Hey everyone,

Over the last month, I’ve been analyzing access logs and crawler behavior specifically looking at how AI search engines (OpenAI's GPTBot, Anthropic's ClaudeBot, and PerplexityBot) interact with WooCommerce product catalogs.

Since LLM-driven shopping and search is picking up, a few observations stood out:

1. Heavy HTML vs Markdown (Crawl Efficiency)
Product pages with heavy builders (Elementor/Divi) and 60+ scripts get crawled noticeably slower. When providing a clean, token-efficient /llms.txt or markdown feed, the bot hits almost the entire catalog in a single pass without getting choked by server limits.

2. Perplexity relies heavily on FAQ schemas
For product recommendation queries ("best X for Y under $Z"), Perplexity seems to prioritize structured FAQ and product metadata (JSON-LD) over standard paragraph descriptions.

3. Bot Spoofing is real
A surprisingly high number of scraper bots pretend to have user-agents like GPTBot or ClaudeBot to bypass standard blocks. If you're analyzing this, running Forward-Confirmed Reverse DNS (FCrDNS) checks on their IPs is mandatory, otherwise half your "AI crawler" data is just random scraping bots.

4. Revenue is starting to trickle in
By passing native UTM tags into order attribution (utm_source=chatgpt / perplexity), we saw about 8 orders ($800+) across a couple of stores that originated directly from AI chat sessions.

Are any of you currently preparing your store architectures for generative engine optimization (GEO), or are you treating it as standard SEO for now? Curious to hear how other store owners/devs are approaching this.

Edit: Thanks for the thoughtful feedback. To be clear, these were small-sample observations from limited testing, not controlled results, and shouldn't be read as proof that llms.txt or FAQ schema improves AI visibility. Disclosure: I'm the developer of a WordPress plugin in this space (Zoventic GEO). Method and sample sizes to follow if I run a proper holdout test.