When we first shipped AI Visibility, the index covered the first 100 products in a store. That was a deliberate limit while we learned what assistants actually do with the file, and for a small catalog it was enough. For most of the merchants who installed it, it was not.
From today, every store on the app indexes up to 500 products — titles, prices, categories, collections and store policies — and the index regenerates whenever any of them change.
Why the limit existed at all
An LLMs.txt file is read in one pass. Everything in it competes for the same attention, and a file padded with thousands of near-identical variants is measurably worse than a shorter file that describes a catalog cleanly. Our first release capped the count while we worked out where the useful ceiling sat.
What we found is that the ceiling is much higher than 100, but that it is about structure more than length. A file listing 500 well-described products performs well. A file listing 500 products whose titles are all "Product — Variant 3" does not, regardless of how many entries it contains.
The limit was never really about size. It was about how much of a catalog a model can read before the description stops describing anything.
What changed underneath
Raising the number meant rebuilding how the file is produced. The old generator rebuilt the entire document on a schedule. The new one tracks changes per product and rewrites only the affected sections, which is what makes a 500-entry file practical to keep current rather than merely possible to produce once.
- Products, prices, categories and collections are pulled directly from Shopify
- The index rebuilds when a product, price or policy changes, not on a fixed timer
- Store policies — shipping, returns, exchanges — are described alongside the catalog
- AI-written store and product descriptions are refreshed in the same pass
What merchants need to do
Nothing. Stores already running the app are moving to the higher limit automatically, and the first rebuild happens the next time your catalog changes. There is no setting to enable and no file to upload.
If your catalog is larger than 500 products, the app selects the set it indexes by collection membership and recency rather than truncating the list arbitrarily. We would rather describe half a catalog well than gesture at all of it.
What comes next
The number is not the interesting part. The next release is about how well each entry is written: which attributes an assistant actually uses when it answers a shopping question, and which ones are noise. That work is ongoing and will ship as an improvement to the same file rather than a new feature to configure.