Promptless serves AI readers a markdown version of its pages, which is exactly what it sells. But when we fetched the site with an AI-agent reader on November 9, 2026, /blog, /changelog and the Agent Score tool page /agent-score all returned HTTP 404, with and without a trailing slash and on repeat fetches, and the Starport docs page answered "This page does not exist in its markdown form." The same reader loaded /pricing, every individual post, every monthly changelog, /agent-score.md and public Agent Score reports without trouble. So the pages that show momentum (25 monthly changelogs since October 2024) and the free tool at the top of the funnel are the ones an assistant cannot open. These URLs are in the sitemap and the navigation, and the homepage invites visitors to ask AI assistants about Promptless. An assistant asked "what has Promptless shipped lately?" or "run Agent Score on my docs" hits an error and gets its answer somewhere else.
First action
Request /blog, /changelog, /agent-score and /docs/for-docs/starport with a markdown Accept header, confirm the 404s, and find the rule that returns them.
How to measure it
Agent-readable sitemap coverage
- Now
- –
- Target
- 100% of 234 URLs
- By
- 1 week
Steps
Reproduce
Request each of the 234 sitemap URLs as a browser and as a markdown-requesting agent, and list every one that differs in status code.
Founding Engineer 2 hours
Find the rule
Locate the content-negotiation or rewrite rule that returns 404 when no .md file exists.
Co-founder & CTO 2 hours
Add the fallback
Serve the HTML page with a 200 whenever the markdown version is missing, instead of an error.
Founding Engineer half a day
Generate the missing markdown
Build blog and changelog index pages from the existing llms.txt lists, map /agent-score to agent-score.md, and export the Starport overview.
Founding Engineer 1 day
Guard it in the build
Add a deploy check that requests every sitemap URL as an agent and fails on any 404, alongside the afdocs checks Agent Score already runs.
Co-founder & CTO half a day
Fix parity
Close the two markdown and HTML parity failures Agent Score reports on the governance and triggers docs.
Forward-deployed technical writer 1 day
Score the whole site
Run Agent Score on promptless.ai as well as /docs and publish the report.
Co-founder & CEO 1 hour
Watch the referrals
Track visits from AI assistants to the blog, changelog and Agent Score in the analytics tool, weekly.
Co-founder & CEO weekly
Why it works
Change the rule that serves markdown so that it never returns 404 when only the markdown version is missing: return the normal HTML page with a 200, or a markdown page generated from what already exists. Blog and changelog listings can be built from /blog/llms.txt and /changelog/llms.txt, /agent-score can return the content of /agent-score.md, and Starport needs its overview page exported. Then add a check to the deploy pipeline that requests every sitemap URL the way an agent does and fails the build on any 404. Finally, run Agent Score on promptless.ai itself (not only /docs) and publish the A+ report as proof that the company follows its own advice.
Promptless's own blog and Agent Score argue that agents are now a primary reader of docs and product sites. Every page that answers an agent with an error is a missed citation in an assistant's answer, and it contradicts the product's own pitch in front of the most technical buyers. The fix is small, can be tested in an afternoon, and protects every future page.
Ready to use
ChecklistAgent-readable site check
[ ] /blog returns 200 to a markdown-requesting agent [ ] /changelog returns 200 to a markdown-requesting agent [ ] /agent-score returns the tool description (agent-score.md) [ ] /docs/for-docs/starport has a markdown version [ ] No sitemap URL returns 404 to agents but 200 to browsers [ ] Deploy check fails on any agent-only 404 [ ] Markdown and HTML parity fixed on the 2 flagged docs pages [ ] Agent Score run on promptless.ai, not only /docs
Risks and how to handle them
Serving HTML to agents breaks the markdown promise
Treat HTML as the fallback only; generate markdown for the listing pages so agents get the clean version.
Assistants cache the old 404s
Resubmit the sitemap and link the fixed pages from llms.txt so crawlers fetch them again.
Evidence
- Your websiteAI-reader check (November 9, 2026, each URL fetched at least twice, with and without a trailing slash): /blog, /changelog and the Agent Score tool page /agent-score returned HTTP 404, and /docs/for-docs/starport returned "404 – Page Not Found" with "This page does not exist in its markdown form." The same reader got full content from /pricing, /pricing.md, /agent-score.md, individual blog posts, each monthly changelog and public Agent Score reports. /blog/, /changelog/ and /agent-score/ are in the sitemap, Blog and Changelog are in the site navigation, and llms.txt lists Starport and Agent Score as pages.Source
- Your websiteSitemap (234 URLs, checked November 9, 2026): 58 blog posts, of which one is a customer story (Vellum); 25 monthly changelog entries from October 2024 to October 2026; 135 docs pages; one use-case page (/use-cases/technical-writer-power-tool/); free tools Agent Score and a Broken Link Report; event pages /lavacon/, /bingo/ and /bingo-leaderboard/; no comparison or alternatives pages.Source
- Your websiteAgent Score: a free tool, "no signup", that scores from 0 to 100 how well AI agents can read a docs site, using 28 checks in 7 categories from the open-source afdocs project; every report is public, with fixes, score history and an SVG badge. The leaderboard lists 23 scored sites. Promptless's own docs score 100 (A+, scan of October 19, 2026) but fail the markdown and HTML parity check on 2 of 50 pages. Agent Score "checks infrastructure, not content quality", and the page points to a content strategy audit with a free 14-day trial.Source
- Your websiteHomepage: headline "Continuously improve your AI workforce and docs" with two products, Promptless for Docs ("Write the docs. Skip the busywork.") and Promptless for Agent Instructions; 14 customer logos including Megaport, Mezmo, Vitess, Helm, Aptible, Runpod, Flatfile, Latitude, Mautic and Vellum; testimonials from a Senior Technical Writer at Runpod ("My team literally calls me a 10x tech writer"), the CTO of Latitude.sh ("basically pays for itself right away") and the Head of Developer Docs at Prove ("a solo tech writer's godsend"); buttons that open AI assistants to ask about Promptless; "Every plan includes a 14-day free trial"; Agent Instructions claims typical 30-day results of token spend down 18%, first-attempt completion up 15%, wall-clock time down 32% and human interruptions down 42%, with "white-glove onboarding with a forward-deployed AI engineer and technical writer".Source