For most of 2026, the GEO industry has been arguing about effectiveness without evidence. Providers claimed results. Clients doubted them. No third-party data existed to settle the question.
On July 29, that changed. A GEO brand monitoring team published the first empirical stress-test of what happens when an AI search platform changes its algorithm — 2,182 brand visibility measurements, 22,510 citation links analyzed, 30 business scenarios tracked, every single day from July 1 through July 29. The test covered the period when Doubao (ByteDance's 600-million-user AI platform) implemented a major algorithm update in mid-July.
The results are not ambiguous. They are the closest thing to proof that the GEO industry has produced.
📉 The Crash — July 13-17
The monitoring data shows a synchronized collapse. Beginning around July 11, brand mention rates began to decline across all monitored scenarios. July 12 saw a brief rebound. Then July 13 through 17 — five consecutive days — every monitored scenario across two industries (education and mental health services) fell into a simultaneous trough.
This was not a gradual adjustment. It was a platform-level event. Doubao changed its source weighting rules, and for roughly five to seven days, brand visibility in AI answers was depressed across the board. No brand was immune. The question was not whether brands would be affected. It was whether they would recover.
♻️ The Recovery — Who Came Back and Who Didn't
Starting July 18, the data split.
The brands that recovered — and in most cases surpassed their pre-crash mention rates, reaching 98-100% by month-end — shared a common structure. They had official website content pages indexed by the platform. They had presence on vertical industry portals. They had "how-to" and "selection guide" content that AI models could cite as authoritative reference material.
The brands that did not recover shared a different structure. Their citation profile was dominated by soft-article placements — content published on Sohu, Toutiao, and similar platforms, with no official website content pages. One unrecovered brand accumulated 388 citations from such sources in a single month — but after the algorithm change, none of them counted. From July 22 through 28, its mention rate was 0-25% on five of seven days. It never returned.
The contrast could not be clearer. 388 soft-article citations — zero recovery. 58 citations from a single competitor's industry guide content — stable visibility throughout. The algorithm change was not a content quality filter. It was a source identity filter. The AI stopped treating self-media platforms as authoritative, and every brand whose visibility depended on them lost it.
🔬 What the Test Proved — and What It Didn't
Three conclusions emerge from the data with a level of confidence that the GEO industry has never been able to offer before.
One: content assets win. Soft-article placements lose. This is not opinion. It is the empirical outcome of 2,182 measurements. Brands whose AI visibility was built on owned content — official websites, industry portals, structured knowledge bases — experienced a 5-7 day dip and then recovered. Brands whose visibility was built on rented content — paid articles on third-party self-media platforms — experienced a crash from which they did not recover.
Two: authority is the only durable moat. Across the entire test period, brands with authoritative source verification — Baidu Baike entries, ICP-filed official websites, third-party news citations from recognized outlets — were the most stable. Their mention rates fluctuated the least. Their recovery was the fastest. The algorithm change did not penalize them. It filtered out the noise around them.
Three: the monitoring matters as much as the optimization. One finding in the data that had nothing to do with brand strategy: complaint platforms — 黑猫投诉 (consumer complaints), 12315 (government consumer protection), 企查查 (business registry) — appeared in AI answers 7 times during the test period, with 6 instances showing substantively negative conclusions (consecutive losses, regulatory warnings). The battlefield for brand reputation has moved from search engine result pages to AI answer text. If you are not monitoring what the AI says about your brand, you are invisible to your own reputation risk.
🌏 What This Means for Overseas Brands
The stress-test data is the most actionable intelligence on GEO effectiveness that has ever been publicly available anywhere. For an overseas brand evaluating its China AI search strategy, it translates into three specific imperatives.
First, do not build your AI visibility on rented land. Soft-article placements on third-party platforms are the GEO equivalent of building a house on someone else's foundation. When the platform changes its source rules — and platforms will continue to change their source rules — that visibility disappears. Own your content real estate: official website, Baidu Baike, Baijiahao. These are the assets that survived the July stress test.
Second, the quality of your citations matters more than their quantity. The unrecovered brand had 388 citations. Its competitor had 58. The competitor is now the AI's default answer. The math is brutal but clear: one well-structured industry guide on an owned platform outweighs nearly four hundred soft-article placements on rented platforms. GEO is not a volume game. It never was. The stress test just proved it.
Third, you need the infrastructure before you need the content. Every brand that survived the stress test had the same foundation: official digital identity (Baike), owned publishing channel (Baijiahao), and cross-referenced authority signals (schema-marked website, news citations). Building that foundation requires what most overseas brands lack at entry — a Chinese business entity. The brands that solve this structural problem first will build everything else on solid ground. The brands that skip it and start publishing will build on sand.
📋 What the Stress Test Teaches Every Overseas Brand
- Don't build visibility on rented land. Soft-article placements disappear when platforms change source rules. Own your content — official site, Baike, Baijiahao.
- Quality beats quantity, decisively. One well-structured industry guide (58 citations) outperformed 388 soft-article placements. The math is settled.
- Infrastructure before content. Every brand that survived had the same foundation: verified digital identity, owned publishing channel, cross-referenced authority. Build that first.
🏢 The BPP Perspective
The July stress test is the strongest validation BPP's approach has received from an external source. For six months, we have been telling overseas brands the same thing: your AI visibility in China depends on the assets you own, not the content you rent. The stress test has now provided the empirical proof.
The window for building those assets is not indefinite. Doubao has announced it will clear all C-level user-created intelligent agent data by October 15 — a date that likely coincides with another round of source rule adjustments. The platforms are iterating faster than most international marketing calendars can accommodate. The brands that establish their owned, verified, cross-referenced digital identity now will be the ones whose visibility survives the next algorithm change.
The stress test did not reveal anything new about GEO. It revealed what was always true — and proved it with data.