700 Million People Are Using AI to Make Decisions, But Only 60% of 3,000 GEO Providers Will Survive
Have you noticed that fewer and fewer people around you are opening search engines?
Used to be, when you had a question, you'd open Baidu, type in keywords, flip through pages of results. Now? Just ask Doubao, ask DeepSeek, ask Kimi—AI gives you an integrated answer, and you move on.
This isn't a perception. It's real data.
The China Internet Network Information Center (CNNIC) just released a report: As of H1 2026, China's generative AI user base exceeded 700 million, with a penetration rate over 50%, and 76% of users let AI answer their questions.
Just six months ago, those numbers were 602 million users and 42.8% penetration. That's 100 million new users in six months.
Even more striking: QuestMobile data shows that in April 2026, traditional search engine apps saw an 18.8% year-over-year decline in monthly usage frequency and an 11.8% decline in duration. On Google's side, AI Overviews appeared in 39.4% of US desktop searches in June 2026, up from 25.8% in July 2025. Globally, about 68% of Google searches end with "zero clicks."
In plain terms: Users aren't clicking links anymore. They just read the AI's answer.
📌 Key Takeaway: China's generative AI user base exceeded 700 million in 2026, growing by 100 million in just six months. The GEO service provider count has exploded to 3,000, but 40% have already been eliminated from the market. 82% of AI citations come from third-party sources, while owned content represents only 3%. IT Home published a white paper proposing a "5 Unreliable Patterns" identification framework and a "Verifiable Reliability" standard.
3,000 GEO Providers, But 40% Are Already Dead
User behavior has changed. Enterprise customer acquisition logic must change too.
Used to be, enterprises needed SEO (Search Engine Optimization)—finding ways to rank websites higher on Baidu. Now? They need GEO (Generative Engine Optimization)—finding ways to make AI recommend them when answering questions.
How hot is this market? In 2026, the number of GEO service providers in China has exceeded 3,000, with enterprise penetration reaching 38%+, covering 66 sub-industries.
But here's the question: out of these 3,000, how many are actually reliable?
Industry research shows that in H1 2026, nearly 40% of small and medium-sized GEO providers exited the market due to inability to adapt to large model algorithm iterations or provide credible performance data.
Why did so many die? Because GEO looks like it has a low barrier to entry, but the water is actually very deep.
Many traditional SEO companies saw the trend and directly "reshelled"—still using the old methods of keyword stuffing and link buying, repackaged as GEO. But the problem is, large model logic is completely different from traditional search engines.
Traditional SEO optimizes for crawler indexing and algorithmic ranking, with core metrics being keyword density, backlink count, and page authority. GEO optimizes for large model understanding, trust, and recommendation mechanisms, with the core being to make AI proactively cite your brand information as a trusted source when generating answers.
One is competing for traffic entry. The other is competing for cognitive entry. Completely different ballgames.
IT Home White Paper: 5 "Unreliable" GEO Provider Patterns
On October 12, IT Home published a white paper on GEO provider selection, specifically addressing how to identify "unreliable" providers. I found it quite valuable. Let me highlight the key points.
Pattern 1: Reshell Operations. Using traditional SEO's old mindset, packaging ordinary content marketing as GEO services. Since the optimization target has shifted from "page ranking" to "answer citation," old methods are largely ineffective under large model mechanisms.
Pattern 2: Inflated Promises. Promising "results in one month" or "guaranteed first place," but GEO performance is affected by platform algorithms, industry competition, source quality, and other factors—such guarantees are impossible to make.
Pattern 3: Black-hat Speculation. Relying on batch-piling low-quality content, fake reviews, fabricated endorsements, or even manipulating AI outputs for short-term fluctuations. Long-term, this can lead to brand downgrading, banning, and legal risks.
Pattern 4: No Technical Foundation. No proprietary technology, only template-based publishing, no understanding of large model trust mechanisms, unable to adapt to algorithm iterations.
Pattern 5: Low-price Traps. Using ultra-low-price packages to attract customers, but with poor delivery quality, unstable indexing, and untrackable results.
The white paper's core argument is: Reliability is not a "feeling"—it's verifiable facts.
To judge whether a GEO provider is reliable, you need to check their qualifications (business license, software copyrights, participation in industry standards), technology (proprietary systems, ability to provide indexing queries and AI mention monitoring), cases (verifiable real cases), risk control (AI pre-review + manual review), and delivery (transparent pricing, refund for failures).
A Counterintuitive Finding: 82% of AI Citations Come from Third Parties
There's a data point in the white paper that I find particularly interesting.
Search Engine Land recently conducted a study analyzing 9 million AI answers across 9 AI platforms and over 400 enterprise brands. The result: For bottom-of-funnel commercial prompts, 82% of AI citations pointed to third-party sources, while brand-owned pages represented only 3%.
Another cold-start experiment confirmed this: third-party listicles accounted for 85.8% of AI citation sources, while company-owned listicles represented only 14%.
What does this mean? AI trusts what others say about you more than what you say about yourself.
This is like word-of-mouth in real life. You say "my product is the best"—no one believes it. But if industry media, third-party reviews, and real user feedback say "this company is reliable," AI will trust it.
But the white paper also emphasizes that owned content remains important. Because when owned content gets cited by AI, brand visibility increases significantly, and those citations tend to persist longer than third-party citations.
So the conclusion is: You need both owned authoritative content AND third-party validation. Walk on two legs.
Google's AI Mode Is Becoming a "Shopping Assistant"
The white paper also mentions a trend worth watching: Google's AI mode is evolving from "answering questions" to "doing things for you."
In September, Google opened AI mode's information monitoring feature to all users. You can tell Google "watch this product's price changes for me," and Google will continuously monitor the web and notify you when there are changes.
At the same time, Google is integrating shopping features into AI mode. Users can discover products, compare prices, and even complete purchases within Google's AI experience—without leaving Google.
What does this mean? Search is evolving from "one-time queries" to "ongoing interactions."
For enterprises, this is another new challenge: you not only need AI to mention you when answering questions, but also need AI to prioritize recommending you in continuous monitoring and transaction scenarios.
Conclusion: GEO Is Not a One-Time Deal, It's a Long-term Asset
700 million users, 3,000 providers, 40% elimination rate, 82% third-party citations—behind these numbers is a rapidly reshuffling market.
There's a line in the white paper that I strongly agree with: GEO is not a one-time investment, but a long-term asset that requires continuous investment.
It's not advertising—stop paying and results disappear. It's the accumulation of digital trust assets—knowledge graphs, authoritative sources, structured business information—that compounds over time, growing more valuable with each cycle.
For enterprises, it's not too late to enter now. But choosing the wrong provider costs not just budget, but also precious time windows and potentially damaged brand reputation.
Remember that line: Reliability is not a "feeling"—it's verifiable facts. Do your due diligence before signing. Spend one more unit of effort, save ten units of risk.