If we see the GEO industry of the past year as a new species growing at high speed, then 315 was more like a collective health check. The problems were not appearing for the first time; this time, the lesions were finally seen.
GEO is originally neutral. Generative Engine Optimization should help companies organize true, authoritative, and verifiable information into content forms that AI can more easily understand, call, and recombine. It should not fabricate facts; it should organize facts. It should not manipulate judgment; it should improve understanding. It should not pollute the information environment; it should improve information accessibility.
But in reality, part of the industry is going off course. Some people have turned GEO into another kind of "answer engineering": not optimizing content around facts, but designing information around desired outcomes; not helping AI understand brands more accurately, but trying to make AI more inclined to output preset conclusions; not building trustworthy content, but mass-producing content shells that look trustworthy.
That is why we want to say: GEO is sick.
By "sick," we do not mean the direction of GEO itself is wrong. We mean some practices have already become pathological. The illness is not in the technology, but in the method; not in the goal, but in the path; not in "letting AI see you," but in "trying to make AI see only you."
If this were only a traffic game in consumer goods or the broad internet sector, the problem would already be serious enough. Once it enters life sciences and healthcare, where evidence, professionalism, and truthfulness are highly sensitive, the issue is no longer just communication distortion. It may become cognitive misdirection, because what is affected is not only brand exposure, but also factual expression, evidence understanding, professional judgment, and even the starting point from which different roles understand the same question.
So where exactly is GEO sick?
First, it is sick in goal misalignment.
For many people, the first reaction to GEO is not "how can AI understand me more accurately," but "how can AI recommend me." The two goals look close, but they are fundamentally different.
The former is based on content construction: using higher-quality, more structured, and more verifiable information to improve AI's correct recognition of companies, products, viewpoints, and evidence. The latter is closer to outcome manipulation, aiming directly at the generated conclusion.
When GEO slips from "helping AI understand true information" to "competing for AI output conclusions," it starts to move away from content optimization and toward cognitive manipulation. At that point, content is no longer a carrier of facts, but a tool for influencing generation results. On the surface it looks like communication; in essence, it is pressure on the information environment.
A healthy GEO project should first answer: have we provided a more truthful, clearer, and more complete information foundation? Not: do we have more chances to become the name inside the answer?
Second, it is sick in hollow content.
Many so-called GEO contents today look abundant, but are hollow. They have high opinion density and low evidence density. The expression appears complete, but the sources are vague. They look like they "understand the industry," but once you ask about origin, applicable boundaries, version updates, or evidence level, they cannot stand up.
The biggest risk of this kind of content is not that it is badly written, but that it looks real. Wrapped in professional language, industry phrasing, and structured packaging, it creates an illusion of credibility. For ordinary consumers, this may cause misunderstanding. For fields that depend heavily on professional information, that illusion itself is a risk.
Especially in healthcare, content is never merely a wording issue. What evidence level supports a conclusion, what study design a viewpoint is based on, and whether a statement applies to a specific population, scenario, or point in time are not packaging issues. They are factual issues.
If GEO produces content that is highly readable but weak in evidence, then it is not optimizing information quality. It is only optimizing the appearance of information.
Third, it is sick in disguised authority.
More dangerous than hollow content is content disguised as authority.
A typical issue in the industry now is that large amounts of content try to imitate expert tone, consensus language, evaluation structure, and institutional expression, making information look "verified." But looking like something is not the same as being it. When content increasingly resembles authority but lacks clear sources, original basis, and traceability, GEO begins to slide from information organization into information disguise.
This is a very important warning signal. AI does not naturally equal truth detection when processing information. It is better at recognizing patterns, organizing text, and synthesizing statements, but only if the information environment it touches is healthy enough. If the external world is full of pseudo-professionalism, pseudo-citations, and pseudo-consensus, AI's answers may be very confident in language yet fragile in fact.
For brands, disguising authority may exchange for short-term visibility. For the industry, however, it overdraws trust. It overdraws platform trust, user trust, and the company's own long-term cognitive assets.
Fourth, it is sick in "feeding dependence."
Some GEO projects are essentially not building knowledge, but piling information. They believe that as long as they publish enough, distribute widely enough, and make the structure look "AI-friendly" enough, they will have a chance to gain higher weight in generative engines.
This thinking looks highly executable on the surface, but it is classic feeding dependence. It assumes AI understanding can be stacked through content volume, while ignoring a more fundamental issue: AI does not only look at who speaks louder. It also looks at whether information is consistent, verifiable, and corroborated by multiple sources.
Large volumes of homogeneous content, rewritten content, and low-information-density content may look like content prosperity, but in fact they create noise. In the short term, they may bring some visibility. In the long term, they may push the company itself into content disorder: many wordings, many versions, many citations, but no clear, unified, trustworthy knowledge base.
When a company outputs more and more information externally but finds it harder and harder to answer "which version is accurate," "where did this sentence come from," and "what is the basis for this judgment," the problem is no longer only communication. It is a content system problem.
Fifth, it is sick in monitoring only "whether we were mentioned," not "whether we were described correctly."
This is the most fundamental blind spot in many GEO projects today.
Some companies have begun paying attention to brand visibility in generative engines, and that is correct. The problem is that many monitoring systems still remain at shallow indicators such as whether the brand appeared, how many times it appeared, what its position was, and how many scenarios it covered. They do not continue asking: was the mention accurate? Were key facts omitted? Were concepts confused? Was evidence mismatched? Were there wrong associations among brand, product, disease, mechanism, and indication?
This is why we say many GEO projects do not lack traffic. They lack diagnosis.
Without diagnosis, there is no real correction. Without correction, there is no "optimization." Counting mentions without checking accuracy is like looking only at exposure while ignoring misdiagnosis. The data may look attractive, but the system's problems have not been solved; they may even be concealed.
For the life sciences industry, this is especially critical. The real value here is never merely "being answered," but "being answered correctly"; not merely "being seen," but "being accurately understood"; not "generating more content," but "reducing wrong cognition."
So what is GEO's real root illness?
In our view, on the surface it is a deviation in communication methods. At a deeper level, it is disorder in the content system.
Many companies are not unable to do GEO. They are not yet ready to do healthy GEO. What they lack is not a few prompts, several distribution tricks, or a handful of AI-rewritten articles, but a content infrastructure that can support AI's correct understanding: a unified knowledge base, traceable evidence chains, clear version management, structured mapping between internal and external content, and continuous monitoring and correction mechanisms for misinformation.
If these foundations are absent, so-called GEO can easily degenerate into another form of content placement. It may create presence, but it is difficult to build credibility. It may produce short-term results, but it is difficult to form long-term assets.
Healthy GEO should not be built on "how to influence answers," but on "how to govern content." It is first knowledge engineering, then communication engineering. It is first trustworthy expression, then traffic distribution.
This is also why GEO today should no longer be understood as a simple marketing action. It is more like a new dividing line. On one side, companies continue using old traffic thinking to compete for answers in the AI era. On the other, companies begin to admit that what they truly compete for is not whether AI mentions them, but whether they can provide a more trustworthy content foundation when AI understands the industry, organizes facts, and generates answers.
If the core question in past content competition was "who says more," then in the generative engine era the core question is becoming "who deserves to be trusted more."
This is GEO's real diagnostic conclusion.
GEO has no original sin. What is truly sick are methods that try to influence AI cognition with pseudo-content, pseudo-authority, and pseudo-consensus. They distort "optimization," alienate "visibility," and push a direction that should help information flow more accurately toward a higher-order manufacturing of information noise.
For companies, especially life sciences companies, the real question today is no longer "should we do GEO," but "is the GEO we are doing building trustworthy cognition, or creating new cognitive risk?"
The next step, more important than criticism, is prescribing treatment.
GEO's problems cannot be solved by saying "oppose false information." They need a new treatment logic: monitor first, then diagnose; clarify the lesions first, then fill evidence gaps; rebuild the trustworthy content system first, then discuss external cognitive optimization.
This is also the next question we want to continue discussing:
When GEO is sick, what prescription should companies truly write?
