Why auto-reply bots get you banned (and what to do instead)

Every few weeks someone posts a version of the same story: they wired an AI tool to reply automatically to Reddit or X threads that mentioned their keywords, it worked for a fortnight, and then the comments stopped appearing to anyone but them. One widely-read account is titled, plainly, “I got shadowbanned 17 times before I figured out Reddit’s unwritten rules” on Indie Hackers.^[5]

The usual diagnosis is that the replies were low quality. That is often true and it is not the mechanism. The mechanism is that automated posting is detectable as a pattern, independently of what any single comment says.

What actually triggers a shadowban?

Platform anti-spam systems are not reading your comment and judging its helpfulness. They are looking at account-level behaviour, and automation produces a distinctive shape:

• Timing regularity. A human replies in bursts with long gaps. A cron job replies at intervals. Interval variance is one of the cheapest signals to compute.
• Latency after post creation. Consistently commenting within a minute or two of a thread appearing is not something a person browsing does repeatedly.
• Topical narrowness. An account that only ever appears in threads containing three or four commercial keywords looks like what it is.
• Link and phrase repetition. The same domain, or the same sentence skeleton with a noun swapped, across many threads.
• No non-commercial history. An account with no participation outside its own funnel has nothing to weigh against the promotional signal.

Reddit’s own Content Policy prohibits spam and manipulation^[1], and individual subreddits layer their own self-promotion rules on top — which is why a comment can be removed by a moderator even when it breaks no site-wide rule. Enforcement is mostly invisible: a shadowban means your comments render normally to you and to nobody else, so the first symptom is usually silence rather than a warning.

Is it the content or the pattern?

Both, but they fail in different ways and on different timescales. A bad comment gets removed by a moderator, once. A bad pattern degrades the account, permanently, and takes every future comment down with it — including the good ones you write by hand.

This is the part that makes full automation a poor trade. The upside is bounded by how many threads exist; the downside is the account you use for everything else.

What do the tools in this category actually do?

We ran a market scan of AI post-and-comment lead trackers in June 2026 — twelve searches and six page fetches across roughly seventy sources — specifically to map where each product sits on this question.^[6] The category splits cleanly:

The loudest complaint in the whole category, across every community source we read, was spam blowback from the first row: deleted comments, shadowbans, suspended accounts.^[6]Not “the AI wrote something embarrassing.”

What should a tool do instead of posting?

The useful division of labour puts the machine on everything except the send:

1. Find and score. Read whole threads, not just the top-level post, and score buyer intent against a profile you defined — so you spend your attention only on conversations that are actually about buying something.
2. Draft. Produce a reply that answers the question first. A draft is a starting point that costs nothing if you discard it.
3. Stop. Hand it to a human, in the browser, on their own account.

Everything that creates ban exposure — timing, cadence, which threads you choose to enter, whether this particular one deserves a comment at all — stays with the person whose account is at risk. That is not a compliance footnote. It is the whole design.

Does draft-only actually scale?

Less than automation, and that is the point. The binding constraint on this channel was never how many comments you could physically post; it was how many threads are worth commenting in. Scoring and drafting removes the expensive part — reading everything to find the few — and leaves you with a short list and pre-written first drafts.

If your plan needs hundreds of replies a day to work, this channel is the wrong one.

GrowthFuse is built on exactly this split: it reads the threads and writes the draft, and you post it yourself from your own account through the Chrome extension. Nothing is ever posted for you.