Using Jev for SEO

SEO at scale is a pile of small, repeatable judgments: what does this keyword intend, is this page thin, should these two URLs link to each other, is this generated page worth publishing? Jev is TypeSafe AI's System One model — it answers each of those as a typed, calibrated value in roughly 70–500ms, so you can run them across thousands of URLs for a fraction of a cent each instead of prompting an LLM and parsing prose.

Jev doesn't write content and it isn't a rank tracker — it's a decision model. You hand it some state (a keyword, a page, a pair of URLs) and one or more typed questions, and it returns structured values — choice, score, noul with calibrated confidence — in a single non-autoregressive pass. That makes it a good fit for the judgment steps that sit between your crawler/CMS and your publishing pipeline, where you need a consistent, machine-readable answer rather than a paragraph.

Where Jev fits an SEO workflow

• Keyword-intent classification — a choice over informational / commercial / transactional / navigational, so you template the right page type for each query
• Content-quality and thin-content gate — a score or noul on whether a page is substantive enough to publish or index
• Internal-link discovery — for a candidate source/target pair, a noul 'should these link?' plus a suggested anchor, run across the whole site
• Cannibalization / duplicate check — a noul 'do these two pages target the same intent?' so you merge or canonicalize instead of competing with yourself
• Programmatic-SEO publish gate — score each generated page for real, non-duplicate value before it ever goes live
• SERP-title / snippet fit — a score on whether a draft title actually matches the query it's meant to rank for

Classify keyword intent in code

// Classify search intent + decide page-worthiness, one typed call per keyword.
async function triageKeywords(keywords) {
  return Promise.all(keywords.map(async (keyword) => {
    const res = await fetch("https://jevtypesafeai.com/api/v1/decide", {
      method: "POST",
      headers: {
        "Content-Type": "application/json",
        Authorization: `Bearer ${process.env.JEV_KEY}`, // jv_live_...
      },
      body: JSON.stringify({
        state: { keyword },
        questions: {
          intent: {
            type: "choice",
            instructions: "What is the searcher's primary intent?",
            options: ["informational", "commercial", "transactional", "navigational"],
          },
          worthAPage: {
            type: "noul",
            instructions: "Is this keyword worth its own landing page (real, non-duplicate intent)?",
          },
        },
      }),
    });
    const { answers } = await res.json();
    return { keyword, intent: answers.intent.choice, build: answers.worthAPage.noul };
  }));
}

answers.intent.choice is locked to the options you listed — it can never come back as free text or an out-of-set label — and answers.worthAPage.noul is a calibrated 0–1 (RLCD training makes the number trustworthy), so you can gate on a threshold like 'build if > 0.6' and have it behave consistently across a 10,000-row keyword export. Run the rows in parallel and the whole batch costs a fraction of a cent per keyword with output tokens free.

Gate a programmatic-SEO pipeline

The real failure mode of programmatic SEO is shipping thousands of near-identical thin pages that get the whole site demoted. Put Jev in front of publish: score each generated page for substance and ask a noul 'is this meaningfully distinct from pages we already have?'. Only pages that clear both thresholds go live; the rest get held for a human or merged. It's a cheap, typed quality floor that scales with the page count instead of fighting it.

Why a typed pass beats prompting an LLM

Try it without writing code

The Internal Link Finder crawls your site and uses Jev to surface where you should add internal links — with a relevance verdict, the relationship, and a suggested anchor, exportable to CSV. Or open the playground and run a keyword-intent or content-quality question on your own text first. When you're ready to wire it in, point your pipeline at POST https://jevtypesafeai.com/api/v1/decide with a jv_live_ key (official upstream POST https://api.typesafe.ai/v1/systemone).

FAQ

See also: Internal Link Finder tool · Jev as a judge · Jev classifier · Jev for RAG · Playground