Websites often use dynamic rendering to show content to AI bots that do not execute JavaScript. This involves creating a static copy of a page for bots, while regular users see the version rendered in their browser. Testing shows that these copies can contain outdated information, such as old prices or incorrect dates, which remain invisible to the site owner.
Data from Vercel and MERJ in 2024 indicates that most AI crawlers do not execute JavaScript. While JS files accounted for 11.5% of ChatGPT requests and 23.8% of Claude requests, the bots did not process them. Gemini is an exception, as it uses Googlebot infrastructure to perform rendering. Web Almanac 2025 reports a 16–18% median difference in word count on pages before and after JavaScript execution.
Automated testing using Python and Playwright revealed that relying on similarity percentages is insufficient for verifying content accuracy. A page with an outdated price achieved a 97% similarity score compared to the user version, despite the critical data discrepancy. Other issues include the insertion of unauthorized promotional text into bot-facing copies, which AI models might then cite as independent assessments.
To mitigate risks, experts recommend server-side rendering or static site generation over dynamic copies. If copies are necessary, developers should implement automated CI checks to compare specific data points like prices and dates, set strict cache expiration policies, and identify bots via verified IP lists rather than User-Agent strings, which can be easily spoofed.
