JIANRUNTECH / GEO-SCORE
324 well-known sites, scored on whether an AI answer engine can find, parse, trust and cite them. One open rubric, one public tool, up to eight pages per site: 66 sites were scored on fewer than four pages and are marked thin sample. Nothing here is a black box — re-run it yourself.
Source: the geo-score benchmark, measured with geo-score 1.5.0 on 2026-10-02; data: results.json at tag v1.8.3.
73 sites have a gate check at zero — a retrieval crawler cannot get the content, so nothing else on the page matters. Three very different problems live inside that number.
Their robots.txt disallows AI retrieval crawlers by name. A site that does not want to be quoted is not misconfigured, and this rubric reports it as it is.aftership.com, amazon.com, arstechnica.com, drugs.com, ebay.com, elegoo.com, figma.com, healthline.com, jdcloud.com, june.so, kuehne-nagel.com, lexisnexis.com, mercadolibre.com, notion.com, phoronix.com, qcloud.com, techcrunch.com, theregister.com, theverge.com, wired.com
The server refuses crawler user-agents. The probes carry the crawlers' user-agent strings but come from the auditor's own address, unverified, so a server that checks a crawler by its IP address may refuse them and still serve the real crawler. From outside, a policy and a bot-protection default look the same.airbnb.com, bambulab.com, clubfactory.com, cma-cgm.com, deepseek.com, hashicorp.com, hims.com, lenovo.com, lwn.net, marriott.com, shopline.com, stratechery.com
Body copy only exists after JavaScript runs. The content is there and a browser sees it; a retrieval crawler gets an empty shell. This group almost certainly did not choose it. 30 of them were scored on fewer than 4 pages.01.ai, 4px.com, agoda.com, airtable.com, aliexpress.com, baichuan-ai.com, bigmodel.cn, clevelandclinic.org, cohere.com, continue.dev, datahawk.co, dingtalk.com, drizzle.team, duolingo.com, fanruan.com, feishu.cn, freecodecamp.org, hioscar.com, jushuitan.com, keepa.com, khanacademy.org, lianlianpay.com, megvii.com, narwal.com, nostarch.com, oceanwing.com, payoneer.cn, rakuten.com, sellersprite.com, sf-express.com, shein.com, shopee.com, skyscanner.net, solidjs.com, stepfun.com, trendyol.com, walmart.com, weaviate.io, wps.cn, xiaomi.com, yuque.com
Bars are counts of sites per ten-point band. The rubric is not graded on a curve; this is the distribution that falls out of it.
No install, no dependencies, no account: Python 3.8+ is all it needs.
curl -sL https://raw.githubusercontent.com/jianruntech/geo-score/v1.8.3/cli/geo_score.py \ | python3 - yoursite.com
Add --explain for the evidence behind every check and what the next tier asks for, or --compare competitor.com to see the two side by side.
Click a column to sort. The score is readiness — whether an engine can cite the site, not whether one does.
| SITE | SCORE | BAND | SECTOR | BIGGEST GAPS |
|---|
| SECTOR | SITES | MEDIAN | |
|---|---|---|---|
| Cloud & DevOps | 25 | 77 | |
| Payments / Fintech | 12 | 69 | |
| AI | 29 | 67 | |
| Marketing / CRM | 17 | 66 | |
| Developer tools | 38 | 60 | |
| SaaS / Productivity | 22 | 58 | |
| Education | 14 | 51 | |
| Healthcare | 13 | 46 | |
| China / AI | 13 | 40 | |
| China / consumer | 23 | 40 | |
| Cross-border services | 21 | 40 | |
| Logistics / Freight | 11 | 40 | |
| Media / Publishing | 13 | 40 | |
| China / enterprise | 10 | 38 | |
| Manufacturing / Industrial | 11 | 38 | |
| China / no-code | 11 | 35 | |
| Ecommerce | 15 | 32 | |
| Professional services | 8 | — | n < 10 |
| Travel | 9 | — | n < 10 |
| China / cloud | 9 | — | n < 10 |