okaihub

Cold Email for AI Founders: What Still Works in 2026

Deliverability, list quality and reply rates for cold email in 2026 — what changed, what still works, and the mistakes that put you in spam.

Published 2026-08-07 · 4 min read

Cold email still works for AI products, and it still fails for the same reasons it always did: the list is wrong, the message is about you, or the domain cannot deliver. What changed is that the third problem got much worse, because inbox providers now enforce authentication that most founders do not set up.

This is the practical version, in the order that matters.

Deliverability first, because nothing else survives without it

Before writing a single message, check that your sending domain is properly authenticated. The three records are non-negotiable:

RecordWhat it doesIf missing
SPFLists which servers may send for your domainMessages are treated as spoofed and filtered
DKIMCryptographically signs each messageNo way to prove the message was not altered in transit
DMARCTells receivers what to do when SPF or DKIM failsProviders apply their own default, which is usually the spam folder
Set these up on the sending domain and verify with a testing tool before sending anything. This is an afternoon of work and it is the difference between the inbox and the spam folder.
  • Use a separate sending subdomain, not your root domain. If a campaign damages the sender reputation, you want that isolated from the domain your product and your website live on.
  • Warm up gradually. Start at a few dozen messages a day and increase over two to three weeks. A domain that goes from zero to a thousand sends overnight is behaving exactly like a spam operation.
  • Never buy a list. Beyond the obvious quality problem, purchased lists contain spam traps, and hitting one damages a sender reputation that takes months to rebuild.
  • Include a working unsubscribe. It is a legal requirement in most jurisdictions and it also reduces spam complaints, which are the single most damaging signal.

The list is the whole game

A hundred well-chosen recipients outperform ten thousand scraped ones, and not narrowly. The reason is that relevance drives reply rate, and reply rate drives sender reputation, which drives deliverability for everything after.

Build the list from evidence of the problem rather than from job titles:

  • People who asked a question about the task publicly, in a forum, community or social thread.
  • Companies hiring for the role your product assists — the job description is a description of the pain.
  • Users of adjacent tools, visible through reviews, community participation or public case studies.
  • People who attended a specific event or joined a specific community where the problem is discussed.

For each name, record where you found them and what specifically made them a candidate. That note becomes the first line of the email, and it is the difference between a message that gets read and one that gets deleted.

The message

Short. One idea. One ask. The structure that survives:

  1. 1

    Line one: the specific reason you are writing

    Reference the thing they posted, the role they are hiring for, or the tool they use. It must be true and it must be about them. This line determines whether the rest gets read.

  2. 2

    Line two or three: the problem, in their words

    Use the phrasing you collected during research. Not your product's category language — the way they describe it when complaining to a colleague.

  3. 3

    Line four: one sentence about what you built

    What it does, not what it is. No adjectives, no positioning language, no claim about being better.

  4. 4

    Line five: a low-cost ask

    "Worth a look?" or "Can I send a two-minute video?" Not a 30-minute call. The first reply should cost them almost nothing.

Follow-up, where most of the replies come from

A single message rarely gets a reply, and the first follow-up often does. Two or three follow-ups spaced several days apart is normal and is not harassment; five in a week is.

  • Follow-up one (3-4 days): one line, plus something new — a relevant example, a short note about a specific use case, an answer to an objection they did not raise.
  • Follow-up two (a week later): the useful thing on its own. A short guide, a template, a finding. No ask attached.
  • Break-up (a week after that): one sentence closing the loop, no guilt. Leave the door open and stop.

Every follow-up must add something. Repeating the same message with "just bumping this" is the pattern that gets marked as spam, both by the provider and by the reader.

What to measure

MetricWhat it tells youIf it is bad
Delivery rateWhether you reached the inbox at allFix authentication and warm-up before anything else
Reply rateWhether the list and the first line are rightNarrow the list; rewrite line one
Positive reply rateWhether the offer is interestingChange the ask, not the copy
Spam complaint rateWhether you are damaging the domainStop sending immediately and rebuild the list
Unsubscribe rateMild signal of mismatchUsually acceptable; watch the trend
Reply rate is the number to optimise. Open rates are unreliable because many clients prefetch images and register opens nobody performed.

When cold email is the wrong channel

  • Your buyers are consumers rather than businesses, and you have no legitimate way to reach them.
  • The product needs a demonstration to be understood — a text email cannot carry it, and a video reply rate is far lower.
  • You have no evidence-based list and would have to buy or scrape one.
  • Your market is small enough that you could message every person in it individually on a platform they already use. Do that instead; it works better and it does not risk a domain.

Cold email is a targeting exercise with a delivery problem attached. Founders usually work on the copy and should be working on the list.

More in Cold start