When customers ask AI,
make sure it describes you accurately

No. Publishing articles is at most one of the actions. Complete AEO includes taking a baseline, unifying the brand’s facts, finding the questions buyers ask, writing pages for those questions that can be cited, making the website crawlable and readable for AI, and adding credible third-party sources, then re-testing every week and fixing errors. Publish articles without changing any of this, and what AI reads is still scattered and contradictory.

There are three cases. One, gate problems — crawlers blocked, pages that are only scripts — are a few days of technical work; rerun the rubric the same day and you can confirm whether the gate is cleared. When AI engines come back to recrawl is up to them, not us. Two, a page answering one question AI is asked often has to wait to be indexed and crawled; it takes a few weeks before you can tell whether AI cites it. Three, competing through search rankings for a category’s biggest keyword is measured in months, often six months to a year. Acceptance is written only in terms of the AIV baseline and the lift. We track traffic, leads and inquiries with you, but they are not acceptance criteria — there is no attribution baseline at the start, and writing business metrics into acceptance is irresponsible for both sides. Nor do we promise rankings or vanity traffic.

Yes. Doubao, DeepSeek, Tencent Yuanbao, Kimi and ERNIE Bot each have their own source habits: some fetch web pages directly, others rely mainly on content from their own platforms. Based on which ones you want to cover, we decide where content is published and how the website changes. Among the Chinese assistants, those that can be tested through an API are tested that way; the rest are sampled by hand.

No, and nobody can. AI answers change with the model, the time, the region and the wording; ask the same question twice and you may get two different answers. The contract states the deliverables and the score measured on the open-source rubric, from where it starts to where it should reach; recommendations and rankings we review with you every week, but they are not written into the contract.

Yes, but the job is different: accuracy. When AI gets your prices, terms, credentials or scope of service wrong, it is misleading your customers on your behalf, and they will not tell you. Every week we ask each AI engine the same fixed questions, log what it gets wrong, trace each error to its source and fix it there until the AI gets it right. The dashboard shows, week by week, how many errors were found and how many are fixed.

They're half right. Google's own documentation does say that SEO practices still apply to its AI search results, with no special files or extra markup required. But that document governs Google alone. In the full-year data from our founder's own product, chatgpt.com as a single referrer already sends more visitors than google.com — the engine that matters most to exporters isn't covered by that document. The other half: whether you call it GEO or SEO is a naming question; the work is the same work. We tested 324 well-known websites, and AI crawlers can't read about a quarter of them — crawlers blocked, script-only pages, outright errors. Sites aimed at the Chinese market score a median of 36, 22 points below the rest. Those sites need fixing whatever you call it. So we don't argue about names; we argue with things you can check: an open-source rubric, a public 324-site leaderboard, a rule that any re-run within ±5 points is noise, and a promise of measurable readiness rather than traffic. If you doubt GEO, the fastest test is to run the open rubric on your own site — the score comes out of your machine, not ours.

Correct: your own claims are not evidence. In the rubric, brand credibility is worth 18 points, and 12 of them can only come from third parties: independent mentions (3 points for independent coverage or a review, 4 for being cited consistently in several places), third-party listings (full marks only at five or more), and a knowledge-graph entry (points only if a public knowledge base such as Wikidata has an entry for you; otherwise zero). More important still: when the machine cannot run the two checks that need off-site lookups, they are not counted in the total. A tool that wanted to flatter itself would mark them as passed; we take them out of the total instead, and you can run it yourself to verify. So the real value of a diagnostic is separating “crawlers cannot read you at all” from “they can read you but nobody cites you” — the first is a few days of technical work, the second takes months, and the fixes are completely different.

Because the rubric produces numbers, not judgment. The score you get from running it yourself is the same one we get — that is the whole point of open-sourcing it: every point we give you can be checked. The snapshot adds three things the rubric cannot do. One, reading the result: a rerun drifts by ±5 naturally, and gate checks flip with a site's bot-blocking policy, so someone has to tell real gaps from noise. Two, putting it back into your business: which of the nine stages to fix first — the rubric has no idea what you sell. Three, saying whether a fix is worth it and how much work it takes — the scoring is open, the fixing is our job. If you can run a command, run it. If you cannot, or you ran it and do not know what to do next, ask us. No charge.