The 2026 AI Search Readability Index: 50 Top Tech & DTC Brands Audited

Key takeaways

1. Crawler rules are silent, not hostile

Not one measured site blocks OAI-SearchBot or PerplexityBot from its home page. 2 block another answer-engine crawler (Amazonbot). The bigger pattern is that 36 of 46 sites name no AI crawler at all, so every bot falls back to the generic * rules. Unless those rules block them, that lets them in, but it states no intent: there is no separate decision for search crawlers and training crawlers, and Crawlable's governance pillar scores that 0 of 20. Naming retrieval bots and training bots in separate groups takes a few lines. The platform guides have the code.

2. Most sites have an llms.txt; most of those fail the spec

14 sites serve an llms.txt that passes Crawlable's validator: one # heading, a > summary, ## sections, and absolute links with a description after each. 35 of 46 serve a file at all; 21 serve a file that misses at least one of those, most often the summary line or the link descriptions, and 11 serve none. 8 of the files that fail are Shopify's auto-generated "Agent Instructions", which every store has served by default since May 2026: it documents checkout endpoints for shopping agents rather than mapping the catalog. A store can replace it with a theme template; the Shopify guide shows how.

3. Empty JavaScript shells: 14 of 46 sites

10 of 46 sites earned the full 35 readability points. Of the 36 that did not, 22 had no empty page at all and lost points to short or markup-heavy pages. But 14 of 46 sites served at least one audited page as an empty JavaScript shell, where the text arrives only after the browser runs the app. Among the 26 sites detected as Next.js, 11 had at least one such page. The extreme case was v0.app, where 37 of 40 audited pages were shells. The Next.js guide shows the usual cause, content fetched after the page mounts, and the fix. Shells cost more than their own pages: the rendering gate capped 4 sites at 64.

4. Schema depth: 27 of 46 sites complete

27 of 46 sites carry a complete rich entity, such as a SoftwareApplication with an offer or an article with its author, and earn all 25 schema points. 12 of 46 sites have nothing beyond Organization or WebSite markup, or no valid JSON-LD at all.