Cases and how we report results
Results recorded in four layers,
not one screenshot
AI mentioning you, getting you right, recommending you and bringing you business are four different things. We record each layer separately: what it is measured with, when, which parts go into acceptance and which we only track together. Every number below states its source and limits, and wherever something can be recomputed, we show how.
01 Seen
AI starts mentioning you accurately
02 Understood
No longer gets you wrong
03 Recommended
On the shortlist
04 Chosen
Brings inquiries, visits, deals
Four layers of results, and how each is measured
Being mentioned by AI is not the same as being recommended; being recommended is not the same as a sale. A screenshot of a ranking only shows one answer, at one moment, for one account and one phrasing.
| Layer | What it means | How we measure it |
|---|---|---|
| 1 · Seen | When buyers ask AI, it starts mentioning your brand and main business accurately | AIV score: whether AI can crawl, understand and cite your website; a fixed set of questions asked every week, recording whether you are mentioned |
| 2 · Understood | Your positioning, products, prices and service scope are no longer described wrongly | In each weekly sample we separately record “what was wrong”, then check which page the AI cited and start fixing from that page |
| 3 · Recommended | When buyers ask “which is best” or “which should I choose”, you are on the shortlist | Mention and citation rates for the same set of questions, looked at as rates over several weeks, not single answers |
| 4 · Chosen | Recommendations bring inquiries, store visits or deals | Forms record which website a visitor came from; when a deal closes, we ask “How did you hear about us?”, with AI assistants as one of the options |
Contract acceptance looks only at the AIV score in the first layer: a baseline before work starts, then re-tests on the same public rubric; a difference of less than 5 points between two runs is not a conclusion. The other three layers we review with you every month, outside acceptance. At the start there is no attribution baseline; write traffic and sales into the contract, and on acceptance day neither side can say clearly what was achieved.
Our own product: AI became the largest source of visitors
This is a product the founder built, not a client case. The on-site AEO work took 4 weeks; a few weeks later, visitors from AI engines grew noticeably. It is a single site, and new model launches and promotions happened in the same period, so we do not report exactly how many weeks it took.
| Metric | Reading | Benchmark |
|---|---|---|
| AI engines' share of annual visitors | 43% | Organic search 20% |
| Visitors from AI engines | 12,238 | Search 5,597, about 2.2× |
| chatgpt.com alone | 3,809 | google.com 3,671 |
| Most-visited page | 11,207 | Homepage 10,277 |
| Social and video, full year | 40 | No paid spend |
| AIV score, re-tested on public rubric v1.1 | 81 | Median of 324 sites: 55 |
Data: Jan–Sep 2026, PostHog channel types, about 28.3K visitors; AIV score re-tested on 2026-09-16. Limits: these are visitors, not sign-ups; 38% of the year’s visits were direct and cannot be attributed; new model launches and promotions happened in the same period. So we do not credit AEO with all of the growth, and we do not use it to promise clients traffic.
Public leaderboard: 324 websites measured
We scored 370 well-known websites with the open-source geo-score, and 324 could be measured. The list, every score and the algorithm are all public, and anyone can rerun them to check.
| Reading | Value | Note |
|---|---|---|
| Median | 55 | The 324 websites that could be scored |
| Capped by a gate | 73 | Can score 40 at most; about a quarter of sites |
| Body text only appears with JavaScript | 41 | AI crawlers get an empty shell |
| AI crawlers blocked in robots.txt | 20 | Some sites choose not to be cited |
| Errors served to crawlers | 12 | Opens in a browser, but not for crawlers |
Measured 2026-10-02, geo-score 1.5.0. See the full leaderboard · See the scoring criteria and code
Our own websites, measured with the same yardstick
The score is the same whoever runs it. You can recompute both scores below yourself with the same command.
| Website | AIV score | When measured |
|---|---|---|
| jianruntech.com | 96 | 2026-10-07，geo-score 1.8.0 |
| citespot.si | 90 | 2026-10-09, launch day, geo-score 1.8.0 |
The AIV score measures readiness: whether AI can crawl, understand, trust and cite the website — not how many times it has been cited. How the rubric is calculated
Client cases
Client cases will appear here once clients have given written consent, stating the background, what was done, which layer the results reached and how they were measured. Without consent, no client names, chat records or business data are shown. Until then, we do not invent cases here, nor borrow other people’s. Students’ reviews are on the GEO Consultant Program page (in Chinese).
Attribution principles
Between an AI recommendation and a sale there are several steps; we only claim the part the evidence can support.
AI search is often one touchpoint on a buyer’s path to a decision. Unless the evidence is strong, we do not credit an entire sale to a single recommendation.
Mentions, recommendations, inquiries and closed deals are recorded separately; a result in one layer is not treated as evidence for the next.
Results depend on industry, competition, product, execution and timing. What one site achieved is never described as “typical”, nor treated as a promise to anyone else.
Company names, personal identities, chat records and business data without full authorization are anonymized or left out.
AI answers change with the model, time, region, account and wording. Every reading states when and under what conditions it was measured.
FAQ
Not listed here? Leave your website and let’s talk.
Why don’t you show screenshots of clients ranking in AI answers?
A screenshot only shows one answer, at one moment, for one account and one phrasing; someone else asking, or asking two days later, may get something different. What we record is rates over several weeks for a fixed set of questions on fixed engines, plus the AIV score before and after the work.
Can your numbers be checked?
Everything that can be recomputed can be checked: the AIV score uses the open-source geo-score and comes out the same whoever runs it; the leaderboard’s list, every score and the raw data are in a public repository. For our own product’s visitor data, the basis and limits are written out above.
Will AEO definitely bring in sales?
Not necessarily, and we do not promise that. AEO raises the chance of being mentioned accurately and recommended by AI; whether buyers end up choosing you also depends on your product, price, reputation and how well you handle the inquiries. So we track sales with you, but they are not part of acceptance.
When will there be client cases?
Once clients agree in writing to make them public. Until then, this page shows only data from our own product, the public leaderboard and our own websites — we do not invent cases, and we do not borrow other people’s.
First, see which layer your site is on
Send us your website. We measure it with the open-source rubric and tell you which layer you are stuck on and what to fix first. Reply within 48 hours; if it is not a fit, we will say so.