Who can AI actually see?
We ran 26 websites, from household names to small businesses and startups, through the Protony GEO Grader, the same tool on our home page. Each site's home page was fetched the way an AI crawler fetches it (server-rendered HTML, no JavaScript) and scored across six categories, 100 points total. Average score: 81. Only 7 earned an A.
| # | Site | Score | Grade | AI /20 | Structured /20 | Content /20 | Discoverability /15 | Authority /15 | Technical /10 | Biggest gap |
|---|---|---|---|---|---|---|---|---|---|---|
| 1 | zendesk.com | 96 | A | 20 | 20 | 20 | 15 | 11 | 10 | No author attribution |
| 2 | atlassian.com | 95 | A | 20 | 20 | 17 | 13 | 15 | 10 | No author attribution |
| 3 | stripe.com | 93 | A | 20 | 20 | 17 | 15 | 11 | 10 | No publication date |
| 4 | wix.com | 92 | A | 20 | 20 | 20 | 15 | 7 | 10 | No author attribution |
| 5 | salesforce.com | 91 | A | 20 | 17 | 20 | 15 | 11 | 8 | No author attribution |
| 6 | hubspot.com | 90 | A | 20 | 20 | 20 | 13 | 7 | 10 | No author attribution |
| 7 | webflow.com | 90 | A | 20 | 20 | 17 | 12 | 11 | 10 | No publication date |
| 8 | asana.com | 86 | B | 20 | 20 | 14 | 15 | 7 | 10 | No publication date |
| 9 | klaviyo.com | 86 | B | 8 | 20 | 20 | 13 | 15 | 10 | AI crawlers blocked |
| 10 | intercom.com | 85 | B | 20 | 20 | 15 | 13 | 7 | 10 | No question headings |
| 11 | notion.so | 84 | B | 20 | 20 | 12 | 15 | 7 | 10 | No question headings |
| 12 | waveapps.com | 82 | B | 16 | 17 | 20 | 12 | 7 | 10 | No llms.txt |
| 13 | zoom.us | 82 | B | 16 | 17 | 15 | 13 | 11 | 10 | No question headings |
| 14 | mailchimp.com | 81 | B | 20 | 17 | 12 | 15 | 7 | 10 | No question headings |
| 15 | squarespace.com | 80 | B | 8 | 20 | 20 | 15 | 7 | 10 | AI crawlers blocked |
| 16 | workday.com | 79 | C | 16 | 20 | 15 | 15 | 7 | 6 | No question headings |
| 17 | nike.com | 78 | C | 16 | 15 | 15 | 15 | 7 | 10 | No question headings |
| 18 | slack.com | 77 | C | 20 | 6 | 17 | 13 | 11 | 10 | No JSON-LD schema |
| 19 | dropbox.com | 75 | C | 20 | 14 | 12 | 12 | 7 | 10 | No social metadata |
| 20 | hellobonsai.com | 74 | C | 20 | 17 | 12 | 12 | 3 | 10 | No question headings |
| 21 | sendfox.com | 71 | C | 20 | 6 | 17 | 15 | 3 | 10 | No JSON-LD schema |
| 22 | shopify.com | 71 | C | 20 | 6 | 17 | 15 | 3 | 10 | No JSON-LD schema |
| 23 | jobber.com | 70 | C | 20 | 6 | 15 | 12 | 7 | 10 | No JSON-LD schema |
| 24 | monday.com | 66 | D | 8 | 20 | 9 | 12 | 7 | 10 | AI crawlers blocked |
| 25 | gumroad.com | 64 | D | 20 | 3 | 9 | 15 | 7 | 10 | No JSON-LD schema |
| 26 | carrd.co | 57 | F | 16 | 0 | 14 | 10 | 7 | 10 | No JSON-LD schema |
Columns: AI crawler access /20, Structured data /20, Content structure /20, Discoverability /15, Authority signals /15, Technical health /10. Scores reflect the public home page on 2026-11-03 and change as sites change. Want your company added or re-scored? Ask us.
What the index says
Size does not predict the score. nike.com (78) sits four points above hellobonsai.com (74), a tool for freelancers. slack.com and dropbox.com, with engineering teams in the thousands, land in the mid-70s next to two-person SaaS companies. The spread among big brands, from monday.com at 66 to zendesk.com at 96, is wider than any gap between big and small. We wrote up why in GEO Is Moneyball.
Blocking crawlers is the most expensive mistake. monday.com, squarespace.com and klaviyo.com all block at least one AI crawler and lose 12 points for it, the single largest penalty in the grader, because nothing else on the page matters if the engine cannot read it. That one rule is why monday.com, a public company, grades a D.
Small sites lose points on schema, not on being small. carrd.co (57), gumroad.com (64), jobber.com and sendfox.com all publish no JSON-LD schema. Their pages read fine to a human and say almost nothing to a machine. Add the schema and every one of them passes several household names on this list.
The leaders have boring fundamentals. zendesk.com, atlassian.com and stripe.com do nothing exotic. Full crawler access, complete schema, one H1, question-style headings, clean metadata. That is the whole playbook, and almost nobody runs all of it.
Everyone is missing the same easy points. No publication date and no named author show up in the top gap for a third of the index. These are afternoon fixes.
Method
Each home page was fetched once with a crawler user agent and no JavaScript execution, which is what most AI retrieval systems see. robots.txt and llms.txt were checked at the domain root. Scoring rules are published in What is GEO. Two brands we attempted (godaddy.com, servicenow.com) returned HTTP 403 to the crawler and were excluded rather than scored zero. The index is free to cite with a link to this page.