0.3percent
Why emails go to spam: email deliverability for a small business
Why emails go to spam, shown on one invented heating and cooling company whose monthly note goes to fourteen hundred past customers and draws five spam complaints. What the CAN-SPAM Act requires and what it does not, the 0.3 percent ceiling Google and Yahoo both publish and why no sender can work out its own rate, what SPF, DKIM and DMARC really prove, what an open rate measures, and seven checks to run on your own sending in an afternoon.
Why emails go to spam has a published answer: the mailbox provider decides they are unwanted or cannot tell who sent them, and Google and Yahoo publish what they require of every sender, chief among it a sending domain that proves who it is and a spam complaint rate below 0.3 percent. This page gives you, free, the method we use when we set up email marketing for a small business: the rules in their own words, the arithmetic of that ceiling and an afternoon's audit of your own sending. Below, one invented heating and cooling company sends its monthly note to fourteen hundred past customers, and you will see why the obvious fixes, a better subject line or a new sending tool, are the wrong first move, and why the cheapest lever on email deliverability is a shorter list. The method works the same for a plumbing company, a dental office, a salon or a real estate team.
Tony's heating and cooling company (invented). Tony owns a heating and cooling company with six people: himself, Beth in the office and four technicians. Once a month Beth sends a note to everyone who has ever booked a job: fourteen hundred people. Tony, Beth, their customers and every figure about the company are made up; their sending runs through this page as one worked example of a real method.
Tony's heating and cooling company (invented). The October note went out on a Tuesday morning. It was a decent piece of work: when to change a furnace filter, what a tune-up checks, and two sentences at the bottom in Tony's own voice about a cracked heat exchanger one of his technicians had found the week before. Nobody wrote back to complain. Five people pressed one button, the one that reports a message as spam. Pressing that button is a complaint of a kind that takes under a second, starts no conversation, and never reaches Tony.
Tony's heating and cooling company (invented). The November note went to the same fourteen hundred people. The sending tool reported that it had been sent to fourteen hundred people, which is what the sending tool can see. Some number smaller than fourteen hundred received it. A smaller number than that ever saw it in the place where a person reads things. Nobody at the company looked: the five complaints sat in a column of the sending tool's report that nobody read, and nothing in that report could show the notes that never reached an inbox. When the replies thinned out, Beth's first idea was a livelier subject line; her second was a new sending tool.
Everything on this page is free to use: the seven checks, the rules quoted from the documents that set them, and the calculator. What we sell is setting up the sending and the audience rules with you, and making sure somebody owns them afterwards.
In short
- The federal statute that governs commercial email does not require anybody's permission to send it. It requires that the message says what it is. It must carry a physical address. And it must offer an exit that keeps working for at least thirty days. Permission is not the law's demand. It is the mailbox providers' demand, and they are stricter.
- Google and Yahoo both publish the same ceiling for how often people may report you: three tenths of one percent. On Google's page it sits in the requirements for all senders rather than in the bulk section. So a one person office is inside it. Three complaints in a thousand delivered messages is the line itself.
- The number you cannot work out is your own. Yahoo states that the rate is calculated in their system on mail delivered to the inbox. That is a denominator no sender can see. What you can do is send less, to fewer people, more carefully. That is the entire craft.
Three things this is not
The neighbours own most of this street.
Not the busywork
A chain can fire when a form is submitted, so that nobody has to copy a name into a second system. That is a real and valuable thing, and it is a different product. That question is about work. It is answered by whether a step a person used to do now happens without them. Nothing in this article will save you an hour.
Not the old list
Going back to people who contacted you years ago and went quiet is its own project with its own rules. Those rules have dates in them. That question is about permission that may have gone stale. This one assumes the permission is current. It asks the harder thing: what you send those people, how often, and whether it reaches them.
Not the ask
Asking for a review at the right moment is a single message with a single job. It has a bright legal line running through the middle of it. That question is about one moment. This one is about the two hundred moments in between. There is no event to react to, and you are deciding whether to be in someone's morning at all.
What is marketing automation, once the busywork is taken out?
Marketing automation, put plainly, is a standing instruction about people you are not currently talking to: a rule written once about who hears from you, what they hear and when, which then runs for years without anyone looking at it. It is not a task that runs itself, and not a message you decide to send. Three other pages on this site cover the nearby ground, the busywork, the old list and the review request, and the cards above say where each one stops.
That changes what the risks are. A workflow that copies a name into a second system either works or does not, and you find out on the day. A rule about who deserves a message from you keeps running while the people it describes change, and you can go years without finding out.
What a campaign decides for you
Four decisions, and you only ever edit the second.
Who it goes to
The audience is a query, and a query written once runs forever against a database that keeps changing. The customer whose furnace was replaced last spring still matches the needs-a-new-furnace segment, because nobody wrote the rule that takes them out of it. When a campaign embarrasses someone, look first at who it reached, and only then at what it said. And the audience is the half nobody opens again.
What it says
The only part anybody edits, and the part that carries the least risk when it is wrong. A dull message to the right person on the right morning is forgettable. A brilliant message to a customer whose job you finished in March reads as evidence that you do not know who they are. Being forgotten is a far cheaper outcome than that.
When it arrives
Behaviour triggers move this from a calendar to a reaction. That is a real improvement, and it is also where the strangeness comes from. Someone looks at the page about new boilers on a Sunday evening. What arrives on Monday tells them exactly how closely they are being watched. How long to wait is a design decision, and it is a decision whether or not anyone makes it on purpose.
Whether it arrives at all
This is the one nobody chooses. It is decided by a company you have no contract with. They use a rule they publish and a number they will not show you. It is the reason the other three matter more than they look. Every one of them is an input into a reputation you cannot inspect. The effects arrive weeks later on a different send.
The fourth card is the whole of this page. The first three are decisions you make. The fourth is a decision made about you, by an organisation that will not tell you the result, and it is made on the basis of the first three.
Tony's heating and cooling company (invented). Tony's audience rule was written the week the sending tool was set up, four years ago: everyone who has ever booked a job. It has no exit. The customer who moved out of the area, the landlord who sold the building and the man who asked Beth to stop sending him the notes are all still in it.
Your turn. Open your largest audience and write down, in one sentence, the rule that puts a person into it and the rule that takes them out.
Do you need permission to email customers in the United States?
A business does not need prior permission to send a commercial email under the federal CAN-SPAM Act, which sets conditions on the message rather than on the relationship; the permission that matters in practice is the mailbox providers' demand, set out further down this page.
It is easy to assume American email marketing law works like a consent regime, in which you may not email someone unless they agreed. The federal statute does not work that way. The CAN-SPAM Act at 15 U.S.C. 7704 sets conditions on the message rather than on the relationship. There is no clause in it requiring prior permission to send a commercial message.
What it does require is worth knowing exactly, because a sending tool does not do all of it for you on its own.
The message must contain "a functioning return electronic mail address or other Internet-based mechanism, clearly and conspicuously displayed". A recipient can use it to ask not to receive future messages. That mechanism must remain "capable of receiving such messages or communications for no less than 30 days after the transmission of the original message". So an unsubscribe link that dies when a campaign is archived is not a technicality.
Once someone uses it, section 7704(a)(4) gives you a deadline. It becomes unlawful for the sender to send them a further commercial message in the scope of that request "more than 10 business days after the receipt of such request". Ten business days is roughly a fortnight. That sounds generous, and it is exactly why senders trip over it: a fortnight is long enough for a sequence someone is already inside to fire once more.
And section 7704(a)(5) requires three things in the message itself. One is "clear and conspicuous identification that the message is an advertisement or solicitation". One is clear notice of the opportunity to decline further messages. One is "a valid physical postal address of the sender", in the footer, where mail reaches you. The first of the three does not apply when the recipient has given prior affirmative consent to receive the message; the other two always do.
There is a rule underneath the statute that is more practical than the statute. It is the one to check your own tools against. Under 16 CFR 316.5, neither a sender nor anyone acting for them "may require that any recipient pay any fee, provide any information other than the recipient's electronic mail address and opt-out preferences, or take any other steps except sending a reply electronic mail message or visiting a single Internet Web page" in order to opt out or to have that opt out honoured.
Read that against the last preference centre you were sent to. A single web page. No login. No required survey about why you are leaving. No account.
Tony's heating and cooling company (invented). Beth checks the October note against the statute. The unsubscribe link works. The footer carries the office phone number and no postal address, and the address goes in that afternoon.
Your turn. Open your last marketing email and find the postal address and the unsubscribe link without scrolling past them.
Is a newsletter an advertisement under CAN-SPAM?
Whether a newsletter counts as an advertisement under CAN-SPAM is decided by how a reasonable recipient would read its subject line and its body, not by what you intended. The duties above attach to a "commercial electronic mail message", and the natural instinct is to think a useful newsletter is not one: it contains advice, and it is not selling anything. But the FTC's rule at 16 CFR 316.3 sets out how the primary purpose of a mixed message is decided.
The rule has two paragraphs for a mixed message. Where promotion sits beside transactional or relationship content, such as a receipt or a booking confirmation, the message is deemed commercial if a reader would take the subject line as promotion, or if the transactional content does not appear "in whole or in substantial part, at the beginning of the body of the message". A newsletter of advice falls under the other paragraph, for promotion beside content that is not transactional. There, too, the first test is the subject line: whether "a recipient reasonably interpreting the subject line of the electronic mail message would likely conclude that the message contains the commercial advertisement or promotion of a commercial product or service". The second is whether a recipient "reasonably interpreting the body of the message would likely conclude that the primary purpose" of the message is promotion, judged by factors such as where the promotion sits, how much of the message it takes and how type and colour highlight it.
The subject line, read by a stranger, is the first test, and the body is read the same way.
So a useful newsletter with a subject line about furnace filters is one thing. The same newsletter with a subject line about booking a tune-up is another. The difference is decided by someone who has never met you. What follows from that is simple, and it is not a burden. Put the postal address and the unsubscribe in everything, and stop trying to work out which of your emails count.
Tony's heating and cooling company (invented). The October subject line was "Book your fall tune-up before the first cold night". A stranger would read that as a promotion, so the note was an advertisement, whatever the filter advice inside it was.
Your turn. Read your last three subject lines as a stranger would, and mark each one an advertisement or not.
In a real estate office, specifically
A real estate team's monthly market note is the same case. Take an invented one: a note with the median price in three towns, what closed against what was asked, and two sentences about an inspection that fell through is useful, and real estate marketing automation sends one to every past client. With a subject line about the market, it may be read as information; with a subject line about booking a home valuation, a stranger would read it as a promotion. Put the postal address and the unsubscribe in every note, and the question stops mattering.

Who decides whether an email goes to spam?
The mailbox provider decides whether an email goes to spam: the company that runs the recipient's inbox, such as Google for a Gmail address or Yahoo for a Yahoo one, judges each message after your sending tool has handed it over. Two of them, Google and Yahoo, publish the rules they apply to incoming mail, in plain English on a public page.
Unless you are their customer, you have no agreement with them. What governs the relationship from your side is a document on a help site: the rules they publish, which the next sections read.
You can count what share of your own list sits behind those two in an afternoon. It is worth counting before you decide how much of this applies to you.
The system
From something happening to somebody reading it.
Six hops, and your software stops at the third. The last three are the ones that decide whether any of the first three mattered. They are done by companies you have no agreement with. The only one of them you can influence is the fourth. The first two hops are the ones a sending tool shows you.
Scroll to follow the chain
- The trigger: Something happened
- The audience: Who still matches
- The send: Your software's last act
- The signature: Who says this is yours
- The judgement: Someone else's rule
- The inbox: Or the other folder
Tony's heating and cooling company (invented). Beth sorts the fourteen hundred addresses by what comes after the @ sign. In this invented example about eight hundred end in a Gmail address and about two hundred in a Yahoo one, so roughly seven of every ten notes are judged by one of the two published rule books.
Your turn. Sort your own list by the part after the @ sign and count the two largest groups.
Do Google's sender rules apply to a small business?
Google's sender rules do apply to a small business, because the requirements that matter most for a small sender sit in the list headed "Requirements for all senders", not in the list for senders of "5,000 or more messages per day". Google's page for email sender guidelines splits into those two lists, and it is worth reading which items are in which.
If you have heard the whole thing called the bulk sender requirements, the natural conclusion is that a small business is exempt from all of it. Look at which list the important items are in. Under "Requirements for all senders": set up SPF or DKIM authentication for your sending domains, ensure valid forward and reverse DNS records, use a TLS connection, format messages according to RFC 5322, and "keep spam rates reported in Postmaster Tools below 0.3%".
The 0.3 percent is in the all senders list. A one person office sending twice a month is inside it.
The larger list adds the things that are about volume. SPF and DKIM together rather than either. A published DMARC policy. Alignment between the domain in your From header and the authenticated domain. And one-click unsubscribe. Yahoo's sender best practices states the same requirements in its own words. It spells out two details Google's page does not: "honor unsubscribes within 2 days", and an unsubscribe process that is obvious, visible and "doesn't require users to log in".
The evidence
How long you have to stop, after somebody asks you to
Two deadlines for the same act, each taken from the document that sets it. 15 U.S.C. 7704(a)(4), linked in the section above, makes a further commercial message unlawful more than ten business days after the request. Yahoo's published sender requirements, linked below, say to honor unsubscribes within two days. Source: Yahoo Sender Best Practices, honor unsubscribes within 2 days.
The two bars are not in the same unit, and the difference runs in the safe direction. Ten business days covers a longer stretch of calendar than ten days does. So the drawn gap is smaller than the real one rather than larger. There is no third bar for Google, and that is a checked absence rather than an omission. Their sender guidelines page carries no figure in days anywhere on it. Read this as which document is actually setting your deadline, not as a scale of penalties. The statute is enforceable and the requirement is not. The requirement is still the one that will change what happens to your mail.
On the unsubscribe, the mailbox provider's rule is stricter than the regulator's: two days against ten business days. And one more line from Google's own page is worth sitting with, because it is the mailbox provider itself speaking: "Consider unsubscribing recipients who don't open or read your messages."
Tony's heating and cooling company (invented). Tony had assumed the rules were for big senders. Fourteen hundred notes a month is far below five thousand a day, and every item in the all senders list still applies to him.
Your turn. Read the all senders list on Google's page, and put a mark beside each item you can confirm for your own domain today.
What do SPF, DKIM and DMARC do for email deliverability?
SPF, DKIM and DMARC prove who sent an email, and none of them makes a mailbox provider treat the email better: they remove a reason to reject it. They are sold as a deliverability package, and setting them up is worth doing. What is worth understanding is what each one claims, and their own specifications are clear about it.
SPF is RFC 7208, and its abstract says what it is for. Existing protocols "place no restriction on what a sending host can use as the MAIL FROM of a message". SPF is the mechanism by which a domain owner "can explicitly authorize the hosts that are allowed to use their domain names, and a receiving host can check such authorization". It authorises machines. It says nothing about the message.
DKIM is RFC 6376, and it "permits a person, role, or organization that owns the signing domain to claim some responsibility for a message by associating the domain with the message". Claim responsibility. Not vouch for, not certify, not recommend. It is a signature. A signature on a bad letter is still a signature.
DMARC is RFC 7489. It ties the two to the domain in your From line: a message passes when SPF or DKIM passes for a domain aligned with that one, and the owner's published policy tells a receiver what to do when it does not. Its abstract contains one sentence that belongs beside every offer to fix your deliverability.
DMARC does not produce or encourage elevated delivery privilege of authenticated email.
That is the specification saying, in its own summary, that doing all of this correctly buys you nothing in the way of better treatment. What it buys is the absence of a penalty. It also lets a receiver tell the difference between you and someone forging your domain.
Which means setting them up is worth doing at any size and none of them is the work. They establish who is speaking. Whether what you said was wanted is a separate judgement, made afterwards, on evidence you supply every time you send.
Tony's heating and cooling company (invented). Beth looks up the domain's records. The notes pass DKIM, but they are signed by the sending tool's own domain rather than Tony's, and Tony's domain publishes no DMARC policy. At fourteen hundred notes a month that meets both providers' lists for every sender, which ask for SPF or DKIM; alignment and DMARC sit in the larger lists. Beth sets them up anyway, so that the domain in the From line is the one that vouches for the note.
Your turn. Ask whoever set up your sending which domain signs your messages, and whether it is the one in your From line.
What is one-click unsubscribe, and why does it protect the sender?
One-click unsubscribe is an exit a recipient can use without a confirmation page, and it protects the sender because a person who cannot leave easily can still press the button that reports the message as spam. It sounds like a favour to the recipient. It is the mechanism that protects your reputation from the people who want to leave.
It has its own specification, RFC 8058. The reasoning is written down in the introduction rather than left to be worked out. The problem it solves is technical, as its introduction explains: anti-spam software often fetches every link in a message's headers on its own, so a plain unsubscribe link in a header could be set off by a machine rather than a person. Senders answered by putting a confirmation page behind it, and "that makes the unsubscription process more complex than a single click".
Then the specification explains why that mattered, and this is the passage to take away:
Operators of broadcast marketing lists tend to be primarily concerned about deliverability of their mail... Hence, the mailers want to make it as easy as possible for recipients to unsubscribe; if an unsubscription process is too difficult, the recipient's alternative is to report mail from the sender as junk until the mail no longer appears in the recipient's inbox.
That is an internet standards document stating the business case for an easy exit, on the sender's behalf. Every extra step between someone deciding to leave and being gone gives them another option. That option is the button that counts against the three tenths of a percent.
There is one detail with a practical result. Section 4 of the same document requires that the message carry a valid DKIM signature covering the unsubscribe headers. Without it, it says, "the mail receiver SHOULD NOT offer a one-click unsubscribe for that message". So the authentication work in the section above is not a separate project from the unsubscribe work. Get the signature wrong and the safest exit route is quietly not offered to your recipients at all.
Tony's heating and cooling company (invented). Beth sends herself a note and tries to leave through the link in the footer. It opens a page that requires her to sign in to manage her preferences. Under 16 CFR 316.5 that is a step the rule does not allow, and under RFC 8058's own reasoning it is a step that turns a polite leaver into a complaint. The one-click exit in the message's headers is a separate mechanism, so she asks the sending tool's support whether the notes carry it.
Your turn. Send yourself your last campaign and try to unsubscribe. Count the pages.
What spam rate sends emails to spam?
The spam rate Google and Yahoo both publish as the ceiling is 0.3 percent: a sender is asked to stay below three complaints for every thousand messages, and cannot work out its own rate against it, because each provider counts only its own complaints against the mail it delivered to an inbox. Yahoo's sender best practices state it in as many words: "Keep your spam rate below 0.3%". Google puts it in the all senders list.
On a thousand delivered messages, three complaints is the line itself, not an allowance. Not three unsubscribes, which are a healthy sign. Three people pressing the button that says this should not be here.
Tony's heating and cooling company (invented). Beth cannot work out either provider's rate from her five complaints: each provider counts its own, against the mail it delivered. What she can say is that five on fourteen hundred notes is already more than three in a thousand, before a single note that never reached an inbox is taken out of the count.
In your numbers
How many people have to press the button before you are over the line?
The list as it actually is, not the segment you intend to build.
Gmail and Yahoo both publish the same figure. Count the domains in your own list rather than guessing at this one. It is an afternoon's work, and it decides how much of this section applies to you.
Count everything, including the ones a sequence fires without anybody deciding.
Complaints in one send that reach the published ceiling
3.4complaints in one send
- People on the listyour 1,400
- 1,400people
- Whose provider publishes a ceilingyour 80%
- 1,120covered messages
- At the published ceiling0.3%, the figure Google and Yahoo both publish
- 3.4complaints in one send
- Across a month of sendingyour 2
- 6.7complaints a month
- Across a year12 months
- 81complaints a year
The headline is the third row rather than the yearly figure. It is there because a single send is the unit a person actually decides about, and because the number is small enough to be startling. Shares of people produce fractions, and half a complaint is not a thing. Read two and a half as between two and three. Three things this deliberately refuses. It will not work out your actual spam rate, and it cannot. Yahoo states the rate is calculated in their system on mail that was delivered to the inbox, which is a denominator no sender can see. So what you have here is a ceiling in your own units rather than a measurement of you. There is no open rate and no click rate anywhere in this chain. The section above shows what an open actually counts, and it is not attention. And there is no row for what a damaged sending reputation costs in money. We found no published figure for that with a method under it, and this article is not going to be the first to make one up.
Now the honest part, and it is the reason no calculator can give you your own rate, only the ceiling in your own numbers.
Yahoo states how the figure is derived: "Spam rate is calculated in our system based on mail delivered to the inbox". The same line tells senders to keep that in mind when referring to it. Read it carefully. The denominator, which is the bottom of the fraction, is not what you sent. It is what they delivered to an inbox. Mail that was rejected or filtered before arriving is not in the bottom of that fraction.
That has an uncomfortable result. As your reputation falls, more of your mail is filtered. That shrinks the denominator. A smaller denominator raises the rate worked out from the complaints you still receive. The measurement moves against you at exactly the moment you would want it to hold still, and there is no version of the sums you can do at your own end that reproduces it.
Google publishes a sender's own rate to that sender through Postmaster Tools, which is free and needs one record added to your sending domain's DNS. It is the most useful single step on this page in practice, and the one place we know of where a small sender can see Google's own reading of its spam rate.
Your turn. Find out whether anyone in your business can add a record to your sending domain's DNS, and write down their name.
What does an email open rate measure?
An email open rate measures how often a mail client fetched a small tracking image, not how often a person read the message. Nobody tells your software that a message was read. The message contains a small image hosted on the sender's server, the kind of tracking the paper below studies. If the recipient's mail client fetches that image, the fetch is recorded and reported to you as an open.
That mechanism has two failure modes in opposite directions. Clients that block remote images never register an open for a message someone read carefully. Clients that fetch images on their own register one for a message someone deleted from the preview pane.
The evidence
What a corpus of 12,618 commercial emails was carrying besides the message
Three shares of the same corpus of 12,618 commercial emails from 902 senders. A crawler collected them by subscribing to mailing lists on 15,700 websites. They were then opened in a real email client while the network traffic was recorded. Source: Steven Englehardt, Jeffrey Han and Arvind Narayanan, I never signed up for this! Privacy implications of email tracking, Proceedings on Privacy Enhancing Technologies 2018.
The authors state their own limit, and it is the one that matters here. They write: "our corpus of emails is not intended to be representative, and we are unable to draw conclusions about the extent of tracking in the typical user's mailbox." So read this as what commercial mailing lists were doing rather than as what your own provider does. The study is also from 2018. The third bar in particular describes a practice that mailbox providers have since made harder rather than one that has grown. What has not changed is the mechanism, and the mechanism is the reason it is here. None of this is added by the sender. All of it arrives with the tooling. And the researchers' own list of open questions includes whether a sender setting up a campaign is even told that the tracking is there.
Read the chart with the note under it, because the authors are careful about what they are and are not claiming. What survives the caveats is the mechanism rather than the rate. A commercial email is a page, and that page loads resources from other companies. How many of those a sender put there deliberately is exactly what the researchers could not resolve. They listed it as an open question at the end of their own paper: when a sender sets up a campaign with a mailing list manager, is the tracking disclosed to the sender at all.
Be careful about what that does and does not license you to say. The same paper reports that most of the address leaks it found, 62 percent by its own heuristics, were intentional on the sender's part. So this is not a story in which senders are uniformly innocent. What it is is a strong argument for finding out which case you are in, and that takes minutes rather than a project. Send yourself a campaign. Open it in a browser with the developer tools showing. Read the list of domains the message contacts. Each is a company your recipients' mail programs will contact when they open your note, with your name on it.
Tony's heating and cooling company (invented). Tony's open rate had fallen from one month to the next, and he had read it as a verdict on his writing. It was also what a filtered message looks like from the inside: a note that never reached an inbox cannot be opened.
Your turn. Open your last campaign with the developer tools showing and write down the domains it contacts.

Why can later email campaigns land in spam when the first did not?
Later email campaigns can land in spam when the first did not because the mailbox provider judges each send partly on how people reacted to the ones before it, so every send adds to a record the next one is graded against. Yahoo's page says it plainly: "Sending email to users who are not reading them, or who report them as spam, will harm your delivery metrics and reputation." It is not fatigue, and it is not the copy going stale.
So every send to a list is graded partly against a record you are writing as you go. Three results follow, and none of them are visible from inside a dashboard.
The first is that mistakes are paid for later and somewhere else. A badly aimed send in March does not cost you March. It costs you part of the delivery of the April message, to the people who were pleased to hear from you. Nothing in your reporting will connect the two events.
The second is that adding people to a list is not free even when they never respond. Every address that did not want the message is a source of the evidence that grades you. Google's own page tells senders to consider unsubscribing recipients who don't open or read their messages, and Yahoo's says the same of users who are not reading.
The third is the one that changes behaviour, and it is why the close of this page is what it is. The cheapest lever available to you is not better writing, a better subject line or a better tool. It is a shorter list. Taking off the people who did not want the message changes the one input you control, and the change carries into every later send, not only this one.
Tony's heating and cooling company (invented). This is where Beth's two first ideas fail. A livelier subject line changes what a stranger reads; it does not change who receives the note. A new sending tool starts on a domain whose record is already written. The list is the input that moves.
Your turn. Count the people in your largest audience who have not booked, replied or bought in three years.
How do you check why your emails go to spam, in an afternoon?
To check why your emails go to spam, run seven checks in an afternoon, in this order: the domain's records, the From line, Postmaster Tools, the unsubscribe, the tracking, the audience and the count of sends. None of the checks needs a consultant to run, though fixing what they find may need whoever manages your domain.
- Find out what your domain publishes about itself. Your SPF, DKIM and DMARC records are public DNS entries. Anyone can read them, including you. If nobody in your business knows whether they exist, that is the finding. It is the one to fix first, because everything else depends on it.
- Check the From line against them. Google's list for senders of 5,000 or more messages a day and Yahoo's requirements for bulk senders ask for the domain in your From header to be aligned with the domain that authenticated the message. If your mail says it is from you but is signed by your mailing tool's own domain, it would fail that requirement at that size, and it is easier to fix before you get there.
- Set up Google Postmaster Tools. Google's own sender guidelines point every sender to it: the spam rate they ask you to keep below 0.3 percent is the rate "reported in Postmaster Tools". It is free, and it needs a DNS record. It is where Google shows a sender its own spam rate, as Google works it out. If it shows no data yet, that is not a clean bill of health.
- Send yourself a campaign and try to unsubscribe from it. Count the pages. If it requires you to log in, requires a reason, or takes more than one page, it does not meet the rule at 16 CFR 316.5. It is also turning people who would have left politely into people who press the other button. This tests the link in the footer; ask whoever set up your sending whether the one-click exit is in the headers too.
- Open the same message with developer tools on. Look at the list of domains it contacts. Decide whether you are comfortable with it. There is no right answer here, and there is certainly a wrong one, which is not knowing.
- Open your largest audience and scroll to the bottom of it. Read the last twenty names. If you recognise someone who moved away, someone who told you to stop, or someone who is now a competitor, your audience rule has no exit condition.
- Count your sends over the last ninety days. Not campaigns. Messages that left the building, including every automated one. Count it before you guess it, because it is the input you control most directly.
Tony's heating and cooling company (invented). This is Beth's whole afternoon, an invented example, with what each check found:
- The domain's records. DKIM passing under the sending tool's domain; no DMARC policy. Not required at Tony's size; set up anyway.
- The From line. Not aligned with the signing domain. Not required at Tony's size; fixed with the first.
- Postmaster Tools. Not set up; nobody had heard of it. Set up the same day.
- The unsubscribe. A sign-in page behind the footer link. A finding.
- The tracking. The note contacted four outside domains, two of which Beth could name. Noted, and asked about.
- The audience. The last twenty names included a man who had asked Beth to stop sending him the notes. A finding.
- The count of sends. Three notes a month, not one: the October newsletter, a tune-up reminder sequence and a review request nobody remembered switching on.
Your turn. Run the seven checks on your own sending this week, and write one line beside each.
What happened when Tony's list got shorter?
Tony's sending changed on what the checks found, and the largest change was the list: fourteen hundred people became about nine hundred, the customers who had booked a job in the last three years or asked to stay on. In this invented example the work took two afternoons of Beth's time and one call with whoever managed the domain.
Tony's heating and cooling company (invented). DMARC was published and the notes were signed with Tony's own domain, so the From line and the signature now agree. The unsubscribe became one page with one button. The man who asked to stop getting the notes was removed, with everyone who had not booked in three years, and the forgotten review sequence was switched off. The December note went to about nine hundred people with a subject line about furnace filters, and the sending tool's report showed one complaint against October's five. Several things changed at once, so that count cannot say which change mattered, and nobody can promise what the next note will do. Tony did not cut the list on a promise. He cut it because every name on it is now someone who might want the note, and in a year he will have twelve months of complaint counts that someone actually read.
Your turn. Decide now which number you will read every month: complaints per send, from your sending tool's report and, once it shows data, from Postmaster Tools.
If you would rather someone read your sending records with you, that is where we start: the next sections say what the work takes and what it will not do.
The honest read
Send us the domain you send marketing email from, and roughly how many people are on the list. We will tell you what your authentication records currently say. We will tell you whether the address in your From line is aligned with them. And we will tell you which of the two published requirements you are already meeting without knowing it.
It is a short reply from a person, and it costs nothing. Everything we would read is published in public DNS. We do not need access to your mailing tool or your list.
What does email marketing automation take, and how long does it take?
Email marketing automation takes three pieces of work: the sending platform, the set-up of audiences and triggers, and getting your sending domain into order, and the third is where the time goes when it goes badly.
The sending platform is a subscription from whoever you use. Then the set-up: audiences, triggers, delays, suppression rules, and the joins to wherever your contact data lives. Then the item that is easy to leave out: getting your sending domain into a state where two things are true. The three records above are correct, and the address in your From line is aligned with them.
If your domain's DNS is somewhere you control and nobody has set up anything unusual, it is an afternoon. If your mail has gone out from three different tools under two different subdomains set up by people who have left, it becomes a long dig. The honest order is to finish it before writing a single campaign rather than after.
The other real need is a person. Audiences rot, and they rot quietly. So someone has to own the rules and look at them on a schedule. That is not a big job, and it is nobody's job by default. That is how a segment defined once outlives everyone's memory of what it was for.
Before spending anything, run the seven checks above. They settle the only question that decides the shape of the project: is your domain already in good order, or has it been quietly piling up settings for six years?
What can marketing automation not do?
Marketing automation cannot create demand, cannot get your mail delivered, and cannot tell you whether anyone read anything.
It does not create demand. Every mechanism described here assumes someone who already has a reason to hear from you. A cadence does not turn a person who is not looking into a person who is. Pointing one at people who never asked is a different activity with a different set of risks.
It does not get your mail delivered. Authentication removes a reason to reject you. Delivery is a judgement made by someone else, on the basis of how the people you sent to reacted last time. The only lever you have on it is who you send to and how often.
It does not tell you whether anyone read anything. Opens are a proxy with known failure modes in both directions. Clicks are a proxy for interest in one link. Neither is a measurement of attention. A dashboard built out of them can look healthy while the mail is being filtered.
It does not fix a message that should not be sent. Behaviour triggers change when something arrives, not whether it should. A poorly judged email delivered at the perfect moment is a poorly judged email that arrives faster.
It does not survive being aimed at everyone. A bigger list looks like a bigger asset, which is why this limit is the hardest to accept. Every send to someone who did not want it is evidence handed to a filtering system. It is scored against your domain. The damage lands on the next message you send to the people who did want it.
Three ways a good build produces nothing
None of them are the software.
A segment nobody ever leaves
Someone defines the past customers audience on the day it is built. Nobody ever writes the rule that removes a person from it. Two years later it contains everyone who has ever filled in anything, including the family who moved out of the area last year. Every campaign is then aimed at a group that has quietly stopped meaning anything. The reporting will look fine, because the reporting counts sends.
A sender address nobody owns
The mail goes out from a subdomain a contractor set up. It is authenticated by records nobody in the business can read. The reputation attached to it belongs to whoever last configured it. The day it goes wrong there is nobody to ask. This is dull. It takes an afternoon to fix while it is small, and it cannot be fixed in a hurry.
Reporting that measures the sending
Sends, opens, clicks. All three are counts of what your own software did, and the first is entirely under your control. A dashboard built out of them will show a healthy campaign right up to the point where the mail stops arriving. That is because nothing in it is a measurement of the only thing that matters. That is whether a person read something and thought better of you for it.
Common questions about email deliverability and marketing automation
Why are my emails going to spam?
Your emails are going to spam when the mailbox provider cannot confirm your domain sent them or when too many recipients report them: check your SPF, DKIM and DMARC records and your From line first, then your list.
What is marketing automation for a small business, in plain terms?
Marketing automation for a small business is a standing instruction about people you are not currently talking to: a rule, written once, about who hears from you, what they hear and when. The software part is a scheduler and a mail sender; the part that decides whether it works is the audience rule, because it keeps running for years against people whose circumstances keep changing.
Do I need permission to email somebody in the United States?
You do not need permission to email somebody under the federal CAN-SPAM Act, which sets rules for the message: it must identify itself as an advertisement unless the recipient gave prior affirmative consent, carry a valid physical postal address and offer a working exit that lasts at least thirty days, with the opt out honoured within ten business days. Permission matters in practice for a different reason: mail to people who did not ask is what generates the complaints that get you filtered.
How is this different from workflow automation?
Marketing automation differs from workflow automation in what fails: workflow automation is about work, a step someone used to do by hand that now happens on its own, and you find out on the day if it breaks; marketing automation is about judgement at scale, who deserves a message, what it says, when it lands. That is why the workflow article is a separate read rather than a section of this one.
Is this the same as reactivating an old database?
Reactivating an old database is not the same thing, and the difference is the consent question: going back to people who contacted you years ago raises questions about permission that has gone stale, and those rules have dates in them. That article covers them; this one is about the people already on your list and what you send them.
Will automated marketing annoy my customers?
Automated marketing will annoy some customers, and the number that matters is the ones who press the spam button, not the ones who unsubscribe: both Google and Yahoo publish the same ceiling of three tenths of one percent, fewer than three complaints in a thousand delivered messages. Sending people things they do not care about is what makes them press it.
Why did my open rate drop?
An open rate can drop because fewer people opened the email, or for two reasons that have nothing to do with your writing: mail clients that block or fetch remote images bend the count in both directions, and mail that is filtered never reaches anyone who could open it. Your spam rate in Postmaster Tools is a better health signal, because it is measured by the party whose opinion decides whether your mail arrives.
How often should I send?
How often to send has no published number worth quoting, and we found none with a method behind it. What can be defended is the direction: every send is evidence handed to a filtering system, so send to fewer people more carefully, and the frequency question becomes a smaller one.
What is the one thing to fix first?
The one thing to fix first is what your sending domain publishes about itself: if your mail passes neither SPF nor DKIM, you are failing a requirement both providers publish for every sender, and every improvement above that layer is applied to mail that may not arrive. An aligned From line and a DMARC policy come next: required in the larger senders' lists, and worth setting up at any size.
Can we talk about email deliverability on a call if my business is outside New York?
We are based in New York, and the call works for an owner anywhere.
What to do first about emails going to spam
The first thing to do about emails going to spam is to read what your domain publishes about itself, and the second is to send less, to fewer people, more carefully.
That runs against how this category is sold. A platform priced per contact has no reason to suggest a smaller list. A dashboard built on sends and opens will always reward more of both. But the mechanism is not in dispute: it is published by the companies that run it. Whether you can reach the people who want to hear from you depends partly on how the people who did not want to hear from you reacted, and who is on the list is the input you control most directly.
Open the last campaign you sent. Read it as the person third from the bottom of the list, the customer whose job you finished two years ago and who has not heard from you since except in a segment. If it does not survive that, the fix is not a better subject line. It is a smaller list. A smaller list is the one change in this entire subject that costs nothing.
There is no price on this page, because the cost tracks four things nobody can guess from an article. How many channels you want joined. Whether your sending domain and its authentication already exist or have to be set up. How much of your contact data can be trusted enough to segment on. And whether anybody in the business will own the audience rules afterwards. The AI assessment is an hour, done with you, and it starts by reading what your domain currently publishes about itself.
Start with the page on marketing automation for small business: what we set up, what you approve, and what it will not do.
Sources
- 15 U.S.C. 7704, the CAN-SPAM Act's requirements for commercial electronic mail, law.cornell.edu. Statute in force, read November 11, 2026.
- 16 CFR 316.3 (primary purpose) and 16 CFR 316.5 (no fee or extra steps to opt out), the FTC's CAN-SPAM rule, law.cornell.edu. Regulation in force, read August 25, 2026.
- Google, Email sender guidelines, support.google.com: the requirements for all senders and for senders of 5,000 or more messages a day, and Postmaster Tools, which the page itself names as where a sender's spam rate is reported. A live help page; read November 11, 2026 and again November 17, 2026.
- Yahoo, Sender best practices, senders.yahooinc.com: the 0.3 percent ceiling, how the rate is calculated, the two-day unsubscribe. A live page; read November 11, 2026.
- RFC 7208 (SPF, 2014), RFC 6376 (DKIM, 2011), RFC 7489 (DMARC, 2015) and RFC 8058 (one-click unsubscribe, 2017), rfc-editor.org. Older sources, kept for the argument: they define the mechanisms named here.
- "I never signed up for this! Privacy implications of email tracking", by Steven Englehardt, Jeffrey Han and Arvind Narayanan, in the Proceedings on Privacy Enhancing Technologies, 2018, petsymposium.org. An older source, kept for the argument: cited for the mechanism, not for today's rates.
Written with an AI assistant and checked by a person.
Keep reading
CRM data cleansing: stale contact records, data enrichment and the field it overwrites 
August 25, 2026
CRM data cleansing: stale contact records, data enrichment and the field it overwrites
Duplicate contacts in a CRM: why a small business gets them and how to fix them 
August 25, 2026
Duplicate contacts in a CRM: why a small business gets them and how to fix them
Lead follow-up for old leads: how to win back past customers 
July 30, 2026
Lead follow-up for old leads: how to win back past customers
