There is a service that will get your brand mentioned by ChatGPT. It works by manufacturing the evidence. Aged Reddit accounts, paid upvotes, threads written to match the questions a language model likes to answer. The pitch is simple. AI cites Reddit more than almost any other source, so buy your way into Reddit and you buy your way into the answer.
I have watched this movie before. So has anyone who did SEO in 2008.
The Same Trick, A New Door
A recent piece on Search Engine Journal called the tactic the new link farm, and the label fits. The mechanics are identical to the old game. A trusted signal gets discovered. The signal gets manufactured at scale. The platform degrades until it has to act. Then the filter drops and everyone who bought in watches their rankings collapse overnight.
Google's Penguin update did exactly this to the link economy. Entire agencies that sold paid links disappeared in a single quarter. The brands that rented their authority instead of building it found out the rental was the whole problem.
AI citations are walking the same path, just faster. Firms reverse engineer the question patterns that models favor, then seed Reddit with branded answers through accounts that look aged and trustworthy. For a few months it works. The model scrapes the thread, repeats the mention, and the dashboard lights up.
Then two things happen at once. Detection improves, because platforms always eventually defend the signal that makes them valuable. And the signal itself rots, because every fake thread makes Reddit a slightly worse source for the model that trusted it.
You Are Bidding Up the Price of Poisoning Your Own Well
That last part is the one people miss. When you manufacture citations, you are not just risking a penalty. You are degrading the exact asset you are trying to extract value from.
Every paid thread lowers the trust the model can place in Reddit. Multiply that across every brand running the same play, and the citation you bought is worth less the moment you buy it. It is a zero-sum game where the winning move shrinks the prize for everyone, including you.
From building GEOflux, I watch which brands actually get pulled into AI answers, and the boring truth holds. Models cite sources the open web already treats as credible. Real participation, specific claims, consistent positioning across pages other people quote. The engine is summarizing a reputation. It is not counting upvotes you rented last Tuesday.
I made this case before, that AI search will not cite you just because you engineered the appearance of authority. The Reddit tactic is the cleanest example yet of engineering the appearance and getting the substance wrong.
What Actually Survives the Cleanup
Genuine community participation survives, because it was never a trick. If your team is in the threads where your category gets discussed, answering real questions with real expertise, that presence holds when the filter drops. It reads as what it is.
The test is uncomfortable but clean. If a platform deleted every account you paid for tomorrow, what would remain of your presence? If the answer is nothing, you do not have AI visibility. You have a rented signal on a countdown.
There is a second cost that never shows up on the invoice. The moment a buyer or a journalist spots an obviously seeded thread with your brand stitched into it, the trust you were trying to manufacture inverts. Now you are the company that games communities to look credible. In a market where people already assume marketers are working an angle, getting caught doing exactly that is expensive in a way no ranking report will ever capture.
This connects to a pattern I keep flagging, that the easy fixes for AI visibility fail the moment data shows up. The llms.txt file almost nobody reads was one. Bought Reddit citations are the next. Both promise a shortcut into the answer box. Both ignore that the model is looking for evidence you matter, not instructions you left at the door.
So the move is not to find the next exploit before it closes. It is to build the thing the exploit is faking. Publish claims specific enough to quote. Earn mentions on sources the model already reads. Show up in the communities you actually serve, under your own name, with something worth citing.
That work is slower. It is also the only version that does not evaporate when the platform defends itself. The brands that win AI visibility in two years will be the ones who spent these months earning citations instead of renting them.
The link farmers of 2008 are a useful ghost to keep around. They were not stupid. They were early to a signal and greedy about how they captured it. The ones who survived stopped gaming the link and started earning it. The same fork is in front of you now, with Reddit threads instead of backlinks. Pick the slow road. It is the only one that still exists after the cleanup.