Product research, listings, creative, ads, customer service, supply chain, content marketing, AI search visibility, team workflow — nine stages in one chain. Wherever you are stuck, that is where we start. Everything begins with a free snapshot.
The nine stages are written for cross-border e-commerce. One of them, AI search visibility, applies to any company that needs to be found by overseas customers — manufacturers, brands and export service firms can start there.
Cross-border e-commerce is not one job; it is nine jobs chained together. In every stage someone has used AI to make the work lighter, and in every stage someone has bought a tool and never used it. Here is what we can deliver: nine stages, each one scoped, commissioned and signed off on its own. These are not nine chapters in a slide deck. You do not need all nine. Start with the one that pinches hardest.
Today whoever owns new products spends every morning scrolling best-seller lists, checking competitors, pulling data and talking to suppliers. Decisions run on gut feel and scattered screenshots, and the thread breaks whenever someone new takes over.
With AI the scrolling goes to a system: scheduled pulls of category best-sellers and trends, price and listing monitoring on key SKUs, and candidate reports that take your own supply-chain constraints into account. It runs on its own and only flags exceptions to a person — people move from working the screen to working with suppliers.
Today listings are rewritten per storefront, outsourced or machine-translated and pushed live, keywords are done separately for each site, titles are edited one by one in the back office, and nobody knows whether the edit helped.
With AI we first turn human know-how into business fields a machine can read (category, selling points, target market, promotion, content angle), then chain prompts, templates and validation into a pipeline. Trending search terms are matched against existing titles to surface missing keywords. This is a process change: output that can be batched, rolled back and compared.
Today product images depend on freelancers or the design queue. One new product needs white-background, lifestyle, dimension and A+ images, each through several rounds, and in peak season there is never enough creative.
With AI do not stop at choosing an image model. Plug generation into your existing product and asset flow so every image lands in a company asset library instead of on someone's laptop. Consistency is written into acceptance: person, clothing and scene all have to match, plus fidelity and commercial usability. How tight that bar is depends on where the image runs — a feed ad that lives a week can be looser; a product detail page cannot.
Today ads are run by watching: read the report, pause the weak ad groups, add budget to the good ones, change bids — two or three hours a day, and nothing happens at night or on weekends.
With AI watching becomes rules and alerts: weak performers pause automatically or go to a person, strong ones get budget, bids move when rank moves, and a daily ad report writes itself. The next step is wiring two stages together — when a SKU's stock drops below a threshold the system throttles or pauses its ads, so inventory starts feeding back into ad decisions.
Today support is staffed by time zone, non-English storefronts limp along on machine translation, bad reviews and returns are translated by hand, and feedback takes forever to reach product and sourcing.
With AI we do not build a chatbot; we build layers. The business-process layer fixes the path (detect language → inquiry → collect key facts → build a case card → hand to a person) and never lets the model improvise; the model only extracts information, keeps the language consistent, recovers context and catches exceptions. Then conversion is split by language — the market with the most traffic usually turns out to have the worst lead quality, and without that view the budget stays misallocated.
Today restocking runs on past sales plus gut feel — stock up before peak season, clear it if it does not move; freight options and duties are worked out by hand, and customs paperwork keeps getting sent back.
With AI we do not reinvent your inventory logic. Your restocking rules stay exactly as they are; the system applies them to produce the SKUs to reorder, with timing and quantity, and people move from calculating to approving. Shipping documents and photos are recognized and pre-filled first, with a person doing the final check — the human sign-off stays.
Today everyone knows short video, creators and store content should be happening. In practice a batch gets made whenever the boss pushes, with no steady output and nothing that accumulates.
With AI content is a production line, not inspiration: product selling points and creator content attributes drive a first screen that gives business development a sharper shortlist; then creator traits, selling points and proven hits are combined into a few shooting scripts the creator can choose from — useful to the creator, not just efficient for us. Follow-up becomes system-assigned: who has not received samples, whose deadline is close — automatic reminders first, a person only if nothing moves.
Today the website and store are built for classic SEO. Whether AI search cites you, nobody checks — and nobody knows how to. The classic SEO toolkit is close to useless against the new rules.
With AI we start with an access map: replay your access rules the way crawlers actually apply them and see who gets in and who is blocked (AI crawlers come in three kinds — the ones that feed answers, the ones that maintain the ecosystem, and the ones that only collect training data — and the first must not be blocked along with the rest). Then we take a 0–100 baseline and fix upward from the “can you be found at all” layer; a month later we compare the score check by check. The rubric is open on GitHub; the script has no dependencies and runs in one command, so you can check every point we give you.
The rubric is called geo-score — what it measures, why three gates cap the score, and the results across 314 well-known sites are all on that page. The method is public, so every point we give you can be verified.
Today there is a pile of tools that only a few people can use; the prompts and workflows people build live on personal laptops and leave with them; and the boss cannot tell whether to keep investing.
With AI map the process first, build the agent second. The other way around just stuffs a new tool into an old process. Start with AI literacy for everyone so the team shares a baseline, then have the people who do the work spell out how it is really done, where the exceptions are and when a human must decide — if that cannot be dug out, anything built on top is hollow. Only then do prompts, task cards and test sets go into the knowledge base, and acceptance is by test set, not by how lively the demo looked.
Not every stage deserves AI first. To decide whether something should be done now, check three things:
How to use it Three out of three: do it. Two: design it as human-plus-AI with a manual checkpoint. One or none: leave it for now, even if the boss asked for it by name. During the diagnostic we run all nine stages through this filter with you.
No. Most companies start with one or two stages and expand once they work. We would rather you start with the one that hurts every day and can be measured — once it works, the team will ask for the next one on its own.
Not recommended as a first project: purchasing and production scheduling. Their friction sits outside the company (suppliers) or in compliance, where you cannot change much; scheduling also touches equipment, process, inventory, delivery dates and accountability — high risk, wrong place to start.
One of the nine stages, taken on its own and in depth. Overseas buyers increasingly ask AI “who can make this” — our AEO method (answer engine optimization — what some markets call GEO) turns AI visibility into a single number, the AIV score, and then raises it. The rubric is open source, so anyone can check how we score.
The rubric is called geo-score, MIT-licensed. That page explains what it measures, why three gates cap the score outright, and the results across 314 well-known sites. The method is public: every point we give you can be verified.
The five weights add up to 100. Citable content alone is worth 35, nearly as much as the other four together — whether AI is willing to cite you matters more than whether you have a page.
| Metric | Reading | Benchmark | Basis |
|---|---|---|---|
| AI engines' share of annual visitors | 43% | Organic search 20% | PostHog channel type, 28.3K visitors |
| Visitors from AI engines | 12,238 | Search 5,597 | 2.2× search |
| chatgpt.com as a single referrer | 3,809 | google.com 3,671 | Ahead of Google |
| Carrier page · just one | 11,207 | Homepage 10,277 | #1 on the site |
| Social + video · full year | 40 | — | No paid spend |
| AIV score · public rubric v1.1 | 81 | Median of 314 sites: 56 | Re-scored 2026-09-16 |
NoteFull-year data from the founder's own product, not a client case. Counted in visitors, not sign-ups; 38% direct traffic stays unattributed, so we don't credit AEO with all of the growth. Whether an AI engine cites you depends on competition and query intent, which no external audit can see. We promise measurable readiness, not traffic.
The diagnostic fee is credited toward the first month of ongoing work. Scenario packs stack and can be repurchased; the delivery system includes everything in the packs plus the AIV dashboard. Prices are converted from CNY at the current exchange rate and rounded. Annual retainers are scoped to your stages and your scale, and quoted separately.
Different overseas customers find you in different ways, and the offer has to change with them. We match the scenario pack and the delivery terms to how you actually sell abroad.
What we track together E-commerce: traffic · leads · conversion rate · return rate | Deep tech: inquiries · pipeline · closed deals (no vanity traffic). Acceptance is by AIV score; business metrics are tracked but are not acceptance criteria.
Crawlable, understandable, citable content, brand credibility, answer fit — the five dimensions of the AIV score. We can score any site 0–100, say where the gaps are and write the fix plan. The diagnostic delivers exactly this baseline report. Rubric v1.1, 21 tiered checks, open source on GitHub; the full explanation is on the geo-score page.
Overseas acquisition systems and deep-tech inquiry systems, quoted and contracted per project, monthly maintenance included.
Companies are not short of AI tools. They are short of a way to use them in real work. That second part is what we teach.
A public session for several companies at once, no charge. We show what has actually worked for us; decide afterwards whether to take it further.
Gets the leadership team on the same page, runs an AIV diagnostic live, and produces your company's 90-day roadmap and a list of pitfalls to avoid.
Includes pre-course survey · 3–4 hours in the room · course handbook and prompt templates · 30 days of online follow-up
Tailored by role. Everyone brings a real task and leaves with their own finished output and a task card.
Includes staff interviews and scenario mapping · 6–7 hours of hands-on practice · one task card per person · scenario list + test-set acceptance · 30 days of follow-up
Deep work on a single role, producing a full set of task cards and a prompt library for that role.
Monthly refreshers, designs for new scenarios, and developing your own internal AI lead — so the capability stays inside the company rather than with the instructor.
All four become trainable SOPs and task cards, delivered by our own team, never outsourced.
Seven formats, from a free session to an annual retainer. They line up as one path: most clients come in through an open session or a diagnostic, get one stage working, then decide whether to go further. We will not quote you the most expensive tier at the first meeting.
No charge. One session that lets you see how we think and how we work before deciding whether to talk further.
All nine stages go through the three tests, and you get a baseline report: which stage to fix first, where the gap is today, and where you can expect to land after the fix. On the AEO line the diagnostic produces the AIV score and a 30-day list, and the fee is credited toward the first month of ongoing work.
Half a day to a full day, using existing course modules lightly adapted to your industry. Good for getting everyone to the same starting level, or getting one role moving.
Built around your stages and roles, practiced on your own material. Everyone brings a real task and leaves with their own finished output and a task card.
Two days to finish one concrete thing with your team: rewrite the listings, stand up an asset factory, get ad monitoring running. Produces a full set of task cards and a role prompt library.
After training, when a stage deserves to become a system that runs every day, we build it. Quoted and contracted per project, never bundled into training.
Paid annually, the way a company keeps a law firm on retainer. Ongoing judgment on scenarios, tool updates, support for landing new stages, and a quarterly review and reprioritization.
Cross-border sellers do the math and shop around — nobody in this trade makes money off what the buyer doesn't know. So the prices are on the page: the same tier costs the same for everyone, and if the budget is tight you pick a lower tier instead of negotiating back and forth.
Accordingly, acceptance has to be measurable: on the AEO line the AIV baseline and lift; on the training line a test set plus usage rate, finished output and time saved at 30 days. We only promise what can be measured — traffic, inquiries and revenue we track with you, but they are never acceptance criteria.
Generic training is everywhere. What is scarce is what stays inside the company — three deliverables that are still in use on the job after the course ends.
When Monday morning, before the weekly report Input Competitor list + last week's price sheet (redacted) Check Sample 3 rows against raw back-office data Stop Price gap > 20% → escalate to the ops lead
A Competitor monitoring report works per person → can become a system A First drafts of support emails works per person B Product trend summary knowledge needs organizing C Supply-chain forecasting not ready yet
Method post-course output scored against a test set Metrics adoption rate / finished output / time saved Backup the client may keep a second set of test questions Why a demo can be rehearsed; a test set cannot
| Dimension | Typical AI literacy training | The Jianrun approach |
|---|---|---|
| Course content | The instructor's own workflow, one case for everyone, hard to transfer across backgrounds | Tailored by role, practiced on the client's own material; generic modules are only a fifth of it |
| What trainees do | The instructor operates for two hours; staff watch | Bring a laptop and a real task; complete input, generation, checking and delivery on their own |
| What remains afterwards | A slide deck and a few prompts, quickly forgotten | Task cards, a scenario list and a handbook, reusable the moment they are back at their desk |
| Acceptance | A live demo judged on surface impressions | Scored against a test set, with agreed adoption and output quality; no profit promises |
| Afterwards | The course ends and so does the relationship | The scenario list accumulates; refreshers or custom delivery as needed |
Jianrun is just starting out, and we have no public client cases yet — the numbers above come from a product our founder built, not from client wins. So we are opening the first 5 seats: the list price stays untouched, and you cover it in full with four things instead of cash.
You pay with a case, not with money. This is not a discount: the price and the delivery standard stay the same; only the currency changes. A one-page agreement, responsibilities in writing, both sides committed.
First 5 seats · 5 left
We did the work before we taught it: eleven years inside multinational cross-border e-commerce companies, and two overseas products built by hand — not a decade of talking about AI.
Amazon 5 yrs · Shopee 3 yrs · Coupang 3 yrs, engineer → senior technical expert → technical director.
Launched two overseas products in two months; one passed 10,000 sign-ups — that is the project the AEO method came out of.
University of Electronic Science and Technology of China (alumnus and entrepreneurship mentor) · master's degree, Georgia Institute of Technology · Certified AI Trainer (Advanced) · Shopify Partner · AWS and Google Cloud AI certifications · Google Speaker · digital marketing advisor to several well-known companies
Content operations and design at ZTE, Huawei and other large companies. Now runs AI video editing and social content, and builds the AI image workflows and the content factory hands-on.
Delivered in-house, never outsourced: AI imagery and the content factory are the two things cross-border clients ask about most, and our own team runs both. Faster to change, cheaper to run.
Prices are public and set by tier; the numbers on this page are the quote.
Measured by AIV score. Start with a diagnostic report; the fee is credited toward the first month of ongoing work.
See the product ladderGets the leadership team on the same page, runs an AIV diagnostic live, and produces a 90-day roadmap.
Ask about datesEveryone brings a real task and leaves with their own finished output and a task card.
Free snapshotCourse prices are ranges: existing modules sit at the bottom, deep customization to your roles and material at the top, with small variations by location. Larger custom delivery is quoted per project at US$14,900–74,300. Prices are converted from CNY at the current exchange rate and rounded.
On a tight budget, narrow the scope first — one stage, or the half-day briefing. If we are not the right fit, we will refer you to an instructor who is.
Not listed here? One conversation usually covers it.
Run it through three tests: does it happen at least five times a week, can the input be structured, and can the output be judged objectively? Three out of three: just do it. Two: design it as human-plus-AI with a manual check. One or none: leave it for now. When several stages pass all three, pick the one closest to revenue — product research, listings, ads and content marketing are close; customer service, supply chain and workflow are further away. Get one stage working and the team will ask for the next. Not sure? Start with a free snapshot and we will walk all nine stages with you.
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.
No, which is why every session opens with a skills check. The same room may hold an engineer who writes code with Claude and an admin who has never opened ChatGPT. The rule is to get everyone to the same starting level first, then go deeper; hands-on work is done by role, with different tasks for each.
Before the session we confirm accounts, internet access and the data scope. Where sensitive data is involved we use redacted material or a test environment, and the do-not-upload list is written into the plan. Government and state-owned enterprise clients can run on a domestic Chinese model. We never cross a security boundary for the sake of a good demo.
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 314 well-known websites, and a quarter of them can't be cited by AI at all — crawlers blocked, script-only pages, outright errors. Sites aimed at the Chinese market score a median of 40, 19 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 314-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.
Training is the core business. When a proven scenario deserves to become a running system — a knowledge base, a workflow, an agent, or an overseas acquisition system — we build it as custom delivery, quoted and contracted separately per project.
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.
Send us your store or website. We run the open AIV rubric, then do the part the rubric cannot: which gaps are real for your business and which you can ignore, which stage to fix first, and whether it is worth fixing at all. No website yet? Just tell us which task in the last three months made you feel short-staffed, and we reply with a first read of the nine stages. Reply within 48 hours; if we are not the right fit, we say so.
example.com AIV 46 / 100 · Getting started Three biggest gaps +9 opening passage cannot stand alone +6 no freshness signal +5 no llms.txt Start with: AI search visibility Verdict: all three are a few days of work — worth fixing first