time-of-day data reviewed here (Belkins) shows a real but far smaller morning advantage on a sub-1% base. Do not publish the 30-45% number.
• The multichannel "287%" and "50% more replies" figures have no auditable origin. They surface in outbound-agency content. The one transparent multichannel figure reviewed here is Pin's 2-4x email-plus-LinkedIn lift in recruiting, on its own platform, which is directional and not a general B2B sales benchmark.

What the primary sources actually say

Reply rate is a fraction, and the denominator is the whole argument

Belkins published the most useful sentence in the category this year. Its 2025 reply rate looks "dramatically lower" than prior years, and it explained why in its own methodology note: earlier Belkins studies divided replies by unique openers, and this one divides replies by total emails sent. The stricter denominator is the entire reason 0.45% and "a 5% reply rate" can both be true of the same campaigns. The report is built on 7,530,489 emails and 34,393 tracked replies, excludes auto-replies and bounces, and reports 0.45% overall, 0.51% for the US subset, and 0.57% for founder-level contacts. (Belkins, June 26, 2026)

Instantly, publishing across billions of platform interactions from January 1 to December 18, 2025, reports 3.43% average replies per send, 5.5% at the top quartile, and 10%-plus for elite campaigns. It does not disclose an exact email count or a probability sample. Apollo, writing in May 2026 without a disclosed sample, calls 3-5% the normal range for a well-run campaign, 8%-plus good, and 1% a problem. (Instantly, January 12, 2026; Apollo, May 6, 2026)

These do not reconcile into one number, and they should not be forced to. Belkins divides by every send; Instantly and Apollo describe per-send replies on platform users who are self-selected for at least basic competence. Selection, deliverability, campaign mix, and denominator each move the headline by more than any subject-line tweak. The correct read is a range with its method attached, not a single figure. Anyone quoting "the average cold email reply rate is X%" without the denominator is quoting a fraction with the bottom torn off.

Personalization: the reply effect is real, the opens figure is imported

Two transparent cold datasets show personalization lifting the reply, which is the metric that survives Apple MPP. Pin found emails including the recipient's first name replied at 5.13% versus 2.61% without it, close to double, across a recruiting dataset of more than 4,000,000 messages from over 1,500 organizations sent June 2025 to May 2026. Belkins' 2025 subject-line study found personalized subject lines lifted reply rate from 3% to 7%, and open rate from 35% to 46%. (Pin, 2026; Belkins subject-line study, 2025)

Note the internal tension worth stating out loud: the Belkins subject-line study reports open rates, while the Belkins response-rate study disabled open tracking for the same year because opens are no longer trustworthy. Both can sit in the same house. Read the reply numbers as the load-bearing ones and treat the open numbers as directional.

The viral "26% more opens" belongs to a different world. Experian's 2013 study measured unique open rate on opt-in promotional email. It predates Apple MPP by eight years, it is not cold, and it is not B2B. The number is real and sourced. It is just answering a question nobody in cold outreach is asking. (Experian plc newsroom, 2014)

Send time: a real morning tilt, not a magic minute

Belkins' time-of-day data, on the same 7.5M-email base, found 8am-noon sends replied at 0.54% and early morning (5-8am) at 0.52%, while late-evening and night sends tied for last at 0.40%. It named Wednesday and Thursday mornings as the 2026 winning combination and flagged that its own prior studies had crowned late evening, a reversal it attributes to AI-assisted inbox triage. (Belkins, June 26, 2026)

So there is a real signal: mornings, mid-week, in the recipient's context. What is not supported is the specific "Tuesday 10am" and the claim that timing alone moves replies by 30-45%. The transparent spread here is 0.54% versus 0.40% on a sub-1% base. That is a genuine relative difference and a tiny absolute one, and it is the opposite of a copy-free growth lever.

Length: two honest datasets that disagree

Instantly reports its best-performing campaigns ran under 80 words. Pin reports 150-199 word bodies peaked reply rate in recruiting. Both samples are large. Both are transparent about what they measured. They point in different directions because they sample different populations with different objectives. That is the finding: "shorter always wins" is a preference dressed as a law, and the defensible instruction is to test length on your own list, not to adopt someone else's optimum. (Instantly, January 12, 2026; Pin, 2026)

The classic origin of "shorter wins," a widely cited Boomerang study putting the optimum around 75-100 words, could not be independently retrieved this session and is flagged below rather than repeated as fact.

Automation versus hand-written: manual won in the one transparent test

In Pin's recruiting data, recruiter-written one-off emails replied at 6.31% and automated sequences at 4.96%. In that dataset, hand-written beat automated. This is the cleanest available answer to "does automation win," and it says no, at least for the highest-response messages, at least in recruiting. It does not prove the reverse either. The honest verdict is that channel and craft interact with the audience, and the "AI-personalized beats manual" slogan is unproven at the level of generality it is usually stated. (Pin, 2026)

Deliverability: the one place with a published, non-negotiable standard

This is where the primary sources are firmest, and it is not a tactic, it is infrastructure. Since February 1, 2024, Google requires all senders to Gmail to authenticate with SPF or DKIM, and senders of more than 5,000 messages per day to Gmail to use SPF, DKIM, and DMARC, keep the Postmaster Tools spam rate under 0.3%, provide one-click unsubscribe, maintain valid PTR records, and use TLS. Apollo, echoing this, calls deliverability the primary constraint on cold email in 2026, ahead of copy or subject lines. (Google email sender guidelines; Apollo, May 6, 2026)

The "N inboxes, warmed N days" rules of thumb sit on top of that published floor. Instantly's guidance is representative: warm a new domain from 5-10 emails per day and ramp over 4-6 weeks, spread volume across multiple domains rather than concentrating it, and keep bounce under 2%. These are sensible vendor heuristics. They are not disclosed thresholds from a mailbox provider, and no primary source certifies a specific inbox count or warmup duration as the correct one. (Instantly, January 12, 2026)

Open rate after Apple MPP: broken as a precise number, not literally zero-signal

Apple's Mail Privacy Protection, released in iOS 15 and macOS Monterey in 2021, routes email image loads through Apple's proxy and preloads the tracking pixel, so any message sitting in Apple Mail is reported as opened whether or not the recipient engaged. Mailchimp states plainly that this produces "inflated and inaccurate open rates" and removes reliable location, device, and client data for those contacts. The strongest practitioner signal is behavioral: Belkins turned open tracking off entirely for 2025. "Open rates are dead" overstates it, because non-Apple opens still register and relative movement retains some use, but as a precise campaign metric, opens are no longer safe to optimize against. (Mailchimp Apple MPP FAQ; Belkins, June 26, 2026)

The legal claim: regulated, not banned (and this is not legal advice)

The "cold email is illegal" claim usually misstates the law in both directions.

In the US, CAN-SPAM (15 U.S. Code 7704) does not require prior consent. It prohibits materially false or misleading header information and deceptive subject lines, and it requires a functioning opt-out, a valid physical postal address, and identification of the message as an advertisement. It regulates how you send commercial email; it does not forbid sending it to people who never opted in. The penalties are steep and assessed per email: the FTC's 2025 inflation adjustment set the maximum at 53,088 US dollars per violating email, effective January 17, 2025, up from 51,744 in 2024 and 50,120 in 2023. A single campaign to thousands of recipients can therefore carry seven-figure exposure. The penalty ceiling is verified against the FTC's own notice; whether a specific campaign triggers it is a question for counsel. (15 U.S. Code 7704; FTC inflation-adjusted civil penalties, February 2025)

In the EU and UK, the picture is stricter but still not a ban. GDPR Recital 47 states that processing personal data for direct marketing purposes may be regarded as carried out for a legitimate interest, which is the legal basis most compliant B2B cold email relies on, subject to a documented balancing assessment. UK PECR, per the ICO, lets you email any corporate body (a company, LLP, or government body) without prior consent, while sole traders, some partnerships, individuals, and named personal addresses at a company are treated as individuals who must have consented or be existing customers under the soft opt-in. (GDPR Recital 47; UK ICO, PECR electronic mail marketing)

The defensible summary: cold email is lawful when it targets the right recipients and follows real, specific rules that vary by jurisdiction. The rules are the point, and they are not optional. Anything past this general framing is a question for a qualified professional, and this report does not give legal advice.

Primary versus inferred: what can go behind your name

First-party disclosures or transparent large-sample datasets, safe to state with method attached

• Belkins measured 0.45% replies per total send across 7,530,489 emails in 2025 (US 0.51%, founders 0.57%), on a replies-divided-by-sends denominator it adopted deliberately, and disabled open tracking. State the sample and the denominator every time. (Belkins)
• Instantly measured 3.43% average replies per send (5.5% top quartile, 10%-plus elite) across billions of 2025 interactions, exact email count undisclosed. (Instantly)
• Pin measured, in recruiting, automated email at 4.96%, recruiter-written one-off email at 6.31%, and LinkedIn messages at 17.08% across 4,000,000-plus messages, June 2025 to May 2026. First-name personalization: 5.13% versus 2.61%. Label it recruiting, not B2B sales. (Pin)
• Belkins found mornings (8am-noon) at 0.54% versus late evening at 0.40%, Wednesday and Thursday best. (Belkins)
• Instantly found under 80 words best-performing and 4-7 touchpoints the sequence sweet spot with 58% of replies on step 1; Pin found 150-199 word bodies peaked in recruiting. Cite both when discussing length; they disagree. (Instantly; Pin)
• Google requires SPF or DKIM for all senders and SPF, DKIM, DMARC, spam rate under 0.3%, and one-click unsubscribe for senders over 5,000 per day to Gmail, effective February 1, 2024. (Google)
• Apple MPP reports Apple Mail messages as opened regardless of recipient action, inflating open rates. (Mailchimp)
• CAN-SPAM regulates deception, opt-out, and disclosure and does not require prior consent; GDPR Recital 47 allows direct marketing on a legitimate-interest basis; UK PECR permits emailing corporate bodies without prior consent. (15 U.S. Code 7704; GDPR Recital 47; UK ICO)

Inference or myth: do not present as an established platform fact

• "The cold email reply rate is X%" stated without a denominator.
• "Personalized subject lines get 26% more opens" applied to cold B2B (it is 2013 promotional email).
• "Tuesday 10am is the best send time" and "timing alone lifts replies 30-45%."
• "Shorter emails always win" as a universal rule.
• "AI-personalized or automated always beats manual," or the strict reverse.
• "Multichannel lifts replies by 50% / 287%" as a general B2B benchmark.
• "You need exactly N inboxes warmed for exactly N days" as a disclosed threshold.
• "Open rates are completely unmeasurable" (they are unreliable and inflated, which is not the same as zero-signal).
• "Cold email is illegal."

Dated timeline

Steelman: why "cold email is a dying, sub-1% channel" might be closer to right than we allow

The strongest case against our "it depends on the denominator" framing has three parts, and it deserves an honest hearing.

1. The most transparent, largest-sample, most self-critical dataset in the category returned 0.45%. Belkins is not a skeptic trying to talk cold email down; it sells the service. It still chose the stricter denominator and published a sub-1% number with a note that its old higher numbers were flattering. When the most credible source volunteers the least flattering figure, that is evidence, not noise.
2. The infrastructure floor keeps rising. The 2024 Google and Yahoo requirements, the 0.3% spam ceiling, and MPP breaking the feedback metric all raise the cost and lower the margin of high-volume sending. A channel that needs more domains, more authentication, and more warmup to hit the same inbox is, at the margin, getting harder.
3. The high numbers cluster on self-selected platform users. Instantly's 3.43% and Apollo's 3-5% describe people already paying for outbound tooling and, presumably, already competent. That is survivorship. The true population average, including everyone sending badly, is likely closer to Belkins' floor than to the platform figures.

The honest synthesis:

Per send, across everyone, cold email reply rates are low and probably drifting lower as the infrastructure bar rises. That is not the same as "dead." A targeted, compliant, well-delivered campaign to the right list still returns replies at rates that clear the strict denominator by a wide margin, and the lever is selection and craft, not a viral tactic multiplier. The channel is narrowing, not closing.

The most authoritative sources

1. Belkins cold email response rates - June 26, 2026; 7,530,489 emails, 34,393 replies, transparent denominator change, timing and segment breakdowns, open tracking disabled. The reference dataset for this report.
2. Instantly cold email benchmark 2026 - January 12, 2026; large per-send dataset with reply-rate distribution, length, and sequence findings; exact sample undisclosed.
3. Pin recruiting outreach benchmark - 2026; 4,000,000-plus messages, cross-channel reply rates, personalization and length findings; clearly limited to recruiting.
4. Apollo cold email reply rates - May 6, 2026; practitioner range and deliverability framing, no disclosed sample.
5. Belkins B2B cold email subject-line study - 2025; personalization, question framing, and length effects on open and reply rate.
6. Google email sender guidelines - effective February 1, 2024; the authoritative deliverability standard for sending to Gmail.
7. Mailchimp Apple MPP FAQ - Apple MPP mechanics and its effect on open-rate reliability; pairs with Apple's 2021 newsroom announcement.
8. 15 U.S. Code 7704 (CAN-SPAM) - the statute itself; deception prohibitions, opt-out, and disclosure requirements.
9. GDPR Recital 47 - direct marketing as a potential legitimate interest, with the balancing test.
10. UK ICO, PECR electronic mail marketing - corporate versus individual rules and the soft opt-in, for the UK.
11. Experian Marketing Services 2013 study, via Experian plc newsroom - the origin of the "26%" personalized-subject-line figure, for provenance.

Open questions the primary sources do not answer

• A probability-sampled, cross-vendor B2B cold-email reply rate on a single agreed denominator. None exists; every figure is a vendor's own book of business.
• Whether the morning and mid-week send-time tilt holds outside Belkins' client mix, and by how much on a comparable denominator.
• A transparent, causal test of email length on cold B2B specifically, rather than two vendor observational datasets that disagree.
• A method-disclosed measurement of the multichannel email-plus-LinkedIn lift in B2B sales, not recruiting.
• Whether hand-written still beats automated outside recruiting, and at what volume the tradeoff flips.
• The real distribution of deliverability outcomes since the 2024 sender requirements, separated from copy and targeting effects.
• Jurisdiction-by-jurisdiction cold-email legality with current penalties, which is a legal question for counsel and outside this report's scope.

Verification pass, August 8, 2026

The eight figures the first research pass could not source were run through a second verification. Five are now sourced or corrected, three stay quarantined. None of the report's verdicts moved.

Verified and reinstated:

• The Boomerang email-length study is real. Boomerang's analysis of more than 40 million emails found 75-100 word messages replied at 51%, with 50-125 words the best band and response declining slowly past 125 words. Real and large. Caveat: it measured all email from Boomerang users, not cold B2B specifically, so it stands alongside Instantly's under-80-word and Pin's 150-199-word findings rather than settling them. "Shorter wins" has a real 40M-email source; a single cold-B2B optimum still does not exist. (Boomerang, 2016)
• The CAN-SPAM per-email penalty is 53,088 US dollars, verified, per the FTC's February 2025 inflation adjustment (up from 51,744 in 2024, 50,120 in 2023). Now stated in the legal section above. (FTC, February 2025)
• The Woodpecker 2022 send-time report is real: on 20 million-plus cold emails, the 9-11am local window delivered about 22% higher average response. Sourced, but it is Woodpecker's own 2022 book, and its effect size is far larger than the newer, bigger Saleshandy dataset finds. Treat as a dated, single-vendor data point. (Woodpecker)

Corrected:

• The Saleshandy send-time study is 53.1 million emails, not 100 million, and it argues against the timing hype. Saleshandy analyzed 53.1M cold emails and 60,000 sequences (January to June 2026) and did name Tuesday 9-10am as the top slot. It also reported that the gap between the best and worst sending day is only 2.1%, called timing "one of the smallest levers," and told readers to spend optimization time on targeting and copy instead. This is the largest send-time dataset located, and it strengthens rather than weakens the report's read: mornings and mid-week help a little, and the "30-45% from timing alone" claim is fiction. (Saleshandy, 2026)

Still unverified, stay quarantined:

• Smartlead's specific 2026 figures remain behind a gated PDF that was not opened. The broader 2026 cross-vendor median reply rate (roughly 2 to 4%, compressed from the 3-5% that was normal before 2023) is corroborated across several vendors, but Smartlead's exact numbers are not confirmed.
• The WarmySender "75,000 emails / Tuesday 9-11am 4.8%" figure was not located in a primary source. Saleshandy and Woodpecker now cover this ground with far larger samples; do not cite WarmySender.
• The viral "30-45% more replies from send time," "287% multichannel lift," and "50% more replies" figures still have no auditable source. They remain MYTH. The one real send-time effect size, Saleshandy's 2.1% day gap, is the opposite of a 30-45% lever.

Research notes and limitations

This report prioritizes 2025-2026 material and reads primary platform and regulator documents for deliverability and law. Its two strongest datasets, Belkins and Pin, are vendor books of business, not probability samples, and Pin is recruiting rather than B2B sales, so its channel comparisons transfer only as priors. Instantly and Apollo describe self-selected platform users. Every reply-rate figure in the category is denominator-dependent, and this report's central instruction is to read the denominator before the number.

The absence of a public source is not proof that an effect does not exist. It is proof that the number cannot responsibly be presented as verified. Where a figure could not be sourced this session, it is flagged above rather than ruled false.