AI explanation of this nonsense:
I’ll look up how these forum signup bots are typically built and what operators use them for.They are the old forum-spam industry with a language model bolted on. The accounts and the comments are not from curious users or from an experiment that wandered onto your board. They are cheap automated accounts whose job is to plant something that search engines or other AIs will later treat as a real human mention.
Where they come from
Forum spam has existed since the early 2000s. Tools like XRumer crawled for known forum software, registered accounts, and dropped links in posts and signatures. The labor was often sold as black-hat SEO packages: bulk account creation plus link placement, run from scripts and cheap proxies.
What changed is the text. Old bots posted keyword salad or the same template everywhere, which moderators and spam filters caught easily. Once cheap LLM APIs existed, operators swapped the template for a prompt: read the thread, write a short on-topic reply, sometimes wait a few posts before dropping a link. SentinelOne documented one example, AkiraBot, which used OpenAI-generated messages, bypassed CAPTCHAs, and hit hundreds of thousands of sites. Discourse admins have described the same pattern: accounts that solve captchas, confirm email, fill a plausible profile, then post.
A newer variant is explicit “AI agent” networks (for example the iLands agents that emailed and tried to register on Mastodon in 2026). Those announce themselves. Most of what you see on ordinary forums does not.
What they are trying to do
The comment is usually not the product. It is cover, or a delivery vehicle, for one of these:
- Backlinks and brand mentions. The durable goal is still a link in the post or the signature, or a profile URL, pointing at a client site, an affiliate offer, or a scam SEO service. Even nofollow or UGC links get chased, because operators sell volume, not proven rankings. Google has long listed forum-comment links made to manipulate rankings as link spam; the bots keep running because the marginal cost of another post is near zero.
- Reputation warming. Better bots post a few harmless, on-topic replies first so the account looks lived-in, then drop the promotional link days later. Forum admins have watched exactly that sequence: plausible answers, then an unrelated commercial URL.
- Poisoning AI answers. Forums, Reddit, and Quora are heavily scraped into search and into chatbot answers. Marketing shops now plant promotional sentences so that when someone asks ChatGPT or Google’s AI a buying question, the bot’s line gets quoted back. Research in 2026 showed that a short planted snippet on a user-generated site is often enough to steer an AI agent’s recommendation. That is why the comments are semi-readable and topically aimed rather than pure gibberish: the text has to survive into a training scrape or a live retrieval.
- Account inventory. Some signups never post. Those are stock: aged accounts to sell, or shells for later spam, phishing, or vote manipulation. Creating users is also a crude load test against weak registration forms.
- Direct pitches. A minority are selling the spam itself (“we can get you on page one for $30 a month”) or pushing a product in the thread.
The semi-readable quality is a side effect of the model, not a goal. The prompt is usually thin, the model has no real stake in the thread, and the operator optimizes for “passes a quick human glance and a spam filter,” not for being right. That produces the familiar tone: grammatical, vaguely relevant, slightly off, and empty.
Nobody is running these to improve the forum. The intent is commercial: links, mentions that search and AI systems will count, and accounts that can be reused. The boards are just the cheapest place left where a stranger can still publish text under a fresh username.
