Send three or four follow-ups, spread over two to three weeks, and make each one carry something the last one did not. That is the answer to the copywriting question, and it is the smaller half of the problem. A follow-up sequence sends to the same list four or five times, so it multiplies whatever is already wrong with that list by the number of steps. Before tuning cadence or rewriting a bump, find out how many of those addresses can actually receive mail.
Why does a follow-up sequence multiply a list problem?
A single send exposes a bad address once. A five-step sequence exposes it five times, and the damage is not linear, because mailbox providers weigh repeated failures to the same domain more heavily than isolated ones.
Two failure modes matter, and they behave very differently.
Confirmed dead addresses are the obvious one. In theory a five-step sequence turns each one into five bounce events. In practice most sequencers suppress a contact after a hard bounce, so the multiplication is usually capped. That protection only holds if bounce handling is genuinely wired up and tested, and if the failure comes back as a hard bounce at the gateway rather than later.
Catch-all domains are the less obvious one, and the bigger problem. A catch-all domain accepts mail for every possible address, so the receiving server says yes at the connection stage whether or not the mailbox exists. You will not learn which happened. That also means a catch-all failure often arrives late or not at all, so it may never trip the suppression rule that protects you from the first failure mode. Those rows keep receiving step two, step three, step four and step five. The mechanics of why a verifier cannot resolve these are in how to handle catch-all emails in cold outreach.
This is the part that gets misdiagnosed. When a sequence underperforms, the instinct is to rewrite the copy, because copy is the visible variable. If a meaningful share of the list cannot report back at all, no rewrite will produce a legible result. That is a deliverability problem wearing a copywriting problem's clothes. The bounce rate and deliverability definitions cover the arithmetic underneath.
What we see on our own lists
We are going to be precise about what this data is, because the honest version is less impressive than the version we could imply.
This is not customer data. ReplyLabs verified 3,389 email addresses in early August 2026 where the individual verdicts are still on file, and roughly three quarters of those came from a single list run on a single day by one of our co-founders. It is a small, concentrated sample of our own prospecting, not an industry benchmark, and you should discount it accordingly. We are publishing it because almost nobody publishes theirs, not because it settles anything.
On those 3,389 addresses:
- Deliverable: 2,118 (62.5%). The provider found no reason to reject the address. This is the segment where a non-reply most likely means a person chose not to reply.
- Catch-all: 1,061 (31.3%). The receiving domain accepts everything, so nothing can be resolved either way.
- Undeliverable: 185 (5.5%). The address was rejected.
- Risky: 15 (0.4%) and unknown: 10 (0.3%). Flagged, or the check could not complete.
Note that verification infers rather than proves. A deliverable verdict means no rejection was detected at the time of the check, not that a human reads that mailbox. Addresses also decay: people change jobs, and a verdict from three months ago is a weaker claim than one from this morning.
Take the two middle numbers seriously if your lists look anything like ours. Just under a third of a list returning catch-all means just under a third of your sequence is unmeasurable. Every rate you compute across the whole list inherits that.
How many follow-ups should you send?
Three or four after the initial email, so four or five messages in total, is a defensible default. The reasoning is arithmetic rather than taste.
- Each step multiplies send volume against the same contacts. A five-step sequence means your effective send volume is five times what your list size suggests, against a fixed daily inbox capacity. Our cold email deliverability checklist covers the per-inbox limits that constrains.
- Each step has to justify itself. If you cannot say what step four contains that steps one to three did not, you do not have a fourth step, you have a fourth send.
- Each step raises the chance of a complaint. A spam complaint is materially worse than a non-reply, and repeatedly messaging someone who was never interested is how you earn one.
If you want more contact than that, the better move is usually a smaller and better-targeted list contacted more thoroughly, rather than a larger list contacted more times. ICP scoring for outbound is the version of that which actually reduces total send volume rather than redistributing it.
What should a cold email follow-up actually say?
Not "just bumping this". Not "did you see my last email". Both announce that the message contains nothing, and both make the reader do the work of remembering who you are.
A follow-up should stand on its own. Assume the recipient never opened the first one, and never make message three depend on having read message one. Shapes that hold up:
- A new fact about them. Something specific from their site, their hiring, their product, their public activity. This is the strongest follow-up because it demonstrates you did not simply set a timer.
- A narrower ask. If the first email asked for a call, the follow-up can ask whether this is even the right person. That is far easier to answer, and it often gets forwarded internally.
- A different angle on the same problem. Same offer, different entry point. Useful when you are unsure which of two pains they have.
- A short proof point. A concrete outcome or mechanism, stated plainly. No attachments, no images, no tracking-heavy HTML.
- A clean exit. Make it easy and consequence-free to say no, or to say nothing and be left alone.
The new-fact version is the one people skip, because it needs per-row research rather than one template. That research is the thing worth automating, and it is what personalised cold email at scale is about.
Build a fallback in, though. Per-row research does not resolve for every row: pages are blocked, sites time out, companies have no useful public information. A follow-up that reads "I saw that {{company}} is hiring" with an empty merge field is worse than no follow-up at all. Every row that comes back empty needs a variant that is honest and generic rather than a template with a hole in it.
How long should you wait between follow-ups?
Widen the gaps as you go. Two to four days before the first follow-up, then roughly a week, then longer.
The logic is that early follow-ups catch the person who was simply busy, while later ones catch a change in circumstance, and circumstances take longer to change than calendars do. Compressing five messages into eight days does not create urgency, it creates a pattern that reads as automated, because it is.
The whole sequence sitting inside two to three weeks is a reasonable envelope.
How do you tell a non-reply from a non-delivery?
You mostly cannot, which is the point of verifying first.
A non-reply and a non-delivery look identical in your sequencer: no response. They call for opposite actions. A non-reply from a real person means the message did not land well, so change the message. A non-delivery means the message was never read, so nothing about the copy is being tested at all. Messages filtered to spam produce the same silence and are a third case again.
Verifying before the first send is what separates these. It converts an unknown number of the silences into a known category, so the remaining silence is interpretable. You cannot resolve catch-all addresses this way, which is exactly why knowing what share of your list is catch-all matters.
Should you clean the list again mid-sequence?
If the sequence runs for more than a few weeks, or the list was built more than a couple of months before you started, yes.
Addresses decay continuously as people change roles. A list verified in January and sent in April is not a verified list. Re-verifying before a later step in a long sequence is cheap relative to the reputation cost of discovering the decay through bounces. How to verify an email list covers doing this without exporting to another tool.
Common questions
How many follow-ups is too many? Beyond five total messages you are usually trading sending reputation for diminishing returns. The better question is whether each step carries new information; if it does not, the number is already too high.
Should I follow up in the same thread or start a new one? Same thread keeps context and is easier for the recipient to catch up on. A new thread with a fresh subject line is worth trying once, late in the sequence, on the assumption the earlier thread was never opened.
Does ReplyLabs send follow-ups? No. ReplyLabs does not send email at all. It is a Google Sheets add-on that verifies addresses, enriches rows from company pages, and runs AI prompts across a column. It prepares the list your sequencer sends to. Sending, scheduling and reply tracking happen in your own tool.
What is the single cheapest improvement to a follow-up sequence? Verifying the list before step one, because it is the only change that improves every subsequent step at once. Verification runs at $0.01 per address in ReplyLabs and only charges for addresses that return a result. The full workflow is in cold email from Google Sheets, and the verification step on its own is in email verification in Google Sheets.