A 586-page website audit took 45 seconds and cost 21 cents
The job was not writing, but 8,790 yes or no decisions about which pages should link to each other.
The AI Daily Brief (Formerly The AI Breakdown): Artificial Intelligence News and Analysis · How People Are Actually Using JevBorja, who was running a search-engine audit of his 586-page website, needed to map how the site's pages connect, the record of which pages link to which other pages.
The work sounds simple and is tedious to complete. For every page, someone must decide whether it has a good reason to link to another page, and whether the words to link from are already there.
Jev, a judgment model from the company Typesafe, does not write like a chatbot. It answers small questions, including yes or no questions asked at very high volume.
Deployed on Borja's site, Jev read all 586 pages and rebuilt the map in 45.1 seconds for 21 cents. Claude Opus Five, a frontier model, got through only 21 of those pages and spent $1.43.
Jev suits the work because the audit is a set of classification questions, not writing. Borja said the audit came down to 8,790 yes or no calls. He described each check as, "Does this page have a real reason to link to that one, and is there anchor text already sitting in the copy? That is a classification problem, and we have been paying frontier prices to do it one page at a time." Anchor text means the clickable words of a link.
Questions people ask
What exactly is Jev?
Jev is a judgment model from the company Typesafe. Instead of writing text like a chatbot, it answers small questions at very high speed, such as yes or no, pick the best option, or rate something on a scale you describe.
Why does an internal link map matter?
An internal link map shows how the pages on a website connect to each other. Search engines use those links to understand a site, and readers use them to move between related pages, so missing or weak links can hurt traffic.
How can it be that much cheaper and faster?
Jev asks thousands of simple questions at the same time and returns only short answers, often just a probability. It skips the long reasoning and long written replies that make frontier models slow and costly, and its maker charges only for input, with output tokens free.