Part three

Email: the rules became enforceable, and the open rate stopped meaning anything

The one channel where you own the list. It is also the channel whose headline metric stopped meaning anything in 2021, whose most quoted statistic is a survey of marketers, and whose rules became enforceable in 2024.

  • 55 to 60%

    Of all recorded email opens are affected by Apple's proxy, and are not reliable opens

    Litmus, May 2026 1

  • 11%

    The true open rate once mailbox provider inflation is stripped out, on a second dataset

    DMA UK, 9 July 2026 2

  • 0.3%

    The spam complaint rate that loses you Gmail mitigation. On a list of 1,000, that is three people

    Google 3

  • 5,000

    Messages a day to personal Gmail accounts, across your whole primary domain, and the rules bite

    Google 3

  • ~500

    Marketers self-reporting their own ROI in the survey behind "email returns $36 for every $1"

    Litmus, 16 July 2025 4

  • Google accepts DMARC p=none

    Its own words: "Your DMARC enforcement policy can be set to none." Anyone selling you enforcement as a Gmail requirement is overstating it. 3

  • There is no restaurant open rate

    Three vendors publish one. They say 18.5, 30.03 and 38.52 percent, and not one publishes a sector denominator. 5

  • Emoji and first names were associated with fewer opens

    Across 4.4 billion messages, in the opposite direction to the standard advice. 6

  • Apple replaces your preheader with a summary

    "Instead of previewing the first few lines of each email, they can see summaries." 7

  1. Litmus, Email Client Market Share , May 2026 data, over 1 billion opens. 2nd
  2. DMA UK Email Benchmarking Report 2026, 9 July 2026, read via its sponsor. 2nd
  3. Google, Email sender guidelines and the sender guidelines FAQ. Neither page carries a date. 1st
  4. Litmus, The ROI of Email Marketing , 16 July 2025. 2nd
  5. Campaign Monitor 2021 data, GetResponse 2023 data, Omnisend 2025 data. 2nd
  6. GetResponse, Email Marketing Benchmarks , 4.4 billion messages, 2023 data. 2nd
  7. Apple Newsroom, 10 June 2024. 1st

These are not best practices any more. They are conditions of delivery.

In February 2024 Google and Yahoo turned a decade of advice into requirements. Microsoft joined in May 2025. What follows is what each of them publishes, in their own words.

Requirement Google Yahoo Microsoft, Outlook.com
Who it applies to "Any email sender that sends close to 5,000 messages or more to personal Gmail accounts within a 24-hour period." Counted across the primary domain, so subdomains do not help "We will not specify a volume threshold." Assume it applies to you "Domains sending over 5,000 emails per day," enforced from 5 May 2025
Authentication SPF or DKIM for everyone. Bulk senders: both, plus DMARC "Implement SPF or DKIM at a minimum." Bulk senders: both, plus "a valid DMARC policy with at least p=none" SPF, DKIM and DMARC
DMARC strictness "Your DMARC enforcement policy can be set to none." At least p=none Not specified on the page we could read
Unsubscribe Must "support one-click unsubscribe, and include a clearly visible unsubscribe link," per RFC 2369 and RFC 8058. Processed within two days, per the 2023 announcement One-click unsubscribe. "The Post (RFC 8058) method is highly recommended" Not stated on the page we could read
Spam complaints Below 0.3 percent , recommended below 0.1 percent, calculated daily Below 0.3 percent. Denominator not defined Not stated
Failure Above 0.3 percent you are "ineligible for mitigation." From November 2025, "Gmail is ramping up its enforcement on non-compliant traffic" Enforcement from February 2024 Junk folder first, then rejection "until the DNS records are corrected"
  1. Google, Email sender guidelines and Email sender guidelines FAQ . Neither page publishes a last-updated date; the only dated changelog row is TLS, December 2023. Original announcement 3 October 2023. 1st
  2. Yahoo, Sender Best Practices and Sender FAQs . Neither page carries a date. Latest dates referenced on either are from 2024. 1st
  3. Microsoft, Outlook.com Postmaster policies, effective 5 May 2025. Microsoft's own announcement blog could not be read, so the requirement list is from the postmaster page. 1st
  4. IETF, RFC 8058, January 2017, and RFC 7489, March 2015. 1st

Your open rate has not meant anything since 2021.

Two separate measurements, from two different datasets in two different countries, published two months apart in 2026, arrive at the same conclusion. This is as close to settled as anything in this document gets.

What Apple actually built

Mail Privacy Protection, announced 7 June 2021, "stops senders from using invisible pixels to collect information about the user" and helps people "prevent senders from knowing when they open an email." 12

Read Apple's wording carefully. It does not say opens are blocked. It says the sender cannot know. Mechanically Apple's proxy fetches the tracking pixel whether or not a human opens the message. That is why open rates went up after 2021, not down.

The two measurements

  • Litmus, May 2026, over 1 billion tracked opens. "Apple's MPP now impact roughly 55-60% of all email opens," and those opens "are not considered reliable opens." Apple accounts for 64.66 percent of tracked opens 13
  • DMA UK, 9 July 2026, six email service providers. The true open rate is "just 11 percent" once inflation is stripped, and "roughly three-quarters of recorded opens may not reflect a real, human open at all" 14

What that does to a number you recognise

If 55 to 60 percent of recorded opens are machine pre-fetches, a reported 30 percent open rate sits somewhere near 12 to 14 percent of humans, before you account for anything else. That arithmetic is ours, not Litmus's: Litmus publishes the share of opens affected and does not do the conversion. The DMA's 11 percent, from an entirely different dataset, lands in the same place. 13 , 14

  1. Apple Newsroom, 7 June 2021. Apple's own support documentation on the feature is JavaScript rendered and could not be read on 8 August 2026. 1st
  2. Litmus, Email Client Market Share , May 2026 data, calculated from over 1 billion opens in Litmus Email Analytics. Vendor panel. 2nd
  3. DMA UK, Email Benchmarking Report 2026 , published 9 July 2026, read via Validity's summary of 17 July 2026. Denominator not published. 2nd

There is no restaurant email benchmark. There are three, and they disagree by more than two to one.

Every one of these is a vendor measuring its own customers. Not one publishes a denominator for the sector row it is quoted for.

Publisher Sector open rate Data period Whole-panel sample The problem
Campaign Monitor 2nd 18.5% 2021
51 months old 		Over 100 billion emails, 2021 Sector row denominator not published. Its data predates Apple's proxy saturating the numbers, which is very likely why it is the lowest of the three and probably the closest to a real human open rate
GetResponse 2nd 38.52% 2023
32 months old 		4.4 billion messages, active senders with at least 500 contacts Sector row denominator not published. GetResponse states the caveat itself: growth "can partially be attributed to the growing popularity of Apple's devices, which distort the data with its auto-opens"
Omnisend 2nd 30.03% 2025
3 months old 		Over 20 billion campaign emails, 27,000-plus brands Sector row denominator not published. And "Food and Drink" in an ecommerce panel means food sold online, not restaurants. It is the closest row they publish, and that is all it is
Mailchimp 2nd No sector row at all Campaign and subscriber counts not published. Only stated rule: campaigns to at least 1,000 subscribers Its own page says the data "was last updated in December 2023," 32 months ago. The industries shown are Business and Finance, Non-Profits, Education, Ecommerce and All Users. Any article quoting a "Mailchimp restaurant open rate" is quoting a table Mailchimp does not publish

One number worth keeping, with its confound printed on it

Omnisend's 2025 panel: automated emails earned $3.41 per email sent, against $0.155 for broadcast campaigns. Click-to-open was 11.87 percent against 2.44 percent. That is not "automation is twenty-two times better." Automated mail fires at a moment of intent, to a self-selected person who just did something, which makes the comparison confounded by design. It also gets unsubscribed from nearly three times as often, 0.59 percent against 0.20. What survives is directional and still worth acting on: mail triggered by something the customer did outperforms mail sent because it is Tuesday. 15

  1. Omnisend, Email marketing benchmarks , published 12 May 2026, 2025 data. 2nd
  2. GetResponse, Email Marketing Benchmarks , published 2024, 2023 data. 2nd
  3. Campaign Monitor, Email Marketing Benchmarks , published 31 January 2023, 2021 data. 2nd
  4. Mailchimp, Email Marketing Benchmarks , data last updated December 2023. 2nd

The standard advice, checked.

Read the whole of this page with one caveat attached: every study below is a vendor measuring correlations across its own customers. None randomises. None is an experiment. "This is what happened across GetResponse's customer base" is not "this is what will happen to you."

The advice What the data shows The detail, and what it is worth
Send on the best day Barely matters GetResponse's own conclusion across 4.4 billion messages: "There's little difference in terms of engagement between individual weekdays. While the average open and click-through rates are the highest on Tuesdays, they aren't that far different from what we see on the other days." 19
Send at the best time Unmeasurable now The two reported peaks, 4 to 6am and 5 to 7pm, are almost certainly an artefact of when senders schedule rather than when readers read. And any timing signal derived from opens is scrambled by proxy pre-fetching anyway 19
Mail more often Once a week engaged best Open and click rates fall from one newsletter a week to three, then go noisy: the eleven-a-week cohort shows a higher open rate than the three-a-week cohort. That noise is the tell. It compares different senders, not the same sender changing frequency 19
Put a first name in the subject line Ran the other way 35.78 percent open for personalised subject lines against 41.87 percent for non-personalised, across the same 4.4 billion messages. Do not over-read that in reverse either: the kind of sender who does this may simply be a worse sender. What survives is the negative: no published evidence here that it helps 19
Add an emoji Ran the other way 37.5 percent with an emoji against 42.23 percent without. Same caveat, same conclusion 19
Segment your list On a study from 2017 "Clicks 100.95 percent higher than non-segmented campaigns," from roughly 2,000 users and nearly 9 million recipients. Published 1 February 2017 , which is 114 months ago. It predates Apple's proxy entirely. We found no segmentation study since with a published denominator 20
Plain text beats designed HTML No test exists We looked for one from every vendor, provider and independent source we could reach and found nothing with a stated method and sample. Unevidenced in both directions 21
Use double opt-in Never measured GetResponse publishes adoption only, 10.99 percent double against 89.01 percent single, and explicitly declines a performance comparison. No measurement of the difference was found anywhere 21
  1. GetResponse, Email Marketing Benchmarks , published 2024, 2023 data, 4.4 billion messages from senders with at least 500 contacts. 2nd
  2. Mailchimp, Effects of List Segmentation on Email Marketing Stats , 1 February 2017. 2nd
  3. Searched 8 August 2026. Absence of evidence, stated as such. n/a
  4. Google, Postmaster Tools documentation. No publication date on the page. Retrieved 8 August 2026. 1st

The US and the UK are opposites, and people get this wrong in both directions.

In the US you may email someone who never asked, provided you follow the rules. In the UK you generally may not. Anyone operating in both needs to know which set they are under before they send.

United States: CAN-SPAM is opt-out

There is no federal requirement to obtain consent before sending commercial email in the US. That is the single biggest legal difference from the UK. What is required, per the FTC's own compliance guide: 23

  • "Don't use false or misleading header information"
  • "Don't use deceptive subject lines"
  • "Identify the message as an ad"
  • "Tell recipients where you're located," meaning a valid physical postal address in every message
  • "Tell recipients how to opt out," and "Honor opt-out requests promptly"

The FTC states the penalty as up to $53,088 per individual violating email. That figure is current on the FTC's page as last modified 22 July 2025, and is still in the regulations as of 6 August 2026. We could not confirm a January 2026 inflation adjustment, so treat it as the published figure rather than a verified-current one. 23

United Kingdom: PECR is consent, with one exception

The ICO: "you must not send electronic mail marketing to individuals, unless they have specifically consented." The exception is the soft opt-in, and it has four conditions, all of which must hold: 24

  • You got their details "in the course of a sale or negotiations for a sale"
  • You are marketing "your own similar products or services"
  • You gave a simple opt-out when you first collected the details
  • You give a simple opt-out in every single message

"Negotiations for a sale" is the part most operators miss. Somebody who enquired and did not book can be covered. Somebody who left a business card cannot.

And the B2B split matters: "You can email or text any corporate body." A sole trader or an ordinary partnership counts as an individual and cannot be cold-emailed. That is a Companies House lookup before you send, not a guess. 24

  1. Federal Trade Commission, CAN-SPAM Act: A Compliance Guide for Business , published 9 August 2023, last modified 22 July 2025. Penalty corroborated in the eCFR, current to 6 August 2026. 1st
  2. Information Commissioner's Office, Electronic mail marketing , Guide to PECR, page dated 19 November 2024. 1st
  3. ICO, Guide to PECR landing page, retrieved 8 August 2026; Data (Use and Access) Act 2025 (c. 18), Part 5 Chapter 2, section headings only. 1st

Nobody's AI will discover your newsletter. Their AI will decide whether they read it.

Email has no URL, no index entry and no public retrieval path, so it cannot appear in an AI answer. What has changed is what happens between the inbox and the reader.

Why "optimise your email for AI search" is a category error

AI assistants surface content from the open web and from data a user has explicitly connected. A marketing email is neither. Google states of Workspace content: "Your content is not human reviewed or otherwise used for Generative AI model training outside your domain without permission." 26

Two limits on that, stated honestly. Google's wording is framed around business and education customers, and we found no equivalent first-party wording for consumer Gmail accounts. And an assistant a person has connected to their own mailbox absolutely can read and summarise your email for them . That is not your email appearing in an AI answer. It is their assistant reading their own inbox.

What is actually sitting between you and the reader

  • Apple, 10 June 2024. "Instead of previewing the first few lines of each email, they can see summaries without needing to open a message." Priority Messages puts "the most urgent emails" in a section at the top of the inbox 27
  • Microsoft, updated 20 January 2026. Copilot in Outlook can "Summarize an email thread" and "Prioritize my inbox," including "Which unread emails are most important?" 28
  • Google, updated 11 September 2025. Gmail "automatically groups and highlights emails in the Promotions tab that people are most likely to engage with," and publishes annotation markup that changes how a promotion displays 29

What follows, concretely

Preheader text is worth less than it was, because on an iPhone it is replaced by a machine's paraphrase of your email. An email built around a designed image with very little text summarises into nothing. A plain, concrete, single-subject email summarises into a sentence somebody might act on. None of that is a new trick. It is the same thing the spam thresholds reward: send one clear thing, to people who want it.

  1. Google, Generative AI in Google Workspace Privacy Hub , 26 May 2026. 1st
  2. Apple Newsroom, Introducing Apple Intelligence , 10 June 2024. Apple's own support pages on Mail categories could not be read. 1st
  3. Microsoft, Copilot in Outlook , last updated 20 January 2026. 1st
  4. Google for Developers, Gmail Promotions tab annotations , last updated 11 September 2025. 1st

Six claims checked against the source.

Email has the oldest evidence base of any channel in this document. Several of the numbers still being quoted in 2026 were measured before the thing that broke them.

The claim Verdict What the source actually says
Email returns $36 for every $1 spent A survey of marketers Roughly 500 marketing professionals estimating their own return. And it is not one number: 35 percent said $10 to $36, 30 percent said $36 to $50, 5 percent said over $50. That is 70 percent. The missing 30 percent presumably reported under $10 and are not in the headline. 30
A good open rate is around 30 percent Measures a proxy server Roughly 55 to 60 percent of all recorded opens are affected by Apple's proxy and are "not considered reliable opens." A second dataset puts the true rate at about 11 percent. Report clicks and what happened next. 31
You must get DMARC to enforcement to reach the inbox Google says otherwise Google's own requirement page: "Your DMARC enforcement policy can be set to none." Yahoo asks for "at least p=none." The providers require the record to exist, not the enforcement. 32
Personalised subject lines and emoji lift opens Associated with fewer Across 4.4 billion messages: 35.78 percent with a personalised subject line against 41.87 without, and 37.5 percent with an emoji against 42.23 without. Correlation across different senders, not a test, and it runs the opposite way to the advice. 33
Segmentation doubles your clicks True in 2017 "Clicks 100.95 percent higher," from roughly 2,000 users and 9 million recipients, published 1 February 2017. That is 114 months ago and it predates everything that has since changed about how email is delivered and measured. 34
Optimise your emails to be found by AI Not a thing Email has no URL, no index entry and no public retrieval path. No documented mechanism exists by which a marketing email appears in a public AI answer. What does exist is an assistant summarising your email for the person who already received it. 35
  1. Litmus, The ROI of Email Marketing , 16 July 2025, from the Litmus 2025 State of Email Survey. 2nd
  2. Litmus, May 2026, over 1 billion opens; DMA UK, 9 July 2026. 2nd
  3. Google and Yahoo sender documentation. 1st
  4. GetResponse, 2023 data, 4.4 billion messages. 2nd
  5. Mailchimp, 1 February 2017. 2nd
  6. Google Workspace Privacy Hub, 26 May 2026, plus the absence of any documented retrieval path. 1st

Questions people ask about this chapter

Every answer below is the chapter's own wording, with a link to the page it sits on.

Do the Gmail and Yahoo sender rules apply to a small list?

The safe reading for a small sender is to assume the rules apply at any volume, because Yahoo says they might.

Read the full answer
Is the open rate still worth tracking?

If you keep open rate for anything, keep it as a trend line on your own list, where the inflation is at least roughly constant, and never compare it to a published benchmark.

Read the full answer
What should I report instead of opens?

Report clicks, replies and what happened afterwards: bookings, orders, enquiries.

Read the full answer
What is the industry email benchmark for restaurants?

Use your own last twelve months as the comparison and ignore the tables.

Read the full answer
Do I need consent before emailing in the US?

There is no federal requirement to obtain consent before sending commercial email in the US.

Read the full answer
Will AI recommend my newsletter?

Email has no URL, no index entry and no public retrieval path, so it cannot appear in an AI answer.

Read the full answer

Sources

  1. Litmus, Email Client Market Share , May 2026 data, over 1 billion opens. 2nd
  2. DMA UK Email Benchmarking Report 2026, 9 July 2026, read via its sponsor. 2nd
  3. Google, Email sender guidelines and the sender guidelines FAQ. Neither page carries a date. 1st
  4. Litmus, The ROI of Email Marketing , 16 July 2025. 2nd
  5. Campaign Monitor 2021 data, GetResponse 2023 data, Omnisend 2025 data. 2nd
  6. GetResponse, Email Marketing Benchmarks , 4.4 billion messages, 2023 data. 2nd
  7. Apple Newsroom, 10 June 2024. 1st
  8. Google, Email sender guidelines and Email sender guidelines FAQ . Neither page publishes a last-updated date; the only dated changelog row is TLS, December 2023. Original announcement 3 October 2023. 1st
  9. Yahoo, Sender Best Practices and Sender FAQs . Neither page carries a date. Latest dates referenced on either are from 2024. 1st
  10. Microsoft, Outlook.com Postmaster policies, effective 5 May 2025. Microsoft's own announcement blog could not be read, so the requirement list is from the postmaster page. 1st
  11. IETF, RFC 8058, January 2017, and RFC 7489, March 2015. 1st
  12. Apple Newsroom, 7 June 2021. Apple's own support documentation on the feature is JavaScript rendered and could not be read on 8 August 2026. 1st
  13. Litmus, Email Client Market Share , May 2026 data, calculated from over 1 billion opens in Litmus Email Analytics. Vendor panel. 2nd
  14. DMA UK, Email Benchmarking Report 2026 , published 9 July 2026, read via Validity's summary of 17 July 2026. Denominator not published. 2nd
  15. Omnisend, Email marketing benchmarks , published 12 May 2026, 2025 data. 2nd
  16. GetResponse, Email Marketing Benchmarks , published 2024, 2023 data. 2nd
  17. Campaign Monitor, Email Marketing Benchmarks , published 31 January 2023, 2021 data. 2nd
  18. Mailchimp, Email Marketing Benchmarks , data last updated December 2023. 2nd
  19. GetResponse, Email Marketing Benchmarks , published 2024, 2023 data, 4.4 billion messages from senders with at least 500 contacts. 2nd
  20. Mailchimp, Effects of List Segmentation on Email Marketing Stats , 1 February 2017. 2nd
  21. Searched 8 August 2026. Absence of evidence, stated as such. n/a
  22. Google, Postmaster Tools documentation. No publication date on the page. Retrieved 8 August 2026. 1st
  23. Federal Trade Commission, CAN-SPAM Act: A Compliance Guide for Business , published 9 August 2023, last modified 22 July 2025. Penalty corroborated in the eCFR, current to 6 August 2026. 1st
  24. Information Commissioner's Office, Electronic mail marketing , Guide to PECR, page dated 19 November 2024. 1st
  25. ICO, Guide to PECR landing page, retrieved 8 August 2026; Data (Use and Access) Act 2025 (c. 18), Part 5 Chapter 2, section headings only. 1st
  26. Google, Generative AI in Google Workspace Privacy Hub , 26 May 2026. 1st
  27. Apple Newsroom, Introducing Apple Intelligence , 10 June 2024. Apple's own support pages on Mail categories could not be read. 1st
  28. Microsoft, Copilot in Outlook , last updated 20 January 2026. 1st
  29. Google for Developers, Gmail Promotions tab annotations , last updated 11 September 2025. 1st
  30. Litmus, The ROI of Email Marketing , 16 July 2025, from the Litmus 2025 State of Email Survey. 2nd
  31. Litmus, May 2026, over 1 billion opens; DMA UK, 9 July 2026. 2nd
  32. Google and Yahoo sender documentation. 1st
  33. GetResponse, 2023 data, 4.4 billion messages. 2nd
  34. Mailchimp, 1 February 2017. 2nd
  35. Google Workspace Privacy Hub, 26 May 2026, plus the absence of any documented retrieval path. 1st