I have a friend who operates a home furnishings brand. Upon hearing about the buzz surrounding GEO (Generative Engine Optimization), he decided to give it a shot.

A few days ago, we had dinner together. Curious, I inquired, "How's your GEO campaign performing?"

He replied, "Brand exposure has surged by 260%." Intrigued, I asked, "Who calculated that?" He answered, "The company that optimized it for me."

From his expression, I could tell he was somewhat skeptical of the data, yet it was challenging to verify its accuracy.

This scenario has been a recurring theme in the advertising industry for the past two decades. Media outlets would report their own viewership and circulation figures, leaving companies with no choice but to trust these numbers. It wasn't until third-party monitoring firms entered the fray that the industry became more regulated and transparent.

Now, it's GEO's turn to undergo scrutiny.

600 Million Users Are Ditching Search Boxes for Dialogue Boxes

According to the 57th Internet Report released by the China Internet Network Information Center (CNNIC) in February this year, as of December 2025, China's generative AI user base had skyrocketed to 602 million, boasting a penetration rate of 42.8%.

This means that one out of every two internet users is now leveraging AI for question-and-answer interactions. Just six months ago, that number stood at 515 million—a staggering increase of nearly 100 million in half a year.

This month, QuestMobile's latest rankings reveal that Doubao boasts 382 million monthly active users, Tongyi Qianwen has 167 million, and DeepSeek commands 130 million. Collectively, they approach the 700 million mark.

GEO, or Generative Engine Optimization, revolves around getting AI to mention, recommend, or cite your brand when responding to relevant queries.

Take my home furnishings friend as an example. He used to rely on SEO (Search Engine Optimization), where customers discovered his products through Baidu searches. Now, with GEO, AI mentions his brand in answers related to home furnishings.

Even I find myself occasionally purchasing AI-recommended products, simply to save time.

A 3 Billion-Yuan Business Lacks a Unified Benchmark

The GEO market is expanding at a pace that exceeds expectations.

According to Analysys, the domestic GEO market was valued at just 250 million yuan in 2025 but is projected to soar to 3 billion yuan in 2026—a more than tenfold increase within a year.

iResearch Consulting is even more optimistic, forecasting that the market will reach 51.8 billion yuan by 2030.

Business leaders are swift to respond. Over 70% of companies have already incorporated AI search optimization into their key AI transformation budgets. Brand builders, e-commerce players, and B2B companies are all actively seeking GEO service providers.

While exploding demand and growing supply should be positive developments, there's a fundamental issue: measuring effectiveness is arduous.

Suppose you invest 100,000 yuan in GEO optimization, and the service provider informs you, "Brand mention rate has increased by 40%." They measured it themselves, and you lack a secondary channel to verify the claim.

Moreover, many AI answers don't include clickable links.

AI mentions your brand, but whether users search for you, visit your website, or make a purchase—that connection is severed.

All you see is "AI mentioned me." How much actual traffic and conversion that mention generated? No one knows for sure.

And "optimization" tactics can sometimes be extreme. Everyone recalls the "AI data manipulation" exposed on 315.

Just the day before, on March 14, 16 GEO service providers had signed a "Generative Engine Optimization Industry Self-Discipline Pact" in Beijing, pledging "authenticity" and "positive optimization."

The next day, the gala used a fictional product to demonstrate how meaningless that pact was in practice.

Zen7 Labs Is Building That Benchmark

Liu Tao, co-founder of Zen7 Labs, states that the GEO industry currently excels in optimization but lacks measurement. Optimizers are grading their own work—no one trusts those numbers.

Zen7 Labs is stepping in as the third-party measurer.

Their tool is called "GEO Tag," a product from Zen7 Labs' GEO Analytics.

It's positioned as China's first independent third-party GEO performance measurement product, covering the full link: "AI engine visibility → real traffic → conversion ROI."

Full-link measurement means: from whether AI engines mention your brand, to whether users visit your website, to whether those visitors convert and how much they spend—all data is tracked under the same standard and displayed in a single console for verification.

Breaking it down, there are four key components:

1. Engine-side visibility monitoring.

You configure a list of questions you want to ask. The system regularly queries mainstream AI engines like Doubao, DeepSeek, Qianwen, Wenxin Yiyan, Yuanbao, and Kimi, recording whether your brand is mentioned or your URL is cited, and storing the original AI responses. It generates metrics like article citation rate and brand exposure rate.

This is crucial because many AI engine answers lack clickable links, leaving no trace when users click. For these engines, engine-side query records are the only effect data source. Without proactive queries, you'll never know if AI is recommending you.

2. Site-side behavior tracking.

A JS code snippet embedded on your website records the full visitor behavior chain: landing, browsing, interaction, and conversion.

It uses a GA4-style event model, supporting custom conversion events and amount reporting. Data is stored in an independent database with first-party anonymous identifiers—no device fingerprinting or cross-site tracking.

3. Invalid traffic detection.

Built-in IVT (Invalid Traffic) detection uses eight rules (two certainty levels: "definite" and "suspected") to annotate traffic in real-time during collection and perform batch reviews afterward. Each judgment is traceable to a specific rule number, allowing clients to self-review and adjust thresholds.

This is vital because no one previously knew how much bot and fake traffic GEO brought in.

4. Source-site ticket reconciliation.

You can issue one-time tickets on your domain, which are returned with reported data for verification, keeping a second reconcilable record on your server.

In short, data exists not just in GEO Tag but also on your server—both sides can match, preventing falsification.

As of August 2026, GEO Tag has been running stably in production, collecting 12,555 events across six sites. The numbers aren't huge, but for a newly launched measurement tool, stability matters more than scale.

AI-Era Trust Needs New Infrastructure

In Song Dynasty Quanzhou Port, there was a role called "ya ren" (broker) who weighed goods and assessed quality for both buyers and sellers. Merchants' own weight measurements weren't trusted, but the broker's were accepted by all. The port thrived not because of many ships, but because accounts were clear and transparent.

Peter Drucker famously said, "What gets measured gets improved."

This is precisely what GEO lacks—a scale that both sides trust.

Every shift in how humans access information spawns new trust mechanisms.

In the newspaper era, you trusted the newspaper's credibility.

In the Baidu era, you trusted search rankings and ad labels—though controversial, at least ads were marked "advertisement."

In the AI Q&A era, AI gives you a smooth, natural answer, and you can't tell fact from commercial placement from corrupted data.

This is more dangerous than the search era. Search results were separate links with visible sources. AI answers are blended text with sources shredded and mixed—impossible to discern.

GEO optimization might be the more profitable side now—high prices, strong demand, fast delivery. Abandoning it for measurement seems like a loss in the short term.

But in business history, such "self-amputation" choices are often the only way to build trust.

Otherwise, if the referee can join the game anytime, why watch?

If you have any thoughts, join the discussion in the comments!

If you found this helpful, like, share, recommend the article, and follow "AI Robot Teahouse."