On July 22nd I flipped a switch on my marketplace platform: a counter that logs every visit from every AI crawler, on every store, every day.
I expected to wait a month before there was anything worth writing about. It took a week.
604 AI bot visits across 13 marketplaces. One AI crawler reading one of these stores roughly every 17 minutes, around the clock. Eleven different bots - OpenAI, Anthropic, Amazon, Meta, Perplexity, even TikTok's crawler showed up. The store owners saw none of it. Not because they weren't paying attention, but because these visitors are invisible: AI crawlers don't run JavaScript, so they never show up in Google Analytics, Mixpanel, or anything else a normal site owner looks at. They only exist in the server log, and almost nobody reads their server logs.
If you follow this space at all, you've seen the wave of advice on how to rank in AI search. Almost none of it comes with logs - not because anyone is lying, but because hardly anyone giving the advice can see these bots either. I run the server layer under many stores at once, which means I could actually look. Here's what I found, and what I'd do about it if I ran a marketplace.
What surprised me
First: it's not just ChatGPT. GPTBot came 46 times in the week. Anthropic's ClaudeBot came 168 - almost four times more. Then Amazon's bot, then Meta's. To be clear, that doesn't mean you should optimize for Claude instead (and how would you, anyway?): where the crawlers read and where the buyers ask are two different things, and nobody outside the AI companies can see the second one. What it means is simpler: your listings are feeding the whole AI industry, so as a marketplace owner, let them all in. Eleven different bots showed up in one week, and any one of them can end up putting your store in front of a buyer. Check your robots.txt and your CDN settings - some setups block AI crawlers by default, and you'd never notice.
Data: 13 marketplaces on the Prometora platform, Jul 22-28, 2026. Full methodology and the downloadable dataset in the research write-up.
Second, and this is the one that changed my mind about something I'd said publicly: I was wrong about what gets read. I assumed AI crawlers would favor well-written listings - proper descriptions, good titles. So I compared what they read against what they skipped, inside the same stores. Description quality predicted nothing. Age predicted everything. Listings under a week old: 85% got read. Listings older than three months: 17%. The bots read what is new and what is linked, and a dormant back catalog quietly disappears from AI's view of the world.
616 listings, grouped by age at the time of logging.
Third: the spread between stores was brutal. Same platform, same week - two stores had 100% of their catalog read, five stores had zero. Five of the thirteen only ever got homepage visits. The bots knocked on the front door over and over and never stepped inside. Whatever else you do, if your listings aren't linked and reachable from where a crawler enters, nothing else matters.
What I'd do as an operator
Stop treating AI search as a future thing. The crawlers found these stores the first day the counter ran. The question isn't whether AI reads your site - it's whether it finds anything worth reading.
Ship new listings relentlessly. Freshness was the strongest signal in the data, by far. A marketplace with active sellers gets read deeper every week. This is also the quiet argument for recruiting sellers: they write your AI-search content for you, every day.
Check the front door. Look at whether your listing pages are linked, server-rendered, and in the sitemap. This is boring and it beats every clever trick in the data.
Write for the quote, not the crawl. Good descriptions didn't earn more reads - but reading is not quoting. What an AI does with a page it HAS read almost certainly depends on what's written there. Nobody can measure that part, and anyone who claims they can is selling something.
That last sentence is the honest limit of all of this. I can see the machines reading. Nobody can see them quoting. But the reading comes first, and for the first time I can actually watch it happen.
The full research - methodology, charts, and the raw day-by-bot dataset as a CSV - is here: What AI Crawlers Actually Read: 7 Days of Data From 13 Real Marketplaces
And I made a video walking through the whole week of logs: I Logged Every AI Crawler for 7 Days
The counter is still running. At 30 days I'll write up whether these patterns held - including one experiment already in motion: we took a listing the bots kept walking past, rewrote it, and we're waiting to see if they come back.
- Rasmus
P.S. This is the first issue of The Marketplace Guy. I build Prometora, a marketplace platform, and this is where I write down what running the infrastructure under many marketplaces teaches me. If someone forwarded you this: you can subscribe at themarketplaceguy.beehiiv.com.