Last week I went looking for GEO advice on X, as research for a different article, and what I found was one prompt-pack funnel after another. A single account posted "200+ AI SEO Prompts" four times inside an hour. Every reply had a link-building ebook stapled to it, and clicking through got you offered a free strategy call. Most of those posts had zero likes, and the views they did get looked like bots. The whole thing had the exact texture of 2012-era guest-post spam, except that the product on sale was citations instead of backlinks.
That was when GEO spam stopped being a hypothetical for me. The supply side has already shown up, and the demand side (marketers panicking as their traffic charts sag) is walking in the door right now. If you've watched search for any length of time, you know how this movie goes. I find that encouraging, because knowing the plot means you also know how it ends and who's still standing when it does.
The playbook always repeats
Every new discovery surface the web has produced went through the same four acts. A new system starts deciding what people see, and someone works out that its inputs can be manufactured. An industry then grows up to manufacture them at scale, until the system's owner, watching its results get worse, starts purging and keeps purging forever.
You can trace it through the whole history of search. Directories were stuffed, and so were meta keywords. PageRank attracted link farms, then paid links, then PBNs, and helpful-content signals got AI filler. In each cycle the manipulation industry outran enforcement for a year or two and made real money. Then it got wiped out on somebody else's update cadence, and its customers' visibility went down with it.
Generative engines are the newest system deciding what people see, and I'd argue their input surface is juicier than PageRank ever was. An AI answer cites sources, and a source can be manufactured. It also trusts consensus, which can be seeded. You don't need to be a cynic to spot that arbitrage, only to have lived through SEO.
This cycle has one difference, and it cuts against the spammers: the enforcement loop got faster. Google's purges arrived quarterly and were announced, with a dashboard entry you could set your watch by. A model-side purge arrives whenever a lab ships new weights, silently and worldwide, all at once. The window between a tactic being discovered and being repriced used to be measured in years, yet forum seeding's window closed in about a week this August. Whatever the GEO spam industry builds sits on a floor that can vanish between a Tuesday and a Thursday, and it's the customers who are standing on it.
What GEO spam looks like right now
None of this is speculation. Every specimen is already out in the wild, and most of them are dressed up as legitimate services.
Start with citation-farm content, meaning networks of "review" and "best of" sites that exist purely so AI engines will retrieve them, selling placement the way PBNs sold links. Fake authority domains with official-sounding names appeared once retrieval started favoring official-looking sources. Forum seeding was the workhorse tactic right up until GPT-5.6 killed its ROI overnight. Then there are the agencies with "guaranteed AI citations" on the pricing page, a promise nobody can keep for a simple reason: citations churn about 45% every time an answer regenerates.
My favorite specimen is the self-promotional listicle, because the data on it is so brutal. One 100-query B2B study looked at brands that published their own "best tools" roundups, and 69% of the AI Overview citations those pages earned appeared in answers recommending their competitors. The brand did the work, the engine took the page as inventory, and the answer sent the buyer somewhere else.
Fair warning though: some of this will appear to work for a while, and that's the trap. PBNs "worked" too, right up until the update that turned them into a liability you had to disavow. When you rent visibility from a manipulation vendor, your asset lives on their infrastructure, and their infrastructure is always one enforcement wave away from worthless.
AI search traffic grew 835% this year. Is your content ready?
RankControl generates 26 content formats optimized for ChatGPT, Claude, and Perplexity. Published on your domain, matched to your brand.
A field guide to spotting GEO snake oil
These pitches are landing in founder inboxes right now, so I'd keep five phrases taped to the monitor. What ties them together is simple enough: anything sold as a shortcut to being trusted is, by definition, the thing trust systems are being tuned to catch.
The first is "guaranteed AI citations," and the churn math rules that promise out on its own. Roughly 45% of cited sources change every time an answer regenerates, and model updates reshuffle the rest without notice. Anyone guaranteeing placement in a system like that is either lying or renting you something too fragile to be worth the money.
If a vendor tells you "we know how the algorithm picks sources," ask which one they mean. Seven engines matter, and they visibly disagree with each other, sometimes about the same query in the same week. A vendor who talks about one algorithm is describing a product category they haven't measured.
"Our proprietary network of authority sites" is a citation farm with a brand kit. You'd be paying to attach your name to exactly the infrastructure the next enforcement wave is built to find, and when it burns, the vendor can churn its domains while you can't churn yours.
With "look at these results," check the screenshots for dates and for repetition. An undated screenshot of ChatGPT citing a client is a lottery ticket photographed at the moment it won, since regeneration churn means the same prompt an hour later may say something else. Real measurement shows trend lines across weeks for each engine, and a lone screenshot without them is theater.
The fifth is the "llms.txt optimization package." We ran llms.txt across 12 sites for 90 days and measured nothing, and Google says it doesn't use the file. A vendor who leads with it is selling the easiest deliverable on the menu, which tells you what the rest of the invoice looks like.
Built by the team that got cited in 48 hours.
Content generation, backlink building, AI visibility tracking, and Google rankings. One platform, zero guesswork.
Let me steelman the other side for a second
This is a slight detour, but it matters, because I don't want to pretend the spammers are stupid. They aren't, and the uncomfortable truth is that early manipulation usually does pay for a while. Guest-post spam paid, and PBNs paid. The people who ran them, took their profits and churned domains did fine, and some of the GEO equivalents will do fine too. If you run disposable affiliate sites, the arbitrage might even be rational, since you can burn the domain and keep the margin.
For a SaaS company the math still says no, and my reason is asymmetry rather than morality. Your brand is the one domain you can't churn. Every trend in retrieval right now makes identity more persistent (entity resolution, licensing, official-source weighting), so a burned reputation follows you into every engine that can cross-reference. The affiliate spammer is risking a website, while you'd be risking what your whole company resolves to when a model asks "who are these people and can I trust them?" With a downside that steep, a maybe-profitable tactic is a bad trade even in the months when it works.
The sales deck never mentions that a tactic doesn't need to be caught to fail you. All it takes is for the surface it rents to reprice, the way forum seeding repriced in one August week, and nobody at those agencies refunded the retainers.
The crackdowns already started
I think most marketers haven't fully processed that the first GEO enforcement wave already happened. It wasn't announced as one, which makes it easy to miss.
Look at August as a single event rather than two separate stories. Google ran a spam update that made a top-10 URL 1.8 times more likely to fall out of the top 100 than in a normal week. In the same fortnight, GPT-5.6 rewired ChatGPT's retrieval. Site-scoped queries jumped from under half a percent to 17-23% of its searches and "official" became a favorite retrieval term, while forum citations collapsed by at least 73% even as overall citations rose. The two engines didn't coordinate anything, yet both moved the same way, away from manufactured consensus and toward entities they can verify.
Project that forward and the incentives only sharpen. Answer engines are staking their products on answer quality, and now that ads have entered the picture, a polluted answer costs someone revenue directly. Within a year I'd expect published spam policies for answer manipulation, the way Google eventually formalized link spam. I'd also bet on provenance and verification signals getting explicit weight, because licensing deals already showed that engines will pay for sources they can trust. My last bet is enforcement that runs on the model-update cadence with no changelog, which is worse for spammers than Google's dashboard-announced updates ever were, since you can't file a reconsideration request with a model weight.
What survives every purge
So what do you build when you know the purges are structural? The boring answer is the true one: build what enforcement waves are trying to find. Every crackdown is really a filter searching for evidence that you exist, and verification is the moat, because spam by definition can't survive being checked.
That starts with extraction-ready content on a domain you own, and with entity facts that agree everywhere they appear, from your site and review platforms to partner pages and the knowledge bases models consult. On top of that come real reviews from real users and real mentions in places you don't control. First-party numbers only you can publish belong here too, along with docs that answer questions like someone who's actually met the product. That's basically the four disciplines that replaced the checklist, and a spam vendor can't sell you any of them, since each one requires the inconvenient step of being a real company.
There's one pattern from our side of the glass that I haven't seen anyone else mention. We re-run buyer queries across six engines every week, and what struck me in August wasn't the reshuffle so much as which brands barely moved. The ones with deep corroboration, consistent entity footprints, actual reviews and their own cited data wobbled for a week and then re-settled. The ones riding one strong surface, usually Reddit, dropped and stayed dropped. They were in the same category in the same week and got completely different outcomes, and what decided it was how many independent places the engine could check them, never anyone's cleverness.
My actual advice
I expect the GEO spam wave to be great for exactly one group: companies that were already doing the durable work. Every enforcement cycle clears rented visibility off the shelves and re-ranks whoever's left, so each purge ends up as a free distribution event for the verifiable. The founders getting outranked by farms today are the ones positioned to inherit the shelf, as long as they don't burn the interim buying shortcuts.
In practice, walk away when a vendor promises citations. When a tactic works because it fools a model, give it a mental expiry date, because the model's owner already has. Spend the budget on depth, corroboration and first-party data, and measure per engine every week, because enforcement now arrives silently and the only way to see a purge, or your own inheritance moment, is to be watching when it lands. The clean path is written up step by step in how to get cited by AI search without SEO spam.
If you'd rather have that as a quarter of real work than a vibe, this is my version. In week one, audit where your current AI citations come from and label each source the way an enforcement engineer would: owned, earned, infrastructure or farm. Anything in the farm column is exposure, even if you didn't put it there. Weeks two and three are for cancelling anything contractual that smells like the field guide above. That conversation is awkward, and you should have it anyway.
For the rest of the quarter, pour the reclaimed budget into two deep pieces only you could write, one first-party data asset, and the unglamorous corroboration work of reviews and profile consistency. Then keep the weekly watch running, which takes a few hours by hand, or is the thing RankControl's agents do every week while you build product.
I need to correct that "few hours" figure, though. It's a few hours in the first month. The months after that are where every manual process I've ever watched dies quietly, and the reshuffles won't schedule themselves around your busy weeks. Plan for that however you choose to solve it, and build like you intend to still be cited after the next wave, because somebody in your category will be, and the purge decides who.
See your first AI citation report in under 5 minutes.
No setup calls. No onboarding meetings. Connect your domain and see where AI mentions your brand right now.
