Tear off the fig leaf of "guaranteeing top 3 across the entire network": AI large models are not "telephone poles" for randomly pasting GEO flyers.

A boss wanted to spend money on marketing, but was fabricated a "magical background" that would fall apart upon closer inspection by a GEO service provider out of thin air. This is not a plot from some absurd science - fiction movie, but a real - life commercial absurd drama happening right around us.

Just a few days ago, China National Radio Network lifted the curtain on this absurd drama with an investigative report.

A small - scale international logistics company that was just established in 2024, keenly noticing that more and more inquiry leads were coming from "AI recommendations", was eager to seize a niche in the large - model Q&A ecosystem.

So, with a meager monthly budget of two or three thousand yuan, they outsourced their brand promotion to a GEO (Generative Engine Optimization) service provider that claimed to be able to "handle dozens of keywords, guarantee a top - three ranking, and achieve multi - platform exposure". As a result, instead of achieving the so - called "dominant ranking" effect as expected, this startup was given a "magical background" by the service provider, such as "owning a factory of over 200 square meters" and "grandly holding a ten - year anniversary celebration" through AI - generated content in batches...

Even more ridiculous is that another moving company, deceived by the same sales pitch, spent nearly a month and several thousand yuan on service fees, but still got a zero ranking.

Under the absurd appearance lies the undeniable traffic anxiety of physical business owners in the AI era. As users' search habits irreversibly shift from traditional web engines to large - model AI dialog boxes, where the traffic is, there lies the business. This directly gives rise to the "desperate measures" traffic anxiety of enterprises.

01 Desperate Measures in the Face of Traffic Migration: Why Do People Still Believe in "Looking for a Sword by Marking the Boat" Scams?

The scammers' tricks are nothing new, but they can always accurately collect the IQ tax from devout users.

This year, the 3·15 Gala of China Media Group exposed a GEO cheating software that could automatically generate more than a dozen fake review promotion articles with exaggerated parameters around a non - existent fictional product within minutes, forcibly packaging it as the "industry leader". According to the report of China National Radio Network, some institutions even blatantly put forward low - price sales pitches that seriously violate technical common sense, such as "effective for all industries with 3000 yuan in a quarter" and "effective in seven days, full refund if ineffective".

In the author's view, this is a case of "looking for a sword by marking the boat" in the new business context. You know, the old bottle of traditional SEO can't hold the new wine of large models. However, some GEO service providers are frantically exploiting the traffic anxiety of business owners, using the cheating mindset that has been eliminated by the times to deceive audiences who don't understand algorithms. Large - model AI is becoming like a "telephone pole" plastered with small advertisements.

However, is it really that easy to expose the flaws of large models? An expert from the 360 Intelligence GEO product gave a very incisive in - depth analysis in an interview with China National Radio Network. "Large models are not traditional search engines. The answers they provide are dynamically generated based on corpora, real - time internet access, and user questions."

In his view, the Achilles' heel of traditional search engines lies in crawler crawling and the ranking of fixed web page weights. As long as you stack keywords and create backlinks, you can artificially intervene in the position. But large models are completely different. Their answer results are affected by multiple dimensions, including the training corpus (long - term memory), real - time internet search results (temporary queries), users' various question - asking methods, specific scenario requirements, and even changes in competitor information.

If we compare the two logically, we will come to an extremely absurd conclusion: Those service providers who dare to promise a "top - three ranking across the entire network" in all scenarios are actually guaranteeing that they can rent the same billboard in countless parallel universes at the same time, which is logically impossible and nothing but a pipe dream.

When it comes to high - value core keywords (such as "reliable decoration companies"), the cost of computing power optimization to lock in the top three positions on all platforms and at all time points will far exceed the few - thousand - yuan profit. Therefore, the real delivery method of low - price service providers is extremely crude: they mechanically split core keywords into thousands of long - tail useless words to increase the quantity, and then use AI to massively spread homogeneous and useless content. This is not large - model optimization at all, but the creation of a huge amount of cyber junk on the Internet.

So, how should a truly regular and effective GEO be done? First of all, we need to understand that large models don't refer to the so - called "ranking", but rather the structure of the content, its relevance, and the credibility of the information source.

Therefore, a compliant GEO must meet three strict conditions simultaneously: First, the brand must have a real "presence" across the entire network. If a company has no trace on encyclopedias or authoritative media, the algorithm of large models will not trust you in a short time. Second, avoid highly competitive keywords and focus on long - tail scenarios with extremely low competition and high granularity. This is the key to quickly breaking the deadlock. Finally, and most importantly, "AI - friendly" content must be produced. Large models can't understand abstract prose. Enterprises must abandon empty and false brand stories and instead use highly structured corpora with clear subtitles, direct answers to pain points, and similar FAQ and parameter comparison tables to "feed" AI.

In the author's view, at this stage, GEO is more like a "sowing" process of spring planting and autumn harvesting, rather than an immediate "harvest". It is about gradually establishing an authoritative information source that large models can recognize and trust, rather than playing a "probability game" by keyword stacking.

In this era, doing high - quality content and building a public - domain presence steadily is the key to breaking the deadlock, and there is no shortcut.

02 "AI Poisoning" Like Quenching Thirst with Poison: You Think You're Buying Growth, but Actually You're Buying a Summons

If creating cyber junk is just the GEO scammers "skimming the cream" from the bosses' budgets, then the hidden backlash brought about by such inferior marketing methods is pushing countless small and medium - sized enterprises to the edge of a high - voltage line of life and death.

In order to meet the customers' abnormal demand for low prices and quick delivery, service providers pour toxic corpora into the core information sources of large models by fabricating false "user reviews", inventing non - existent "industry experts", and forging non - existent "data endorsements". This reckless behavior has been clearly identified in the industry as despicable "AI poisoning".

However, large models are not defenseless fools. AI has a strong cross - verification mechanism. Once these fabricated and absurd data are detected by the algorithm, the over - packaged brand will not only quickly lose all credibility but also face severe weight reduction and display restrictions from the platform. In the ever - changing business battlefield, a repair period of several months is long enough for an enterprise to completely miss an entire business opportunity.

From a broader academic perspective, a trust crisis affecting the entire Internet ecosystem is erupting.

Professor Yao Jia from the Institute of Law of the Chinese Academy of Social Sciences once severely criticized the fatal consequences of this chaos with one sentence: "Large models give integrated conclusions. Users think they are smarter and more trustworthy, and it's hard to distinguish when ads are hidden in the answers." This practice of using the natural "intelligent deception" of AI to insert poisonous teaching materials will inevitably lead to a serious "bad money driving out good money" effect, ultimately destroying the trust foundation on which the entire AI era depends.

What's even more fatal is that this behavior is not only regarded as a decline in business ethics but also faces an unavoidable compliance crisis. At the end of April this year, the Cyberspace Administration of China launched a special campaign, clearly and strictly targeting "AI data poisoning" and malicious marketing using GEO technology as the key areas for rectification.

It seems that when an enterprise distorts its marketing actions into data poisoning, what you hire with a few thousand yuan is not a "growth promoter" at all, but a "grave - digger" who will bury your brand with their own hands. You think you're buying traffic and growth at a low price, but actually you're buying a court summons for yourself at full price.

03 "Good Money and Bad Money" from the Perspective of the Industry's Endgame: The End of Traffic is Trust

According to the macro - research report of Guosen Securities, it is predicted that by 2030, the global GEO market size is expected to soar from the current $24 billion to $100 billion.

However, on this new multi - billion - dollar track in its early stage of wild growth, the current situation is a mess. If the entire industry turns a blind eye to this chaos and allows the inferior ecosystem of "the one who can feed more junk gets on the list" to spread wantonly, then what we will destroy with our own hands is the most irreplaceable core asset after humanity enters the AI era - trust.

Fortunately, the power of rationality has begun to fight back. In April this year, nearly 40 professional media, industry organizations, universities, and leading technology companies jointly launched the "Responsible GEO Governance Initiative" in Beijing. This initiative completely tore off the mask of speculators and reached a consensus with great significance for the industry - the essence of GEO is by no means to teach AI how to cheat and lie, but to require enterprises to translate real and high - quality content into structured knowledge that AI machines can understand in a compliant and trustworthy way.

Actually, in this new business world completely reshaped by large models, the highest level of "pleasing AI" is definitely not cheating, but telling the truth. In the eyes of AI, "what ranking" is always just a fragile vanity. Whether the content you provide is real enough and worthy of being recommended to users by the algorithm is the real foundation of a brand.

GEO, which is supposed to be a vast and promising market worth $100 billion, I hope it won't be turned into a disgusting quagmire due to the short - sightedness of some people.

After all, in this era where "algorithms can determine life and death", trust is the most valuable brand asset for every enterprise.

This article is from the WeChat official account "Yongli Business Review". Author: Zeng Xianyong. Republished by 36Kr with permission.