Ask most churches how many people watched a service online for the first time last month and you will get a number. Ask how many of them were contacted afterwards and you usually get a pause.
The honest answer, at almost every church I have worked with, is close to none. And the reason is not that nobody cares. It is that follow-up grows with attendance while staff does not, so it is the first thing to quietly stop happening — and unlike a missed Sunday, nobody notices, because the people it fails to reach never complain. They just do not come back.
That makes the people who came once the largest group in most churches’ digital ministry, and the only one with no plan attached to it.
Why this one is a capacity problem
It is worth being precise, because “we should follow up better” has been said in every staff meeting for thirty years and nothing has changed.
The work is not hard. It is unbounded. Every person who shows up generates a small amount of follow-up work, so the total scales directly with how well your outreach is going — which means the better your digital ministry does, the more certainly this part breaks. Success makes it worse. That is the signature of a capacity ceiling rather than a values problem, and you cannot fix it by caring more or by putting it on a list.
Notice also what kind of work it is. Most of the effort is not writing the message. It is working out who to write to, what they already interacted with, whether anyone has contacted them before, and what question they seemed to be asking. That is assembly, and assembly is exactly the sort of thing a machine is good at.
What AI should and should not do here
Should: figure out who needs contacting this week and surface what is already known about them. Pull the question they asked. Find what your church has already taught about it — which, if you followed last week’s post, is now searchable. Draft a first version so a human is editing rather than staring at an empty box forty times.
Should not: be the relationship. This is the whole line for this series and it is sharpest here, because follow-up is the exact place churches are most tempted to fake it.
An AI-written message that says “I was thinking about you after Sunday” when nobody was thinking about anybody is a lie, and it is a lie told to somebody who is deciding whether your church is trustworthy. People work these out — the rhythm of it, the specificity that is somehow both precise and hollow. And the damage is not that one message failing. It is that every message you send afterwards is now suspect, including the real ones.
So the rule I would hold to: a named human reads and sends every message, and the message never claims a thought or feeling that person did not have. Drafted is fine. Assisted is fine. Automated intimacy is not.
What a good first message looks like
Short. From a person, with their name on it. Within a week, ideally within three days.
It refers to something real — the actual service, the actual question, the actual comment. It does not open with “we’d love to see you Sunday,” because that asks a stranger to do the single hardest thing available as their first step, and it tells them the online thing they just did was a preliminary to the real thing.
And it asks for nothing, or nearly nothing. A question they can answer in one line is the highest-value thing you can include, because a reply is what turns a contact into a conversation. “Was there anything in that you’d want to ask about?” does more work than any invitation.
Then somebody has to actually read the answer, which brings this back to capacity — but a reply from a person who has already reached out once is a very different workload from cold outreach to everyone.
The test for whether this is working
Not how many messages went out. That number will look excellent immediately and will mean nothing.
Count how many people replied, and then how many of those went past a single exchange. That is the number that only moves when the messages were actually worth answering, and it is the one that will tell you honestly whether you have built follow-up or just automated the appearance of it.
Most churches start this and find the reply rate is higher than they expected — people are, it turns out, quite surprised to hear from a church that wants nothing. That surprise is worth thinking about, and it is not really a story about AI at all. AI just made it possible to send the forty messages. Whether they were worth sending was decided by whoever you handed this ministry to, and by whether they were given the room to write like a person.
Part of the weekly AI series. The full guide is AI for churches.
