Roughly a third of the pages AI engines cite are already dead or hallucinated. We know, because we map them. So when a vendor offers to lift your AI visibility by buying your brand onto more low-quality pages, understand what they're really selling: lottery tickets, in a lottery that's already broken.
Most “brand mention” packages sold under the GEO banner are rented citations. A rented citation is a brand mention you paid to place on a page you don't own, and like any rental, you're out the moment the landlord changes the locks. The landlord is the model. The lock change is the next ranking update. When it comes, the mentions you rented stop counting, and the invoices you paid bought nothing. It's black-hat link building in a new costume. Decline it.
We have a stake in saying so. Enginekick earns AI citations for B2B SaaS with zero outreach emails, so the paid-mention pitch is the one we get measured against most. We've also pulled apart enough of these programs to know exactly what's inside them.
~1/3
of AI citations point to dead or hallucinated pages (our mapping)
2012
the year Penguin gutted paid links, the precedent here
12–18mo
our public prediction for an AI authority filter
What GEO vendors are actually selling you
Strip the deck away and the same handful of tactics shows up again and again.
- Manufactured urgency from “studies.” A stat like “70% of AI visibility comes from third-party sources” gets stretched into a mandate to mass-produce mentions. The number might be real. The conclusion drawn from it usually isn't.
- “Partnerships” that are paid-link inventory. Framed as relationship building. In practice it's a rate card for placements.
- PBN mentions at a markup. Brand drops on private blog networks, sold for ten to fifteen times the cost of a normal link.
- Topically irrelevant pages. The site selling you a placement has one page on LMS software and another ranking the best crypto wallets. The model can see that too.
- Reddit astroturfing. Aged accounts post your brand across subreddits it has no business in, half of them removed within a month for breaking community rules.
The client side is where it gets ugly. The agency drops a “placement opportunity” into a Slack channel for approval. A junior marketer who has never evaluated a referring page signs off on a $250 fee for one sentence about your brand. The agency pays the publisher, then bills you to recoup it. The slimy part is the part nobody puts in the proposal. This is not a secret, Gaetano DiNardi documented the same vendor playbook in Search Engine Land. It gets sold as innovation anyway.
Paid mentions are link building in a new costume
Take the costume off and the mechanics are familiar. You pay a fee. A sentence about your brand appears on a page you don't control, usually one already stuffed with commercial anchors pointing at your competitors. That's a paid link with a thesaurus applied to it.
Google spent more than a decade learning to spot inorganic links, starting with the first Penguin update in 2012. It now treats paid placements on low-quality sites as exactly what they are. The AI platforms are earlier in that cycle, not exempt from it, and Google has already signaled that inauthentic mentions aimed at steering AI answers are a problem it intends to handle. The search industry has watched this movie enough times to know how it ends.
And think about what a citation actually is to a model. It's a risk decision: “will I look wrong if I repeat this?” A sentence sitting on a thin page surrounded by paid links to your rivals is a high-risk thing to quote. You're not making yourself safer to cite. You're doing the opposite.
The pages you're renting are the first to disappear
Here's the part the volume pitch ignores. When we map the sources the major engines actually cite in a category, roughly a third of those citations point to pages that are dead or hallucinated, dead links, pages that moved, URLs the model invented that never existed at all. That's the citation graph as it really is, not as a sales dashboard pretends.
So the inventory a mention vendor sells you sits squarely in that disappearing third. You're not buying a seat at the table. You're buying a seat that gets cleared the next time the model refreshes its index. Earned citations behave differently, because the sources that earn them, the real publications, the real communities, the directories buyers actually use, have the durability to survive a refresh. That gap between rented and earned is the whole game.
| Dimension | Rented citation (paid) | Earned citation |
|---|---|---|
| What it is | A sentence you paid to place on a page you don't own | A reference a credible source gives you on its own |
| Where it sits | Thin pages, often beside paid links to rivals | Real publishers, communities and directories buyers use |
| How a model reads it | A paid drop, high-risk to quote | Corroboration, safe to quote |
| Durability | In the disappearing third, gone at the next refresh | Survives index refreshes and authority filters |
| Disclosure | Rarely disclosed, an FTC and policy risk | Nothing to disclose, it was earned |
| Compounds? | No, you re-rent every month | Yes, consensus builds over time |
Mention rate is the wrong scoreboard
The pitch rests on treating AI visibility as a volume game. Buy enough placements, the logic goes, and your “mention rate” climbs. That scoreboard is broken in two ways.
First, visibility is sampling, not counting. Ask ChatGPT “best [your category]” ten times and you'll appear in some subset, four out of ten, then six, then three. One screenshot is noise. What moves that distribution is being described the same way across sources the model already trusts, not the raw number of pages your name sits on.
Second, the model weighs where the mention lives. A drop on a topically irrelevant page loaded with paid competitor links doesn't read as validation, it reads as a paid placement, because that's what it is. Volume without relevance can muddy what the model thinks your brand even is, and entity confusion is far more expensive to undo than a slow start.
The window is closing, and the bill comes due
Some of this works right now. That's worth admitting plainly. It works because LLM citation systems are still immature next to Google's spam detection. For a window, models reward signals a mature system would ignore. Then the countermeasures arrive, the training data gets cleaned, and the cheap mentions stop counting.
Our prediction, on the record: within 12 to 18 months, at least one major engine, ChatGPT, Gemini, Perplexity or Claude, ships an authority filter that discounts paid and low-quality mentions, the way Penguin gutted paid links in 2012. The day it lands, brands that rented their citations watch their mention rate fall off a cliff in a single release. We'll grade this in public, hit or miss.
The brands caught on the wrong side won't just lose the lift. They'll be left explaining a reputation stapled to crypto-wallet listicles and a brand entity the models no longer parse cleanly. There's a legal seam too: the FTC requires a clear disclosure whenever a mention is paid for, and negotiated placements dropped into editorial content almost never carry one. That's worth a conversation with your own counsel before you approve a single invoice, a flag from us, not legal advice.
How to vet a GEO vendor before you pay
When a vendor leans on off-site mentions, three claims do the heavy lifting. Each deserves a hard look.
| The claim | The reality |
|---|---|
| “Most AI brand discovery comes from third-party sources” | Probably true, and beside the point. That third-party sources matter doesn't prove that buying low-quality mentions makes a model recommend you. |
| “Listicles and third-party pages are the lever” | Sometimes. It's not a license to pay your way onto thin, AI-generated listicles that exist to sell placements. |
| “AI search is different, so the old quality rules don't apply” | Google says the opposite about its own AI features. Standard Search policies still govern, with no spam exemption for AI Overviews. |
A few questions cut through the rest. Ask to see the actual pages your brand would appear on. Ask whether the publisher is paid, and whether the placement gets disclosed. Ask what happens to the mention in 90 days. Ask how they measure lift, and whether “lift” means pipeline or a screenshot. The answers sort the operators from the inventory resellers fast.
What actually earns durable AI citations
The durable play is unglamorous, which is exactly why it holds. Competitors can copy your blog overnight. They can't copy being described the same way, by independent sources, across the places a model trusts. Consensus is the only real moat in AI search, and consensus can't be bought by the unit.
Win mentions that are real category validation, from credible, topically relevant brands, reputable publishers, and communities you actually belong in, the kind a model reads as corroboration rather than a paid drop. We made Swydo the most-cited brand against its direct competitors across every major LLM this way, with 100% organic backlinks and not one outreach email, on an SEO foundation that grew AgencyAnalytics from roughly $9M to over $20M in ARR. The full breakdown is in our case studies.
Then make your own pages quotable. Models read passages, not pages, and can't follow a “learn more” link mid-answer. The claim that earns the citation has to be a clean, self-contained sentence sitting where it can be lifted. Here's the difference in practice.
✗ HEDGED, USELESS TO A MODEL
“While outcomes vary by a number of factors, many teams find our platform can help improve reporting efficiency in certain cases.”
✓ CLEAN, SAFE TO QUOTE
“[Your product] turns a week of manual client reporting into a 20-minute job, with Google Ads, GA4 and Meta in one dashboard.”
A model lifts the second one and skips the first. One is a checkable claim it can attribute without looking wrong; the other hedges itself into nothing. That's the core of how generative engine optimization actually works, and it's the opposite of renting space on someone else's spammy page. Then track the right thing: we measure citation share as a distribution across ChatGPT, Perplexity, Gemini and Claude, every prompt sampled and averaged, so model randomness can't masquerade as progress. That whole loop is The Citation Engine.
What to do instead of buying mentions
Cancel the placement invoice. Then put that budget and attention here.
- Find out where you actually stand. Audit how the major engines describe and cite you today, before you spend a dollar changing it. Our free AI visibility checker is a start.
- Rewrite your money pages for liftability. One clean, checkable claim per page, near the top, where a model can quote it without risk.
- Earn presence on the sources that matter. Find the publications, communities and directories the engines actually cite in your category, and become a real part of them.
- Pick sources that survive. Stay out of the disappearing third. Confirm a source is live and trusted before you invest a minute in it.
- Track citation share, not screenshots. Sample every engine over time and watch the distribution, so you know what's working and what's noise.
None of that is rented. All of it compounds.
Straight answers
Do paid brand mentions actually improve AI visibility?
Sometimes, for a while. LLM spam detection is still young, so volume from low-quality pages can nudge how often a model names you. The bump is temporary and risky. Once the platforms tighten their authority signals, paid mentions on thin sites stop counting, and they can leave your brand tied to spam.
Is buying brand mentions against the rules?
It runs against the spirit of every major platform's guidelines, and Google has publicly flagged inauthentic mentions aimed at AI answers. There's an FTC angle too: paid or compensated mentions are supposed to be disclosed, and most negotiated placements aren't. Treat it as a compliance risk, not a clever loophole.
What's the difference between a paid mention and an earned citation?
A paid mention is a placement you bought on a page you don't control, usually sitting next to paid links to your competitors. An earned citation is a reference a credible source gives you on its own, because you are genuinely part of the category. A model reads the first as a paid drop and the second as corroboration.
How do you get cited by ChatGPT without outreach?
Put checkable, self-contained claims on your own pages so a model can lift them without risk, and earn mentions from sources it already trusts, the real communities, reputable publishers and credible brands in your space. Then track citation share across engines over time, not a single screenshot. That is the whole game.
You can't buy your way into being the answer AI gives. You earn it, through real corroboration and pages clean enough to quote, on sources that hold up when the platforms tighten the screws. Rented citations are a fast route to the wrong destination, and the eviction is already on the calendar. Run our free AI visibility checker to see where you stand, or book a strategy call and we'll audit your AI and search visibility live.