ProofStack research · 2026-08-22
The winners without sitemaps: 24% of audited products skip SEO
We probe the site structure of every product in the library — sitemap, llms.txt, page composition. One counter-intuitive statistic fell out: 10 of the 42 audited products have no sitemap at all. Not misconfigured. Nonexistent. Every claim below carries its observation date and can be reproduced with a curl.
The absentees are more respectable than you'd think
These are not obscure products. The ten include Starter Story (one of the few A-grade revenue disclosures in our library), Gumroad (B-grade), Kit (ConvertKit) (B-grade), Carrd, Fathom Analytics, and Readwise (as of its 2026-08-10 snapshot). We re-checked Carrd on 2026-08-22: carrd.co/sitemap.xml 301s to carrd.com, then 404s. Still nothing.
What they share is plain: each has a distribution asset that predates or replaces search — Gumroad's sellers bring their own buyers, Kit grew on creators' existing audiences, Carrd rides its template community and word of mouth, Starter Story is itself a media property. Search is optional when another engine is already turning.
Skipping Google is not skipping machines
Of the ten sitemap-less products, three ship an llms.txt (Fathom, Gumroad, rustdesk; observed 2026-08-07 through 08-17). That combination is worth staring at: no page inventory for search crawlers, but a hand-written machine-readable product summary for AI. Classic SEO gets skipped while AI legibility gets cultivated separately.
Readwise's nine URLs
The best part is a dated change. In our 2026-08-10 snapshot, Readwise had no sitemap. On 2026-08-22 it does — and the entire file contains nine URLs: the homepage, /read, /pricing, /pricing/reader, /cli, /mcp, /openclaw, /privacy, /tos.
Notice what is absent: not one blog post, not one template page, nothing for a keyword crawler to feast on. What is listed is exactly what a machine needs — pricing, a CLI, an MCP server. We can't know their intent, but in effect this is not an SEO sitemap; it reads like a product index written for AI. A product famous for never doing SEO published its first page inventory, and put /mcp on it. That is probably what distribution looks like in 2026.
What you can take away
- A sitemap is a strategy signal, not hygiene. Whether one exists and what it contains reads out a product's acquisition strategy in thirty seconds — 175 template pages is one playbook, nine machine-facing pages is another, nothing at all is a third.
- "No SEO" is a choice that requires capital. These ten can afford the jump because audiences, marketplace effects, or communities distribute for them.
- Machine legibility is decoupling from SEO. llms.txt and MCP showing up on products that ignore search entirely means "being readable and usable by AI" is now its own channel.
What not to copy
If your product has no existing audience, no marketplace effect, and no community, imitating their indifference to SEO doesn't make you cool — it makes you invisible. Skipping search at zero means switching off the one free channel that doesn't care about your follower count. Ask whether you have their replacement engine before you borrow their swagger.
Method
Site structures come from direct fetches of each product's public sitemap and llms.txt on the stated dates; snapshots are visible on the corresponding case pages, and evidence grades link to each case's source list. No ranking or traffic figures appear here — those cannot be verified for free, and we only state what anyone can reproduce.