Over the past year, GEO (generative engine optimization — getting content cited in AI answers) has become a buzzword. Fewer users are in search engines, more are asking AI directly — we agree with that judgment, and the trend is real. But the way the market practices around it works runs counter to its principles. This article states our position clearly: what we endorse, what we don't, and who is suited to do it and who isn't.
First, a contradiction
Most GEO services on the market work by publishing content in bulk, filled with self-descriptions like "industry-leading" and "outstanding quality." But the logic by which AI assembles answers is to find fact blocks to piece together a response — parameters, comparisons, processes, explanations — not to read advertisements. A stretch of evaluative language with no information content has a citation value close to zero; a neutral statement like "when buying, first look at three indicators: thickness tolerance, conductivity and surface tension" is itself a ready-made answer component.
This contradiction is interesting: the direction where soft articles are churned out with brute effort is exactly the direction with the lowest citation rate. After reading a great deal of evaluative content, AI instead relies more on the few sources that have data and are verifiable. Squeezing into the former is easy; being cited by the latter is hard.
Multi-source consistency is real, but submitting articles is not a ticket
Another widely circulated claim is: once a viewpoint is cited by multiple news sources, AI will consider it reliable. Mechanically, that's correct — when the same fact appears in multiple independent sources, credibility gets a boost. But concluding from this that "you must submit articles to a large number of media outlets" is a misstep.
What AI wants is "independent" and "not self-described," not quantity. The same article scattered across dozens of templated sites actually voids the signal. The real path is far cheaper: one article with data and a viewpoint that gets naturally reposted by industry media and vertical sites, and multi-source consistency is achieved. And the very first step is actually free — go ask mainstream AI twenty questions your customers would ask, see whom it currently cites, and fill in whatever is missing. Blindly submitting articles is substituting resource anxiety for judgment.
The Ceiling: accept your fate on big keywords
We must acknowledge a mechanistic fact: AI has a head start bias toward big brands. The frequency of appearance in the training corpus determines that when asked "which product is good," the answer defaults to the well-known names. This is not something optimization can smooth away, and any service claiming it can make an unknown brand win the big category keywords should be viewed with suspicion.
The correct expectation for a small brand is not a comeback, but three smaller yet real things: capturing the question space in its own niche; getting included in comparison lists; and catching the name-verification step — when a customer hears about you at a trade show or from a peer and turns to ask AI "how is this company," your sources need to contain clean fact blocks, not silence.
The Floor: enclosing the long tail
The big category keywords belong to the giants, but in long-tail professional questions, the giants' content is empty. Big companies' official sites write brand narratives, not "how do you control annealing atmosphere" or "where is the cost difference between the two processes." These questions that get asked to death yet have no good answers are the territory of niche companies.
The only criterion for judging whether an industry is suited to GEO is: does this industry's customer use AI for information retrieval before closing a deal. If yes, and the more professional the questions, the more valuable GEO is — B2B manufacturing and industrial goods procurement, professional services, and high-ticket durable consumer goods are all typical scenarios, and their common trait is a large volume of long-tail questions with no one answering them. Conversely, customer acquisition for local instant life services happens within the location-recommendation loop, and the deal depends on instant supply within three kilometers, with AI search taking only a tiny share — selling "AI customer acquisition" to this group is mostly selling anxiety itself.
In closing
We judge that bulk soft articles are an arbitrage window — it exists in the gap before platform governance catches up and before users' discernment develops. The window will eventually close, because advertising pollution harms AI products themselves, and governance is inevitable.
What remains after the window closes is content assets: organizing a company's real accumulated knowledge into fact blocks that machines can read and people are willing to trust. Do this once and it works long-term; what we've been saying for the ten years we've built websites — a website is an asset, not an expense — holds true here as well.
GEO is not a tool for turning small brands into big brands; it is a tool for keeping "companies in a niche that have something real" from going invisible in the AI era. The ceiling is limited, the floor is solid. That is our full judgment at present, and we will act according to it.