Using Jev in Search Engine Optimisation
Recently for my SEO work I've been using "Jev", a new model / API from TypeSafe, built for judging things rather than writing them like normal LLMs do. It's really cheap, so I can scan every page of a client's website with it for a few pennies.
Doing that with normal LLMs is expensive because every question you ask a chat model bills output tokens, and a judgement on every page of a big site means hundreds of calls.
Jev doesn't generate text at all: you send it a page's content plus a list of questions, each with a fixed answer shape - a true/false, a score against a scale you define, or a pick from a list - and it answers every question in one go, each with a probability and a confidence score. It can't make things up, because it can only answer in the shapes you gave it.
The catch is that it's only as good as your questions. Mine ask about the things that matter for search - "does the copy show real experience of the thing?", "does the page answer what someone searching for it actually wants?", "is it concrete facts or AI filler?", yadda yadda - and in my scripts each answer has a weight, so every page comes out with a score out of 100 to give me a priortised list of things to fix.
For rewrites I use ordinary LLM agents on cheap open-weights models: GLM 5.3 Flash is my current favourite, via Neuralwatt which has a "flex" option at 35% off where your requests wait when their servers are busy - fine for setting a batch of rewrites for me to come back to and edit. Each page costs about a penny to research and redraft.
If you want to try Jev yourself, it's on OpenCode Zen with pay-as-you-go credits.
I charge an hourly rate for this, like with everything else I do, so my incentive is to get the most out of every hour I'm paid for. My guess is that most people reading this would still rather not build it themselves, which suits me great - if you want it done on your site, get in touch about my SEO audits, or contact me for a personalised quote.