Chinese Suppliers Are Still Optimizing Alibaba Keywords. Alibaba Already Moved On.

Something caught my eye this week.

Alibaba has been quietly building out a product called Accio, at accio.com, describing itself as an “AI business agent, partnered with Alibaba.” It supports seven languages: English, German, French, Spanish, Portuguese, Korean, and Japanese. The core promise fits in one sentence: “All tasks in one ask, smart sourcing with AI.”

I spent a while poking around in it. My first reaction wasn’t “Alibaba bolted a chat plugin onto its marketplace.” It was a judgment that made me sit up: Alibaba is using an AI agent to rebuild its own B2B traffic entry point from scratch.

For every supplier still optimizing listing keywords on Alibaba.com, that’s a warning at the entry-point level. I want to cover three things: why Alibaba would disrupt itself, what B2B GEO looks like, and why I think this is the most obvious, least occupied blue ocean available right now.

Why Alibaba is cutting a hole in its own traffic

Start with one detail. Accio runs on the same underlying database as Alibaba.com. The supplier library and the product library are fully shared. But the interaction layer above it is completely independent: separate domain, separate navigation, separate multilingual versions, separate brand.

That means Accio is a parallel entry point, not a patch. You can complete a sourcing run through Accio without ever touching Alibaba.com. The agent handles all the upstream work: interpreting the requirement, screening candidates, comparing, recommending.

No platform company builds a product that could eat its own main traffic casually. When I want to know whether something like this is a real strategic bet or a PR gesture, I look at three signals: does it share the underlying data, is the interaction layer fully independent, and does the resource commitment exceed what an experiment would get. Accio checks all three. It shares the entire Alibaba supplier library, it has its own domain and brand, and it shipped seven language UIs at once. This is not an internal team’s demo. This is Alibaba admitting, out loud, that the B2B traffic model of keyword search plus listing display plus ad-slot bidding is failing.

Failing not because it stopped working, but because agents are destined to eat it. What ChatGPT Shopping and Perplexity Shopping are doing on the consumer side will sooner or later be ported to B2B. “Let AI screen out 99% of my candidate suppliers” is a harder requirement for a buyer placing a cross-border order worth hundreds of thousands of dollars than it is for a consumer buying a $39 T-shirt online.

So Alibaba had exactly two options: let a general-purpose agent like ChatGPT take this entry point, or build one itself. Accio is the “build one itself” answer.

Google AI Overviews did the same thing in 2024 when it replaced the ten blue links with generated summaries. Same logic underneath: better to cannibalize your own ad clicks than let someone else do it.

When a platform starts admitting its old traffic model is failing, everyone who invests in optimizing for that platform should start studying the rules of the new entry point.

When AI screens out 99% of the candidates, what gets you picked

Let me make the scenario concrete.

What does traditional Alibaba.com sourcing look like? The buyer types a keyword, the platform returns 200 listings, the buyer manually screens the top 20, opens each one to check product photos, MOQ, certifications, transaction history, and sends a few inquiries. Throughout that process, the screening power sits with the buyer. The platform only ranks.

What does Accio look like? The buyer doesn’t type keywords. He says one sentence: “Find me 3 suppliers who can produce EU CE-certified portable Bluetooth speakers, MOQ ideally under 500, quick to respond.” The agent returns 3 candidates, each with its matching rationale attached. The buyer doesn’t look at 200 listings. He looks at 3.

It looks like the interface changed. What actually changed is where the screening power lives. It moved from the buyer to the algorithm.

For a supplier, the thing you’re optimizing is no longer human eyeballs. It’s an AI filter.

Those are two different logics.

Optimizing for human eyeballs, you put the flashiest product photo first, stuff keywords into the title, write your selling points as long marketing copy, and buy ads to pump GMV so the algorithm gives you more exposure. I’ve watched Chinese suppliers run that formula for close to a decade. They’re very good at it.

Optimizing for an AI filter works completely differently. I call it B2B GEO: the B2B version of generative engine optimization. I’ve been turning over its core dimensions these past few days, and they come out to roughly these five.

First, semantic density. When the agent parses your listing, it needs to extract the core attributes in one pass: material, certifications, MOQ, lead time, key specs. If that information is scattered across five product images, or buried in the middle of a 2,000-word paragraph, the agent mostly can’t grab it, and you get screened out.

Second, extractable differentiation. Can your selling point be compared by an agent? “We are an industry leader” cannot. “0.3% defect rate against an industry average of 1.5%” can. The first is ad copy written for humans. The second is a structured signal written for the agent.

Third, trust signal density. Certifications, order history, customer cases, response speed. These trust signals traditionally rode on a “Gold Supplier” badge. Now they have to be readable by the agent as structured fields. Badges are for humans. Structured fields are for AI.

Fourth, multilingual alignment. Accio supports seven languages, which means the same product’s descriptions have to be consistent across every language version. If the English version says “500 MOQ” and the German version says “minimum order quantity flexible,” the agent judges the information unreliable.

Fifth, image readability. Accio supports image search, which means product photos aren’t decoration, they’re retrieval objects. Clean background, clear subject, no watermark noise. It used to be “make the buyer think it looks good.” Now it’s “make sure the AI can recognize it.”

There’s a counterintuitive point here worth pausing on. A listing that carried high GMV-based weight on Alibaba may actually be at a disadvantage inside Accio. The agent doesn’t look at historical sales. It looks at semantic clarity. A shop that sells extremely well but writes its product descriptions like prose may end up less visible to the agent than a brand-new shop, as long as the new shop’s listing is structured more cleanly.

The compounding you earned on the old platform doesn’t automatically carry over to the new entry point.

The trust chain just went from two parties to three

Here I need to bring in the analytical framework I’ve been using for years: the Trust Gap.

The framework used to describe a two-party relationship, Chinese supplier versus overseas buyer, measuring the gap between what a brand can actually do and what the customer perceives. I’ve written a pile of content about it, mostly about closing that gap with localization, user reviews, and certification endorsements.

Accio made me realize the framework needs another layer. The trust chain went from two parties to three.

It’s now: buyer, agent, supplier.

There’s an agent in the middle. The buyer has to trust the agent before the agent can screen suppliers on the buyer’s behalf. If the buyer doesn’t trust the agent, it doesn’t matter how accurate the agent is. The buyer goes back to reading 200 listings himself.

So what earns the buyer’s trust in the agent? I kept chewing on this while running Accio these past few days, and I arrived at exactly one answer: in an agent scenario, “being able to explain why it chose what it chose” is itself a trust signal, and its weight may be higher than recommendation accuracy.

Think about your own experience with ChatGPT. If it hands you a conclusion without showing its reasoning, would you make an important decision on that conclusion? Probably not. You’d go back and Google it yourself. B2B sourcing carries far more risk than your personal ChatGPT use. An order starts at hundreds of thousands of dollars, and factory audits, samples, and contracts can each blow up on you.

So if Accio only gives conclusions without reasons, buyers will abandon it after two uses. It has to spell out: why these three suppliers, why the others were excluded, which requirement dimensions were matched, which trust signals were recognized.

For a supplier, your job is to help the agent say the reason to pick you out loud. The certification that used to hide inside a product image, the MOQ buried in a long description, the response speed that lived in chat replies: all of it now has to be readable by the agent as structured fields, so the agent can use it as the stated basis for recommending you.

In the B2B AI scenario, the Trust Gap is no longer just “your brand versus the buyer’s perception.” It’s “your structured trust signals versus the trust evidence the agent can extract.” The framework isn’t overturned. The execution layer underneath it has to be rebuilt.

Why I’m calling this the most obvious blue ocean right now

Consumer GEO has been discussed to death over the past six months, in both the Chinese and English internet. Agencies, courses, and tools are everywhere. Open Twitter on any given day and you’ll see at least three posts on “how to get your brand cited by ChatGPT.” That lane is already crowded.

But I searched around, and GEO in the B2B scenario, meaning getting suppliers selected into the candidate set of a B2B AI sourcing agent, is something almost nobody is discussing systematically. Accio itself has almost no deep teardown in the Chinese-language internet, and the English-language discussion of B2B AI sourcing is scattered to the point of barely existing.

I don’t know whether that’s because nobody has noticed, or because everyone thinks it’s too early. What I do know is that Alibaba has already shipped the product, the multilingual UI is done, and buyers in seven target markets are being steered toward this entry point right now. Between a product launching and suppliers feeling the change, there’s a lag. That lag is the window.

If you’re a B2B supplier selling internationally, the smallest useful move I can think of is this: pick one category you know well, take one real sourcing requirement, and run it through both Accio and Alibaba.com. Compare the top 3 results. Watch who the agent picked, what reason it gave if it didn’t pick you, and which dimensions it used to compare. That’s your own first piece of B2B GEO fieldwork, and it’s worth more than any secondhand content.

I’ve been doing exactly this for the past two days. Once I’ve run enough categories, I’ll write up what I find.

The entry point is shifting, and the signposts on the old map will go stale. What makes this shift unusual is that nobody is disrupting Alibaba from the outside. Alibaba is tearing down its own signposts. So while Chinese suppliers keep optimizing their position on the old map, the map itself may no longer be the one in the buyer’s hands.

Keyword rankings won’t stop working tomorrow. But their marginal value started declining today.

FAQ

What is B2B GEO, and how is it different from regular GEO?

B2B GEO is generative engine optimization aimed at B2B AI sourcing agents (like Alibaba’s Accio), with the goal of getting a supplier selected into the agent’s recommended candidate set. Regular GEO optimizes for “getting your brand cited by consumer AI search like ChatGPT and Perplexity.” B2B GEO optimizes for “getting your product listing screened out of a large candidate pool by a B2B sourcing agent.” The endpoints differ: one is a citation in an answer, the other is a direct recommendation decision.

What is Accio, and how does it relate to Alibaba.com?

Accio launched in late 2024 as Alibaba’s AI sourcing engine and now positions itself at accio.com as an “AI business agent, partnered with Alibaba.” It shares the same supplier library and product library as Alibaba.com, but its interaction layer is fully independent: separate domain, separate navigation, seven supported languages. Buyers don’t type keywords. They describe the requirement in a sentence, and the agent returns 3-5 candidate suppliers.

Will Accio replace Alibaba.com? What should Chinese suppliers do?

It won’t fully replace Alibaba.com in the short term, but Accio is Alibaba proactively admitting that the old traffic model (keyword search + listing display + ad-slot bidding) is failing. Suppliers need to optimize in two places in parallel: keep maintaining traditional SEO on Alibaba.com, and start optimizing listings for AI agents at the same time: structure the key attributes, make trust signals explicit, and align information across languages.

What are the core optimization dimensions of B2B GEO?

Five core dimensions: semantic density (can the agent extract core attributes like MOQ, certifications, and lead time in one pass); extractable differentiation (are selling points structured enough to compare: “0.3% defect rate” beats “industry leading”); trust signal density (can certifications, order history, and response speed be read as structured fields); multilingual alignment (the same product’s information must be consistent across language versions); and image readability (product photos need to support the agent’s visual matching).

Why does the agent’s explainability matter more than its accuracy?

In a high-stakes decision scenario like B2B sourcing, if the agent gives conclusions without reasons, buyers won’t dare place orders based on the recommendation. They’ll fall back to manual screening, and the agent’s practical value disappears. The agent has to spell out why these three suppliers, why the others were excluded, and which requirement dimensions were matched. Which means suppliers have to help the agent articulate the reason to pick them: structure the certifications, MOQ, response speed, and the rest, so the agent has something to explain with.

Adam Yang | 10+ years in China-to-global expansion · Independent consultant 中文版: 中国供应商还在优化 Alibaba 关键词排名时，阿里自己已经不玩关键词了