Email autoresponder: what it is and which shape to choose

What an autoresponder actually is

An autoresponder is mail sent by a rule. Something happens — a form is submitted, a file is asked for, a link is clicked, a length of time passes — and software sends a message that was written before the event, without anybody being present. That is the whole definition. It says nothing about how many messages there are, how far apart they arrive, or whether there is a last one.

The name comes from the first version of the idea, which really was a single automatic response: mail arrives, mail goes back. Everything since has kept the name and dropped the constraint, so the same word now covers a receipt, a five-day course and an open-ended marketing sequence. Newer tools tend to call the same feature a sequence, an automation, a workflow or a drip campaign, and none of those words means anything different. They describe a stored message, a rule about when it goes out, a queue and a suppression list.

Which is why the useful question is never whether a tool has an autoresponder. Every tool that sends mail at all has one. The question is what the rule is allowed to be, and that is where tools differ enormously from each other while their feature lists do not.

The three shapes, and why the middle one is different

Almost everything sold as an autoresponder is one of three shapes. A single automatic reply: one event, one message, done. A finite sequence: a fixed number of messages, in order, ending on a day the reader was told about before they joined. An open-ended sequence: messages keyed to behaviour, running until the reader converts, leaves, or is moved somewhere else.

The first is a utility and needs no strategy. The difference between the second and the third is the only part of this worth an afternoon, and it is not a difference in software. It is a difference in what was promised. A finite sequence has an ending that the sender stated out loud, which means it can be measured by how many people got to it. An open-ended sequence has no ending to count arrivals at, so it is measured by conversion instead — which is the right measure for it, and a much easier one to flatter.

That distinction has a practical edge. A finite sequence has to be written in full before the first reader arrives, because the reader was told how long it is. An open-ended one can be extended for ever, and usually is, which is how a sequence nobody has read end to end in two years ends up sending somebody a message that contradicts one they got on Tuesday.

Which shape your idea already is

The test that settles it fastest is to write the last message rather than the first. If it comes out and reads as a conclusion — here is what you can now do, here is what to do next — then the thing is finite and you should say so in the sign-up copy, because a stated ending is the cheapest thing you will ever do to a conversion rate. If the last message will not come, either the subject has no end, in which case what you have is a newsletter, or the ending depends on something the reader does, in which case it is open-ended and will need to read state.

The second test is the trigger, and specifically whether it can happen twice. A finite course has exactly one trigger and it fires once per person: somebody confirmed. Nothing else about the reader matters, and the calendar does the rest. An open-ended sequence is keyed to something that may happen, may repeat, or may never happen at all, and a rule keyed to a maybe has to be able to check what is true now.

Rough signals, when the answer is not obvious:

• You can write down what the reader will be able to do when it is over → finite, five to ten days, and say the number on the page.
• The next message depends on what the reader did inside a product → open-ended, and branching is doing real work rather than being sold.
• The trigger is a date the reader owns rather than the day they joined — a renewal, a booking, an expiry → open-ended, because the schedule is per person and unknowable when you write it.
• You want something that goes on working after you have stopped thinking about it → finite. It is written once and delivered from then on.
• You are about to add a rule about people who did not open message two → stop. Usually the rule would send the same email again, which is resending rather than automation.

What the word hides, and what it costs

Because every tool has an autoresponder, the feature name separates nothing, and the differences that matter are one level down in the scheduler. The first is what a delay counts from. Most builders schedule in relative time — wait one day, wait three days — measured from when the previous message actually left rather than against a local morning. Three delays of twenty-four hours after a confirmation at twenty to midnight put the rest of the week near midnight, and a queue that ran late once drags every later day behind it for ever.

The second is what the trigger does when it fires twice. Open-ended triggers are built to repeat, because behaviour repeats, so a second form submission from an address already on the list can start a second copy of the sequence beside the first. The reader gets two of message three, a few days apart, and nothing reports that it happened. That behaviour is usually one checkbox and it is usually on.

The third is what an edit does to the people already inside. Some tools push a correction to everybody, including the reader who received the old version this morning; some hold each person to the version they entered on, so the fix reaches nobody already running. Neither is wrong and both are surprising, and which one you have decides whether fixing a broken link is a two-minute job or a decision about who gets which version. The companion guide on courses against drip campaigns works through the rest of these defaults, and the one on choosing software turns them into questions you can ask before paying.

An autoresponder is the part of your sending nobody is watching

The thing that makes automated mail valuable is also its whole risk: it goes out when you are not there. A broadcast that looks wrong gets noticed within an hour, because you sent it and you are reading the replies. An autoresponder that started going wrong in March is discovered in July, by accident, and every reader in between got the broken version.

Two consequences follow. The first is that the consent step has to be the strongest part of the whole thing. An address added by a typo or by somebody else receives one unwanted message in a single-reply setup and a week of them from a sequence, and unwanted mail is what produces complaints. Google asks bulk senders to keep the complaint rate in Postmaster Tools below 0.10 per cent and never to reach 0.30 per cent, and a rate is unforgiving at small volumes: at a thousand delivered messages, three complaints is already the outer limit. Requiring a confirmation click before anything else is sent is the only step that removes the typed-wrong, the borrowed and the automated before any of them count against you.

The second is that an autoresponder dates. A sequence written against a price, a platform or a screenshot becomes quietly wrong, and because nobody is watching it go out, nobody notices. Put a review date in the calendar on the day you publish it, and read the whole thing on a phone at least once a year, following every link. Both of those are dull, and both are cheaper than the alternative, which is finding out from a reader.

What to check before the first stranger sees it

None of this needs a trial of a second tool. It needs one address you control and an afternoon, because every failure above is visible from the inside if you go looking and invisible if you do not. Do it in the order below, and do it again after any edit that touches the schedule.

The pass to make with your own address:

• Join on a Friday evening from a phone set to another timezone, and see what hour the next two messages arrive at.
• Submit the form a second time from the same address, and see whether a second copy of the sequence starts a few days later.
• Unsubscribe from one message, then send yourself whatever else you run, and see what still arrives.
• Edit a later message while you are sitting on an earlier one, and see which version you are sent.
• Read every message with images switched off, following every link, because the reader who finds a dead link is the one you never hear from.