Key takeaway
An email marketing best practice is a rule about who you mail and how, and the list splits in two. Four are enforced by the inboxes: authenticate the sending domain, hold complaints below 0.10% and never at 0.30%, ship a one-click unsubscribe that answers a POST, honour opt-outs within days. Everything else is a preference that moves a number. Most guides print both at one weight, so I've watched teams reach the four that decide delivery somewhere around item 22.
What a best practice in email actually means now.
An email marketing best practice is a repeatable rule about who you mail, what you send them, and how you send it, adopted because it reliably improves a result. That's the definition every guide on this topic works from, and it's correct as far as it goes. What it hides is that the phrase now covers two different kinds of thing that got shuffled together. A few of these rules are enforced by the companies that run the inboxes, and breaking one costs you delivery. The rest are preferences that move a rate by a few points, and you could skip any of them for a year without anything breaking.
Almost every page on this subject prints both kinds in one numbered list, ordered by nothing in particular. So somebody reads one, starts at the top, spends the first afternoon rewriting subject lines, and reaches, somewhere around number 22, the four items that decide whether any of the mail arrives at all. I've watched that happen more times than I'd like, and the ordering is the whole problem: the lists aren't wrong, they're flat.
A rule only improves a result if it's applied every time. Most of what gets called best practice is applied when somebody remembers to apply it, which is reliably in month one and rarely in month nine.
Marketing budgets matter less than time and focus. Small teams spread themselves too thin across untested platforms. The reality is much simpler. You must find the channels "bringing in 80 to 90%" of your revenue and double down on them. I made this point on In the Ring with SUMO Heavy. For modern brands, "almost always email will be part of that mix" alongside one or two other core channels.
The four rules that stopped being advice.
Four items on that list aren't advice any more, and they're worth doing in this order before anything else on this page. The penalty for missing one isn't a worse open rate, it's the junk folder or a rejection at the gateway.
- Authenticate the sending domain. SPF and DKIM, plus DMARC once you send volume. Google treats more than 5,000 messages a day to Gmail accounts as bulk sending, and asks bulk senders for all three in its sender guidelines.
- Hold the spam complaint rate down. Google publishes the numbers: below 0.10%, and never at 0.30% or higher. Yahoo asks for below 0.3% in its own sender guidance. It's a rate, so the list decides it, not the copy.
- Ship a one-click unsubscribe that works. An HTTPS
List-UnsubscribeURI, aList-Unsubscribe-Postheader beside it, both DKIM-signed, and an endpoint that answers a POST with no cookies and no redirect, as RFC 8058 sets out. - Honour the opt-out fast. Yahoo asks bulk senders for two days. Build the suppression check into the send path rather than into a nightly job, and the number stops mattering.
A footer link into a preference centre isn't the third item, and a suppression list somebody exports once a month isn't the fourth. Both are working endpoints or they don't count. The records and the header syntax are a separate job, and the deliverability guide covers them properly.
Clearing all four gets the mail accepted. None of it gets the mail read. The variable I've seen move that complaint number fastest is frequency: three or four emails a week to a list used to getting almost nothing is the quickest way past 0.30%.
Automation solves the majority of revenue problems for online stores. We see this constantly. I laid out the math on SaaS District. You only need to "turn on personalized welcome series, personalized win-back series, personalized abandoned carts, and maybe personalized reorder" flows. That tight framework is enough "to get 80% of it right" before you ever touch a manual broadcast.
Permission is the practice the others depend on.
Permission is a record that this specific address asked for this specific mail, with a timestamp and a source you could produce if somebody asked for it. It isn't an assumption that a customer wants the newsletter because they bought something once.
Every rule in the section above is a rate, and permission sits in the denominator of all of them. A bought or scraped list raises the complaint rate and the bounce rate at the same time, which is the one combination nothing rescues: the copy can't fix it, the send time can't fix it, and a better subject line only gets the message opened by more people who never asked for it. That's not a courtesy to the reader. It's arithmetic about your own sending, and it belongs first.
Consent collected once and never re-checked decays. An address that opted in three years ago and hasn't opened anything since is technically permitted, and to a mailbox provider it behaves exactly like an address that never asked. Permission is a state with an age on it, not a checkbox in a signup form.
Double opt-in is the usual answer to that, and it's a trade rather than a commandment. It costs you signups, including real ones who never get round to the confirmation click, and it buys a list whose complaint rate you can predict. At volume, the predictability is worth more than the addresses.
Relevance is a targeting problem, not a copywriting one.
Segmentation is dividing the list by something you already know, so that a message only goes to the people it's for. Personalisation is the same move at the level of one record. Both are queries over data you hold.
Most guides treat them as two separate practices and get both slightly wrong. Inserting a first name into a subject line isn't personalisation, it's a merge field, and readers stopped reading it as attention a long time ago. The decision that moves a result is who gets excluded from the send. That's a question about what you know, not a writing exercise: last purchase, last open, plan, region, whether they already have the thing you're about to announce.
How often to send is almost always a targeting question wearing a scheduling costume. The complaint doesn't come from the person who wanted four emails and got five, it comes from the person who should never have been in the send at all, and mailing everybody less is a blunt fix for that.
A segment is a snapshot, and it's wrong the moment it's exported. Somebody unsubscribes on Tuesday and the file you pulled on Monday doesn't know.
Copywriting cannot fix a fundamentally irrelevant offer. The most effective messages rely entirely on customer behavior. I told listeners on QA Selling Online that our top performers are simple grids "based on what you've bought, added to card or liked" in the past. As of October 2020, these basic recommendation blocks generate the highest "revenue per email sent" across our entire platform.
The practices that are preferences, not rules.
Everything below moves a number rather than deciding whether the mail is delivered. All five are worth doing well, and none of them will save a send that failed one of the four rules above.
- Subject line: the field the reader actually decides on, and the only argument worth having about it is clarity against curiosity, which your own list will settle faster than any rule of thumb.
- Preview text: the second line the inbox renders, usually left as whatever the template's first characters happen to be, which is often an image alt attribute or the words "view in browser".
- Mobile rendering: how the message behaves in a narrow column, which is where a lot of it gets read, and the reason a two-column layout with a small font is a decision rather than a default.
- From name: the field readers recognise before they read anything else, and the reason a no-reply address costs more than it saves, since it teaches people to treat your mail as machinery.
- Send timing: worth testing on your own list and worth very little taken from somebody else's, because the answer turns on who is on that list and what their week looks like.
Doing all five well is still five things somebody has to remember before each send.
What tells you a practice is working.
The numbers in a campaign report split into two kinds, and they answer different questions. Complaint rate, bounce rate, and unsubscribe rate tell you whether you're allowed to keep sending. Click rate and whatever conversion means in your business tell you whether the mail did anything.
Open rate sits awkwardly between them and carries less signal than its prominence suggests, because a privacy proxy can fetch the tracking image without a human ever seeing the message. It's still useful as a relative measure inside one list over time. It's a poor number to make a decision on and a worse one to report upward.
The pair worth looking at weekly is complaint rate and bounce rate, because those are the two the mailbox providers are also looking at. Everything else can wait for the monthly. If the complaint rate is drifting up while the list is growing, the growth is the problem, and no amount of subject-line work shows up against it.
All of these are lagging. By the time the complaint rate moves, the send that caused it has finished and the recipients have already formed their opinion. Reporting tells you what happened. It doesn't stop it happening.
List growth rate rarely gets the attention it deserves. I argued on SaaS District that aggressive collection is the only way to scale. A standard site might convert at 1%. Proper collection changes that math. You end up "changing that conversion rate from 1% to like 3% or 4%" simply by capturing the traffic you already paid for.
A practice you have to remember isn't a practice.
In every guide on this subject, the four rules above included, a best practice is an instruction to a person. Clean the list. Check consent. Confirm it renders. Suppress the people who asked out. Read as instructions, all of it is reasonable.
An instruction to a person is applied at the rate a person applies it. That rate is high in the first month, when the checklist is new and the volume is small, and it drops with every send after that. It drops fastest exactly when it matters most, which is when volume rises and somebody is shipping four campaigns a week against a deadline. I have never seen a pre-send checklist survive its first busy quarter.
The version that survives is the one expressed as something the system does on every send, whether anyone is watching or not. Suppression as a state on the contact rather than a step before the send. A segment as a stored query evaluated when the send runs, rather than a file that was correct on Monday. A pre-send check as a call that returns a result, rather than a printout somebody works down. Every platform in this category was built as a dashboard first, and a screen is a thing a person has to visit. An AI-native email platform inverts that: the endpoint and the agent tool are the product, and the screen is one more client of them. That's why we build every capability as an API endpoint and an MCP tool before it's a screen, rather than adding an agent to a dashboard afterwards. An agent bolted onto a dashboard-first product still has to drive the screens somebody else designed for a human, so the rule stays a step somebody takes. A rule in the call path runs on every call.
You can't automate taste. Nothing decides for you whether the message was worth sending at all, which is why the preferences stay a list of preferences and never become a fifth rule.
If those four rules are currently four things somebody remembers, that's the gap worth closing first, and it closes by moving them into the send itself rather than into a longer checklist. Nitrosend specialises in email, and three of those four run on mechanisms the send carries out: the signing, the unsubscribe endpoint, and suppression on receipt. The complaint rate stays yours, because who's on the list decides it. You run the whole stack from one agent command. Start on the free tier and put something real through it.
Go deeper
Sources
- Google, Email sender guidelines: the bulk-sender threshold of more than 5,000 messages a day to Gmail accounts, the SPF, DKIM, and DMARC requirement, and the 0.10% and 0.30% spam-rate numbers.
- Yahoo, Sender best practices: the spam rate below 0.3%, the authentication requirements for bulk senders, and honouring unsubscribes within two days.
- RFC 8058, Signaling One-Click Functionality for List Email Headers: the header pair, the DKIM signature over it, and the POST endpoint behind one-click unsubscribe.
Common questions
Four of them are enforced rather than advised: authenticate the sending domain with SPF, DKIM, and DMARC, keep the spam complaint rate below 0.10%, ship a one-click unsubscribe header that answers a POST, and honour opt-outs fast. Everything after that, subject lines and segmentation included, is optimisation that moves a number rather than deciding whether the mail arrives.
If you send more than 5,000 messages a day to Gmail accounts, yes, all three. Below that, mailbox providers still ask every sender for SPF or DKIM at minimum, and DMARC is cheap to add once the first two exist. There's no volume at which unauthenticated marketing mail is a good idea.
Below 0.10% is the number to run at, and 0.30% is the ceiling you never want to approach. Yahoo asks for below 0.3%. It's a rate, so it's set by who is on the list rather than by the copy, and if it drifts up while the list grows, the growth is what to look at.
Two days is the practical answer, because that's what Yahoo asks of bulk senders. Design to anything slower and a mailbox provider will have drawn its own conclusion long before you act. Suppress on receipt, in the send path, and the question stops being one anybody tracks.
There isn't an honest single number, because it turns on what you sell and what people signed up for. Treat cadence as a targeting question instead. Complaints come from people who shouldn't have been in the send rather than from people who got one email too many, so tighten who receives each campaign before you cut the calendar.
It's a trade, not a rule. Confirming the address costs you signups, including real people who never click the confirmation link, and it buys a list where the complaint rate and the bounce rate are predictable. At volume that predictability is worth more than the extra addresses. On a small, hand-built list it often isn't.
Treat it as a sequence. Segment by last engagement, send the next campaign to the engaged part only, then re-engage the quiet part in small volumes across several sends while you watch the complaint and bounce rates. Suppress whatever doesn't respond and leave it suppressed. Mailing the whole dormant list once to see who is alive is the move that does the damage.
Usually yes, on separate subdomains. The two classes earn reputation at different rates: receipts and password resets get opened and rarely complained about, campaigns are neither, so mixing them lets one drag the other down. <a href="https://nitrosend.com/transactional-email/best-practices">Transactional email has its own practices</a> worth reading separately.