Is Cheap GEO Worth It? Risks of Paid AI Posting [2026]

US$14. Twenty-eight minutes later, there were 13 new posts on the web. One hour after that, AI search cited one of them.

This isn't a vendor's ad copy. It's the record from an October 2026 research paper that actually bought a GEO service. Cheap GEO posting services really can get cited by AI quickly.

But it's also a path anyone can take. If you can use it today, your competitor can use it tomorrow. And once something is posted, you may not be able to take it back.

This article takes a defensive angle and covers the paper's experiment, the three risks of cheap posting services, and the five questions that tell you whether a proposal is just spam posting.

💡 Want to know how AI describes your brand right now? Start with a Free AI visibility check

Key takeaway: An October 2026 arXiv paper shows researchers paid a GEO vendor US$14, the vendor delivered 13 posts in 28 minutes, and AI search cited one within an hour. But the paper also states this is a single case and doesn't represent market success rates. And the path is open to anyone: after deletion, AI may keep citing the content for days (arXiv 2610.11932, 2026).

US$14, 13 Posts, 1 Hour: A Paper's GEO Service Experiment

On October 8, 2026, a research team including Fudan University published a paper on arXiv, "From Public Posts to AI-Search Citations: Measuring the Fragility of AI Search". It tested 10 AI search platforms, and in one part actually placed an order with a GEO vendor, recording what happened after the money was paid (arXiv 2610.11932, 2026). The paper is marked as an NDSS 2027 conference paper.

GEO (generative engine optimization) is the practice of making content easier for AI search to cite. Some sellers package it as a posting service: the vendor rewrites your material into many posts and publishes them across various public platforms.

The paper's experiment went like this:

Those identifier strings had never appeared in queries on the same topic before the purchase. So the researchers could be sure AI was using the post the vendor had just published.

The topic itself was a fake research topic the researchers invented: harmless, and nonexistent. Going by the times the paper lists, it took about 1 hour 28 minutes from payment (18:16) to the AI citation (19:44).

Where the Paper Draws Its Own Line

Don't jump to conclusions yet. The paper is very restrained about this experiment. The original text says: "This purchase provides one confirmed commercial workflow linking payment, public posting, and production AI-search citation; it does not estimate market-wide success rates."

In plain terms: it proves the path of paying, posting, and getting cited by AI works, but it's only one case. It doesn't mean every posting service on the market gets this result.

There are also two limits related to timing:

• The experiment ran in March to April 2026, and the paper states "exact rates come from one March–April 2026 campaign and may drift with platform updates". The numbers may change as platforms update
• The researchers looked at the ad content of 13 GEO vendors (5 in English-language markets, 8 in Chinese-language markets), but the paper says these are services the vendors claim themselves, with no verified results

We read the whole paper, and what we cared about most were these limits the authors wrote themselves. They decide what the paper can be used to say: it shows a path exists and the bar is low, but it doesn't say the path works for every brand or every topic.

So this article's position is clear: it's not teaching you how to do it. It's helping you understand what you're actually buying with a cheap GEO service.

Why Is AI Search Manipulation Possible? Because AI Often Cites Sites Where Anyone Can Post

AI search manipulation is possible because AI very often cites sites where ordinary people can post. The paper analyzed 10 platforms, 17,211 citations, and 6,356 source domains, and found that of each platform's top 10 most-cited domains, an average of 8.2 are places where ordinary users can post (arXiv 2610.11932, 2026).

The 10 AI search platforms the paper tested are: Grok, Doubao, ChatGPT, Wenxin, Google AI, Yuanbao, Perplexity, DeepSeek, Kimi, and Qwen.

First, look at how concentrated the citations are. The top 20 domains account for 20.5% to 70.8% of each platform's citations. Wenxin is the most concentrated, DeepSeek the most spread out.

Next, look at how high the bar is on these domains. The researchers tested 22 posting platforms matching the cited domains:

• 15 had low or medium barriers for both opening an account and posting
• 18 completed review within minutes, or showed no visible review delay

And taking this path doesn't require touching AI's model, index, or prompts at all. The original text says: "Executing this path requires no privileged access to the model, retrieval index, or application prompt; it relies only on ordinary publication access to public web platforms." Ordinary public posting access is enough.

Citation Decoupled from Publishing Control

The paper gives this phenomenon a name: the citation–governance gap, meaning citation is decoupled from control over publishing.

In other words, AI decides which domains to cite, but whether you can place content on that domain is decided by a different system, and that system's bar may be low. The paper's wording is "a mismatch between citation visibility and publication control".

The paper also points out something people often miss: "Repeated citation alone does not show that a source is trustworthy." Being cited by AI many times doesn't make a source trustworthy.

8 of 10 Platforms Cited a Fabricated Concept, but Only 3 of 10 for a Topic with Existing Discussion

The most commonly repeated number is 8/10, but it has to be read alongside another number. The researchers posted a fabricated concept that didn't exist anywhere online, and 8 of the 10 platforms cited their content within 7 days. The two that didn't were DeepSeek and Qwen. But for an opinion topic that already had lots of online discussion, the signal showed up on only 3/10 platforms, and the fabricated product topic also got only 3/10.

The paper's original text: "signals weaken sharply for the opinion topic (18.8%, 3.5%, 3/10) and remain modest for the fabricated product (36.4%, 4.7%, 3/10)".

Put these two sets of numbers together and the meaning is direct: the more obscure the topic, and the fewer people have written about it, the easier it is to fill the gap. On topics many people already discuss, a few new posts have a hard time drowning out what's there. This also means a brand is often most vulnerable when there's still little information about it online.

Caption: Many sites AI often cites are places where anyone can post. Ordinary posts can flow into AI answers and become cited sources.

The Same Path Can Be Used Against You: Forum Spam and AI Search Misinformation

The path that lets a vendor push posts into AI answers can also let someone else push false claims about you into them. The paper's fabricated concept was cited by 8/10 platforms within 7 days, while a topic with lots of existing discussion got only 3/10. That means the topics easiest to fill are the ones with little information online (arXiv 2610.11932, 2026).

This section is our inference. The paper didn't test it directly. But the logic is simple: this path only needs ordinary public posting access, and everyone has that.

Situations you can imagine look like this:

• A competitor writes a post on a Q&A site, in a neutral-looking tone, saying a certain brand's product has problems
• A former employee posts on a forum with outdated or flatly incorrect insider information
• Unknown people use a few accounts to repeat the same false claim across different platforms

If your brand doesn't have much information online to begin with, this kind of forum spam has a chance to become AI search misinformation within a few days, showing up in answers when someone asks AI whether your company is reliable.

Deleting It Doesn't Make It Disappear Right Away

Worse, even after the original post is deleted, AI doesn't stop citing it right away.

On day 11, the researchers deleted all the posts they had published, then kept querying every day. Doubao was still citing the deleted content on days 12 to 15, and a few other platforms also had scattered citations. The paper's conclusion is "source-page deletion does not immediately clear the citation layer": delete the source page, and the citations at the AI layer don't clear right away.

So you can't handle false information just by asking the other party to delete their post. You also need to put correct, clear, sourced statements somewhere you control. For how to build a single source of truth for your brand and regularly check what forums say about you, see building a brand source of truth.

But Don't Jump at Shadows

The paper also cautions: "This does not mean that every cited UGC page is manipulated, that cited pages are necessarily false, or that production pages can be labeled as paid campaigns without campaign-level evidence."

Not every user post cited by AI has been manipulated, and you can't call a page a paid spam campaign without evidence. When you see AI citing a forum post, first check whether the content is correct, then decide whether to act.

Borrowing another site's authority to pollute search results is a related problem. See how parasite SEO pollutes AI answers.

Three Spam Risks of Cheap GEO: Deleted, Treated as Spam, Losing Control of the Message

The spam risks of cheap GEO posting services fall into three types: posts may be deleted or throttled by the platform, Google may treat them as spam, and once a claim is posted you can't take it back. The paper explicitly describes operations like account farming, coordinated posting, and engagement manipulation with "This tier may violate platform terms" (arXiv 2610.11932, 2026).

Of the three, being deleted and being treated as spam are things that may happen. Losing control of the message is built into the structure.

Caption: Three risks of cheap posting services: posts get deleted, get treated as spam, and the message drifts away where you can't take it back.

Risk 1: Posts and Accounts Can Be Deleted or Throttled by the Platform at Any Time

The paper sorts GEO operations into tiers. One tier uses commercial tools to farm accounts, coordinate posting, and create engagement signals. The paper says this is a common practice among commercial GEO vendors, and that it may violate platform terms.

In their own testing, the researchers also found that some platforms have throttling (shadowban: posts look normal, but other people can't see them), automated account detection, or a strict risk of community moderation removing posts.

Then look at the vendor's accounts. The paper found the posts weren't published from new accounts. Account histories showed they had previously posted on all kinds of unrelated commercial topics. The paper goes on to list features that could be used to detect this kind of operation:

• An account suddenly shifting topics
• Near-identical posts across platforms
• Posting times bunched together
• New pages just published on low-barrier platforms getting cited by AI very quickly

These four points are exactly what cheap posting services most commonly look like.

The researchers have also reported their results to all 10 AI search platforms. Doubao, Qwen, Kimi, ChatGPT, Perplexity, and Google AI replied that they had received the report and would use it as a reference for future product improvements. The paper also gives AI platforms suggestions, including lowering the weight of newly published user content (the paper itself notes these are untested design directions).

So our inference is: low-barrier paths that work today may be demoted later. The paper didn't test this, and what each platform actually changes should be judged by the platform's own announcements. But a practice that has already been reported, with its detection features listed, is hard to expect to keep working.

Risk 2: In Google Search, It May Be Treated as Spam Outright

Google's definition of spam already includes generative AI responses. The original text of "Spam policies for Google web search": "In the context of Google Search, spam refers to techniques used to deceive users or manipulate our Search systems into featuring content prominently, such as attempting to manipulate Search systems into ranking content highly or attempting to manipulate generative AI responses in Google Search." (Google Search Central, 2026)

The consequences of violating it are also spelled out: "Sites that violate our policies may rank lower in results or not appear in results at all." Google detects violations with automated systems and, when needed, human review, which can result in a manual action.

The parts of the policy most relevant to spam posting fall into three categories:

But the full picture matters here, or it's easy to overstate.

First, forums themselves aren't a violation. In the site reputation policy section of the same document, Google lists "Sites designed to allow user-generated content, such as a forum website or comment sections" among the examples that are not inconsistent with the site reputation policy. The text also says "Having third-party content alone isn't inconsistent with the site reputation policy; it's only inconsistent if the third-party content is published on a host site mainly because of that host site's already-established ranking signals." So forum sites themselves don't violate this policy, and posting or joining discussions on a forum doesn't equal violating Google's policies. What violates them is spam accounts, spam posts, large amounts of content with no value, and placing content to borrow another site's ranking signals.

Second, this is a Google Search policy. It governs Google web search, including generative AI responses in Google Search. How ChatGPT, Perplexity, Doubao, and other platforms handle it depends on each platform's own rules. You can't apply this policy to them directly.

When we read this policy, we paid special attention to the fact that Google lists the examples that don't count as violations on the same page, and forum sites are among them. So when you discuss this with a vendor or inside your company, we suggest putting it this way: large-scale, repetitive content meant to manipulate is what causes problems, not posting on forums as such.

Which practices are white hat and which are black hat is summarized in white hat vs black hat SEO.

Risk 3: Losing Control of the Message. Once It's Out, You Can't Take It Back

Being deleted or treated as spam may happen. Losing control of the message comes from the structure of posting services themselves.

Posts a vendor publishes for you sit on the vendor's accounts, not yours. Even the researchers in the paper ran into this. They could delete their own experimental posts, but the 13 posts the vendor published weren't on accounts they controlled. They could only do their best to ask the vendor and the posting platforms to delete them. The original text says "we made best-effort deletion requests".

Extending this to ordinary brands, our inference is: if you pay a vendor to post for you, the right to delete those posts is usually not in your hands either. When the engagement ends, the vendor goes out of business, or the accounts are sold, the posts may still be out there.

Add two more findings from the paper:

• After deletion, AI may keep citing the content for a while (Doubao cited it for 4 more days after deletion)
• Being cited doesn't mean what you wanted to say made it into the answer. After the fabricated product topic was first cited, the citation rate was 36.4%, but the rate at which specific claims from the posts appeared in the answer text was only 4.7%

Put these three points together: what a cheap posting service buys you is a chance to be cited, not AI repeating your version. You control what gets posted, but not how AI uses it, for how long, or when it stops.

As for why cheap packages are cheap and how pricing works in the market, see the price question in how to read a GEO proposal.

How to Tell If a GEO Proposal Is Just Spam Posting: Ask These Five Questions First

Before you go looking up GEO reviews online, ask the vendor these five questions. Each one maps to a specific finding in the paper or in Google's policy. The detection features the paper lists (an account suddenly shifting topics, near-duplicates across platforms) are the basis for questions 2 and 3 (arXiv 2610.11932, 2026).

After each question, you can follow up with one more: can you show me a real example? A vendor who can name specific URLs, accounts, and review processes sounds completely different from one who just says they have lots of experience.

Our suggestion: if two or more of the five questions don't get a clear answer, that's when you should stop. This is our own threshold, not a conclusion of the paper. Adjust it to your own risk tolerance.

These five questions only cover spam posting risk. For everything else to look at in a GEO proposal, see the full evaluation method in how to read a GEO proposal.

The Honest Approach: Make Your Own Website a Primary Source AI Can Cite

A steadier approach than posting services is making your own website a primary source AI is willing to cite. The first key best practice in Google Search Essentials is "Create helpful, reliable, people-first content" (Google Search Essentials, 2025).

This doesn't mean avoiding every outside platform. The same Google page also suggests: "Tell people about your site. Be active in communities where you can tell like-minded people about your services and products that you mention on your site."

The difference is whether you're genuinely joining the conversation, or spam posting at scale with repetitive, fabricated content. The first is promotion Google encourages. The second matches scaled content abuse, spam accounts, and forum spam posts in the policy.

To Be Honest: The Cost of Doing It Right Is Time

There's one result in the paper we don't want to skip. The researchers published 40 posts on a self-hosted WordPress site, and none of them were cited within a week. The original text: "The weakest case was self-hosted WordPress: even 40 posts received no citation within one week."

The paper also stresses that these are thresholds for individual topics, not a general conversion rate ("We report these as per-topic thresholds, not a general conversion rule"). So it doesn't prove your own website is useless, and it doesn't prove your own website works better either. What it shows is that posting where AI rarely cites (the paper puts this self-hosted WordPress site on that side) makes it hard to be seen within a week.

The honest approach isn't a shortcut. It's accumulation. The cost is time, and in return you get three things a posting service can't give you:

• The content is yours: you decide when to edit, delete, or update it
• You decide the message: what's written is facts you've confirmed, not a version the vendor rewrote
• You can fix mistakes yourself: spot an error and fix it the same day, without waiting for someone else to reply

Our approach is to write content on clients' own websites. When we track AI citations for contracted clients, we use an AI citation tracking tool to run a fixed set of questions on a fixed schedule, and look at the long-term direction, not the moment one or two posts get cited.

So how do you write content AI is more likely to cite? Answer the question up front, put the brand name and the fact in the same sentence, and give a source for every number. These practices are covered in content AI wants to cite. Also, being recommended by AI and being cited by AI are two different things. See AI recommended vs cited for the difference.

Want Your Website to Be a Primary Source AI Will Cite?

Posting services work fast, but the results aren't in your hands. AI SEO Hacker can look at what content your website is missing right now and line up the topics to add first. Talk to us first.

Free consultation

FAQ

Does posting on forums or Q&A sites count as spam posting?

No. Google's spam policies (which govern Google Search) explicitly list forum sites or comment sections that allow user-generated content as not inconsistent with the site reputation policy, and Google also suggests being active in communities. What causes problems is large-scale, repetitive content meant to manipulate, such as spam accounts and spam posts on forum threads.

If AI cites it, does that mean the GEO service worked?

Not necessarily. The 2026 paper found that after the fabricated product topic was first cited, the citation rate was 36.4%, but the rate at which specific claims from the posts appeared in AI answer text was only 4.7%. Being cited and having what you wanted to say make it into the answer are two different things. For the difference, see AI recommended vs cited.

Once the paid posts are taken down, will AI stop talking about it?

Not right away. After the researchers deleted all their posts on day 11, Doubao was still citing the deleted content on days 12 to 15. And paid posts mostly sit on the vendor's accounts. Even the researchers could only make best-effort deletion requests, so ask whether posts can be taken down before you sign.

I found AI citing a false post about my brand. What should I do?

First confirm what AI actually says about you. See how to check if AI cites you. Then deal with the source: put clear, sourced, correct statements on your website. For the process, see what to do when AI gets your brand wrong. The paper's experiment shows AI may keep citing for days after deletion.

Do the paper's results apply in Taiwan too?

Use them as a reference, but don't apply them directly. The 10 platforms the paper tested include ChatGPT, Google AI, and Perplexity. The fabricated concept was cited by 8/10 platforms within 7 days, but a topic with existing discussion got only 3/10. The experiment ran in March to April 2026, and the paper also says the numbers may change as platforms update.

The Real Cost of Cheap GEO Services: Fast, but You Can't Take It Back

Cheap GEO services may really be fast: in the paper, US$14, 28 minutes, 13 posts, and an AI citation within an hour. But the same paper also tells us three things: this is only one case; for topics that already have discussion, the signal weakens sharply; and once posts are out, the right to delete them and how AI uses them are both out of your hands.

So when you evaluate a GEO proposal, instead of asking how fast you'll get cited, ask whether you can take it back if something goes wrong.

Cheap shortcuts leave risks on the web for a long time. Content on your own website is what stays your asset in the long run.

🎯 Cheap Shortcuts Can't Be Taken Back. Content on Your Own Website Is Yours

AI SEO Hacker offers an AI-driven SEO/GEO content service that puts content on your own website: from keyword planning and writing sections AI can understand, to a source for every number. Want to talk about your site first? Book a free consultation or leave your requirements.

Free consultation | Contact form

Further Reading

• How to Read a GEO Proposal
• Building a Brand Source of Truth
• Content AI Wants to Cite
• How Parasite SEO Pollutes AI Answers

References

1. From Public Posts to AI-Search Citations: Measuring the Fragility of AI Search - arXiv 2610.11932
2. Spam policies for Google web search - Google Search Central
3. Google Search Essentials - Google Search Central