How machine-readable are creator websites?

What did we measure across creator websites?

When a long-form video creator sets up a website, the site usually exists to serve human visitors. It offers a place to sell merchandise, book speaking appearances, link to social profiles, or invite people to join a mailing list. But as search shifts toward answer engines that read and summarize web pages directly, a website plays a second role. It is where the knowledge in a creator’s videos can live as text on a site the creator owns.

In The Dark Archive, our study of long-form YouTube channels, we wanted to see how well creator websites perform that second role. We evaluated the 528 creator-owned websites we could reach on the open web. We scored each site using our machine readability score, a measurement graded out of 100.

The results showed a wide gap between owning a website and having a site that machines can actually read. Across all 528 sites, the median score was 20, and exactly 0 scored above 70. Even channels that maintain an active written archive did not score above 70.

To understand why these scores are so low, it helps to examine what the score measures.

What do the five parts of the score ask of a site?

The machine readability score is built from five distinct components defined in our glossary. Together, they ask whether a machine can tell what a page is, read its text, reach it, see which video it came from, and expect more pages like it.

In our reading, the five parts reflect what any creator can reasonably ask of their web presence if they want their ideas cited by machines:

• Structured data. This asks for machine-readable markup, such as JSON-LD, that identifies what the page is and connects the written article to the specific video it discusses. We should concede, as we explored in our note on how schema is not the AI citation lever it is often sold as, that the evidence we found does not show markup alone lifting citations. In our reading, it labels the writing; it does not replace it.
• Extractable text. The words and arguments must live directly in the page HTML as readable text. If a creator embeds a video player, puts key takeaways into an image graphic, or relies heavily on client-side scripts that render blank pages to simple crawlers, an automated engine finds very little to read.
• Crawler access. The domain’s robots.txt configuration must permit AI crawlers to inspect its pages. As we showed in our analysis of what blocking AI crawlers costs, access is rarely the gap in practice. Most creator sites already leave their doors wide open to AI bots, even when there is little for those bots to read.
• Attribution. Each written post should explicitly name and link the specific video from which it was adapted. It tells a reader, or a machine, where the text came from, and gives readers a path back to the video.
• Systematic publishing. The written archive must keep pace with the channel’s video releases rather than ending after a brief burst of enthusiasm when the site was first designed.

Why are creator website scores so low?

At least half the sites scored 20 or less, and none scored above 70. Why are the scores so low?

In our reading, the primary reason is that the score rewards an archive, not a homepage. A domain that works as a landing page or a digital business card has little on it to read. On the Dark to Sovereign ladder, a channel whose own domain has no content index (no blog, no episode pages, no feed) is Beacon. Its site may serve visitors well, but it gives a machine little text to extract.

A site with a blog or episode pages can still sit on the Dormant rung, when its posts have stopped or are not drawn from the videos. In our reading, the first loses points on systematic publishing and the second on attribution.

Even the rare Sovereign channels, which publish most of their videos as articles, did not score above 70. Which parts held them back is in the report.

An llms.txt file is not one of the five parts. As our note on llms.txt shows, such a file is no substitute for written pages either.

What can creators do to make their sites readable?

We should be clear about what these scores mean. A low machine readability score does not mean a creator has built a bad website. If a site exists to sell merchandise or take consulting inquiries, it may perform that role well. We make no claim that every creator needs a high score, or that scoring above 70 brings search traffic or income.

However, for creators who produce research-heavy videos, investigative documentaries, or educational explainers, machine readability matters. When the work exists only in video, an engine that cites only text has nothing of yours to quote.

In our reading, creators who want their knowledge to be read and cited by machines can take four straightforward steps:

First, focus on extractable text before any technical optimization. Turn your video scripts, research notes, and explanations into written articles hosted on an owned domain. Make sure the text appears in plain HTML rather than inside complex scripts, image cards, or closed players.

Second, connect every written piece to its source video. State clearly at the top of the post which video the article was drawn from, and provide a direct link to that video. That tells readers, and machines, where the text came from.

Third, maintain an ongoing publishing routine. You do not need to convert your entire back catalogue overnight. Instead, establish a cadence where each new major video upload is paired with a corresponding written article on your site.

Fourth, attend to the basic technical hygiene of your domain. Ensure your robots.txt file does not inadvertently block search crawlers, and add straightforward structured markup that declares the page as an article linked to a video asset. Structured data will not create citations on its own, but it labels what the page is once the written content exists.

Where are the full numbers?

This note presents our findings on the overall machine readability score and what its components require. The full score distribution, the score by tier and the result for each part are in The Dark Archive report.

To see how machine-readable your own channel’s website is, run the free Dark Matter Check.