Contents 4
Then a graph written by a script after the response is a graph those systems never see. That sentence is about arrival. It is not about citation, rank or revenue.
What was measured?
On 17 December 2024 Vercel published, with MERJ, a study of crawler traffic on its network. GPTBot alone had produced 569 million requests in the preceding month. Combined OpenAI and Anthropic traffic was described as about 20 percent of Googlebot volume on that network in the window.
The finding that matters here: none of the major AI crawlers currently render JavaScript. Named: OpenAI (GPTBot, OAI-SearchBot, ChatGPT-User), Anthropic (ClaudeBot), Meta (Meta-ExternalAgent), ByteDance (Bytespider), Perplexity (PerplexityBot). Common Crawl also does not render. ChatGPT and Claude fetched JavaScript files – about 11.5 percent and 23.8 percent of requests – and gave no evidence those files were executed.
As of 21 September 2026 no public re-measurement reversing that finding was located. Absence of a contradiction is not proof that the behaviour is permanent. It is a reason not to design as if the opposite were already true.
How does that split the clients?
| Client | Reads delivered HTML | Waits for scripts |
| --- | --- | --- |
| Googlebot / WRS | Yes, first | Yes, on a later rendering pass |
| Applebot | Yes | JavaScript execution has been measured |
| GPTBot, OAI-SearchBot, ChatGPT-User | Yes | No execution measured, December 2024 |
| ClaudeBot | Yes | No execution measured, December 2024 |
| PerplexityBot | Yes | No execution measured, December 2024 |
| A rendering validator | Yes | Yes – that is why it disagrees with view-source |
What follows?
- JSON-LD inserted only after load is invisible to those fetchers. The Rich Results Test can still be green.
- A page whose visible copy is server-rendered but whose graph is client-rendered is a page those fetchers can quote as text and still miss as an entity.
- A warm browser hides the split. A cold fetch – cache deleted, or a fetcher that never had a cache – does not.
- Write the ld+json block into the HTML the server sends.
What does not follow?
- Schema in the delivered HTML will be cited. Arrival is not use. Use is not citation.
- Google will stop rendering. Googlebot remains a different client.
- Every chat answer was produced by that product’s published crawler hitting your URL on that turn.
- A site without schema is unreadable. The visible HTML is still text.
- A claim made for markup becomes documented.
Next in this set Rich Results · Entity Recognition · Indexing Speed
Elsewhere on this site Entity grounding · Semantic search · Named entity recognition · Client-side rendering check · Large language models · Answer engine optimization · Generative engine optimization · Evidence · E-E-A-T