A Vercel and MERJ crawl study found that none of the major AI crawlers it measured run JavaScript. Gemini is the exception, because it uses Google’s own rendering. The fix is to send your main content as finished HTML from the server.
Key points
- The crawlers for ChatGPT and Claude download JavaScript files but do not run them, so content built in the browser is invisible to them.
- A page can rank well on Google and still be blank to AI answer engines.
- You can test this yourself in two minutes with one command.
- Once the content is visible, pages that lead with the answer, use real numbers and name an author get cited more.
What are SEO, AEO and GEO?
SEO gets your pages ranked in search results. AEO shapes a page so it gives a direct answer to a question. GEO aims to get your brand cited inside answers written by AI tools such as ChatGPT, Gemini and Perplexity.
All three depend on one thing first. The bot has to be able to read your page.
Why can a page rank on Google and still be blank to ChatGPT?
There are two ways to build a page. With client-side rendering, the server sends a nearly empty HTML shell and the visitor’s browser runs JavaScript to fill in the content. With server-side rendering or static generation, the server sends the finished page, so the content is already there before any script runs.
Google runs a separate step that executes JavaScript before it indexes a page, so client-side sites have survived in classic SEO. AI crawlers mostly skip that step. In the month Vercel measured, GPTBot from OpenAI made 569 million requests and ClaudeBot from Anthropic made 370 million. Both downloaded JavaScript files, and neither ran them.
That study measured one month, and crawler behavior can change. So test your own site instead of trusting any single report, including this one.
Which websites are at risk?
Single page apps built with React, Vue or Angular and rendered in the browser are the clearest case. Every product description loaded from an API after the page opens is missing from what the bot receives.
Next.js sites are safe only when the important content is rendered on the server or generated ahead of time. A Next.js page that fetches its main content in the browser has the same problem.
WordPress sends finished HTML by default, which helps. The risk comes from themes, page builders and plugins that pull in pricing tables, reviews or FAQ blocks through JavaScript after the page loads.
For healthcare, fintech, e-commerce and SaaS teams, the content at risk is usually the content that sells. Pricing, plan comparisons, product specs and compliance details are the first things a buyer asks an AI tool about.
How do you test what AI crawlers see?
Run this in a terminal and replace the address with one of your own pages.
curl -s https://yourdomain.com/your-page | grep -i "your headline or price"If the text prints, it is in the HTML the server sends. If nothing prints, a crawler that skips JavaScript will not see it either. Viewing the page source in your browser gives the same answer. The Inspect panel does not, because it shows the page after JavaScript has run.
What else gets a page cited?
Once the bot can read the page, the writing matters. The research points in a few clear directions.
The Princeton GEO study, presented at KDD 2024, found that adding statistics raised visibility in AI answers by up to 40 percent, and adding citations and quotations raised it by up to 41 percent.
SparkToro’s 2026 research found that 44 percent of LLM citations come from the first 30 percent of a page. Put the answer at the top.
A paper posted to arXiv on October 10, 2026 by Mahe Chen and Nick Koudas looked at 158,099 AI answers and 468,037 citations. Within the same website, pages whose titles and paragraphs match the questions people actually ask were cited more often. FAQ blocks, structured data, summary boxes and page speed appeared to help only when comparing one website against another. In plain terms, those items are the entry fee. They do not make one page beat another page on your own site.
Onely found that 76.4 percent of AI-cited content had a named author, so put a real name and bio on every article. AirOps found that about 85 percent of brand mentions in AI search come from third-party pages, so reviews, directories and press coverage count as much as your own site.
These studies come from vendors and researchers with different methods. Use them as direction, and measure your own results.
Does speed still matter?
Yes. Google replaced First Input Delay with Interaction to Next Paint, and a good score is 200 milliseconds or less. Slow pages lose visitors whether they arrive from Google or from an AI answer. Server rendering usually helps here too, because the browser gets real content on the first response instead of waiting for scripts to build it.
A short checklist
- 01Run the curl test on your top ten pages by revenue.
- 02Move pricing, product details and FAQs into server-rendered HTML.
- 03Open each page with a one or two sentence answer to the question it targets.
- 04Add real numbers, named sources and a dated update.
- 05Add an author name and bio to every article.
- 06Check your server logs for GPTBot, ClaudeBot and PerplexityBot to confirm they reach the page.
- 07Track Core Web Vitals, including INP, in Search Console.
Frequently asked questions
Do AI crawlers run JavaScript?
In the Vercel and MERJ study, the crawlers for ChatGPT and Claude did not. Gemini uses Googlebot’s rendering, so it does. Plan for the stricter case and render important content on the server.
Is WordPress good for AI search?
It can be. WordPress sends finished HTML, so content is visible by default. Problems start when plugins load key content through JavaScript. A custom-coded theme with server-rendered content avoids that.
Does schema markup get my page cited?
It helps a site compared with other sites, but the Chen and Koudas paper found no extra lift between pages on the same site. Use it, and do not expect it to do the work of good answers.
Does Google still use JavaScript rendering?
Yes, Google renders JavaScript before indexing. Even so, Google still points to server-side rendering as best practice.
How Dazvix can help
Dazvix is a digital engineering studio. One senior team scopes, builds, tests and measures every project, with no hand-offs. We build custom web apps in React, Next.js and Node.js, and custom WordPress themes and plugins, with server-rendered content, Core Web Vitals work and automated QA built in from the start.
Sources
- Vercel and MERJ, The rise of the AI crawler. Read the study
- TripleDart, Why your site is invisible to ChatGPT and Claude. Read the research
- Chen and Koudas, What drives AI citations, arXiv 2610.15033, October 10, 2026. Read the paper
- WRITER, GEO, AEO and SEO in 2026, which reports the Princeton GEO study, SparkToro, Onely and AirOps figures. Read the article
- Netguru, Web development trends 2026, October 2026. Read the article