AI Fetches Blindly Strip JSON-LD Schema. Perplexity Kills Content Older Than Six Months.

Aggressive payload stripping deployed by AI bots blinds them to standard structured data, reveals Martin Stepanek.

Because JavaScript-heavy frameworks yield a text-to-HTML ratio as low as 0.05%, crawlers for ChatGPT and Claude ruthlessly clean the code to minimize live-fetch computing costs.

They delete all <script> tags by default — which instantly wipes out JSON-LD schema.

In an 8-layer price-hiding test, Claude failed to extract a single product price (0 out of 8) because it purged the foundational markup before processing.

JavaScript-reliant rendering forces LLMs to hallucinate critical data during batch queries.

When a user prompts ChatGPT for multiple products from target.com, the query triggers a live fetch instead of relying on Google's index.

Because the prices exist only in the JS payload, the bot either returns a "price unavailable" error or completely fabricates the numbers.

A $1,300 gaming laptop appeared as $2,100, with the bot fatally confusing an $800 discount flag.

Separately, temporal decay algorithms and server latency execute a hard purge on visibility.

Unlike Google, which ranks 20-year-old pages, AI systems enforce strict freshness thresholds.

Around 65% of AI citations link to content published within the last year, and Perplexity permanently wipes out articles older than six months.

Furthermore, slow server response times trigger a 499 client closed request in Nginx logs; if the live fetch delays the user's real-time answer, the bot simply abandons the connection.