Jev for SEO and GEO

Jev is useful for SEO and GEO when a task is a judgment about text, such as the search intent of a query, whether a page answers it, or which pages belong together. It returns typed answers with probabilities rather than writing anything, so it grades and sorts pages but does not write titles, meta descriptions or copy (System One).

What Jev does in an SEO workflow

TypeSafe does not publish an SEO guide. Its docs do describe the building blocks SEO tools need:

Relevance and ranking. Score how well a candidate matches a query and re-rank results (Example use cases).
Classification into a hierarchy. Walk a taxonomy, such as a site structure, with Choice questions (Hierarchical classification).
Writing checks. Define checks for your writing guidelines and run them in CI (Example use cases).

Because many questions share one state and run in parallel, you can ask everything you need about a page in one request (Speculative fan-out). At $0.042 per million input tokens with free output, cost for a site-wide audit scales with the amount of page text you send (Models). See Jev pricing.

A documented pattern: code measures, Jev judges

TypeSafe's core advice is to keep deterministic rules in code and give the model narrow judgments only (How to build with TypeSafe). For SEO that split is natural:

In code: title length, word count, keyword counts, status codes, schema validity, link counts. Jev does not count reliably and should not do arithmetic (Jev 1.13 jaggedness).
In Jev: does this page serve the intent behind the query, is the answer self-contained, is this page a near-duplicate of that one.

Retrieval work follows the same shape. The re-ranking cookbook uses BM25 to build 30-passage shortlists, then one TypeSafe question per query-candidate pair. Top-1 accuracy rose from 5% to 18% on 40 legal queries (Re-ranking). The line-by-line search cookbook adds a Noul that asks whether the document answers the query at all, which is the question behind a content-gap check (Line-by-line search). More in Search and RAG.

Question designs for SEO and GEO

Choice, search intent: informational, commercial, transactional or navigational, plus an other option in case the list does not fit (Primitives).
Score, intent match: levels from "does not address the query" to "fully answers it", each described in words (Score).
Noul, answer present: "Does `page.body` directly answer `query`?" One condition per Noul (Noul).
Noul, GEO mention: "Does `ai_answer` recommend the brand named in `brand`?" run over saved AI answers you have collected.
Choice, internal links: which of up to 255 candidate pages is the best link target for this paragraph (API reference).

Combine several Scores into one page grade with weights you set in code, as in the composite scoring pattern (Composite scoring).

Limits to plan for

Jev reads text only. Strip HTML to the text and fields you care about, and describe images in text if they matter (Models).
The context limit is 64k tokens per request and 32k for the state plus the longest question (Models). Accuracy also falls as the state fills with irrelevant detail, so send the section a question is about rather than the full page (Jev 1.13 jaggedness).
Pages written to argue for their own classification can move the answer (Jev 1.13 jaggedness).
Jev does not generate text (Jev with coding agents). Use an LLM or a writer for rewrites, and Jev to check the result.
A grade is a judgment about text, not a ranking prediction. Nothing in the TypeSafe docs connects Jev scores to search rankings or AI citations.

Background on the model: What is Jev? and How to use Jev. More marketing builds are in Marketing and social.

Projects doing this

Marketing, sales & socialjev-seoAn agent-first SEO and GEO CLI and MCP server in Rust that pairs DuckDuckGo results with optional Jev scoring for intent, gaps and AI visibility.GitHub · ★ 47 · AkashPriyadarshii

Marketing, sales & socialjevseoScores a page separately for SEO, AEO and GEO, with fixed thresholds in code and Jev answering the questions that need judgment.GitHub · ★ 4 · epergaboni

Marketing, sales & socialjev-seo (AgriciDaniel)Crawls a site from one URL, applies 52 rules plus PageSpeed, and uses Jev page and site judgments to rank fixes.GitHub · AgriciDaniel Marketing, sales & socialNotraTracks whether AI answers mention your brand, where it appears and how it is described, using narrow Jev judgments.GitHub · ★ 219 · usenotra

Marketing, sales & socialpagegradeA Chrome extension that grades page sections for clarity, writing and on-page SEO with Jev Score rubrics.GitHub · ★ 5 · kitze

Marketing, sales & socialSiteClarityAn AI answer-readiness audit that pairs structural checks with Jev meaning judgments, each finding backed by a quote from the page.Site · sanjuacodez

Marketing, sales & socialInternal linking toolA free tool where Jev classifies pages and picks which ones should link to which.Site · Ian Nuttall Marketing, sales & social8 SEO workflows you can put Jev to work onRoss Simmonds covers internal links, cannibalisation, thin content, search intent, content audits, redirect maps, schema checks and AI visibility.X post

Marketing, sales & socialboringtoolskit.comA technical SEO audit cost study in which Jev ranks striking-distance fixes and content gaps by probability.Site

Data & extractiondeslopA CLI and agent skill that scores fetched pages with four Jev probabilities: ad, slop, SEO and derivative.GitHub · yoichiojima-2

Questions

Can Jev write meta descriptions or blog posts?

No. Jev does not generate text. It can grade drafts you or an LLM wrote, for example whether a meta description matches the page, but it will not write them.

What does GEO mean here?

Generative engine optimisation: getting a page quoted or cited by AI assistants. Jev can judge saved AI answers, for example whether one mentions your brand, but you have to collect those answers yourself.

Will Jev scores improve my rankings?

Not on their own. A Jev score is a judgment about the text you sent. TypeSafe makes no claim that its scores predict search rankings, so treat them as audit signals and test changes the usual way.

How big a page can Jev read?

The models page lists 64k tokens per request and 32k tokens for the state plus the longest question. Accuracy drops with irrelevant content, so send only the relevant section.

Should keyword counts go to Jev?

No. TypeSafe says jev-1.13 does not count reliably. Counts, lengths and other measurements belong in code.

More guides: What is Jev?Jev pricing and API cost Jev vs LLMs: when to use which How to use Jev: a quickstart Jev and MCP: using Jev as a Model Context Protocol tool How to use Jev with Claude Code Jev in an agentic harness: where it fits in an agent loop Jev for orchestration: routing requests, models and skills Jev as a judge: evaluating LLM and agent outputs Jev for ads and ad review Jev for marketing and lead scoring Is there an open source Jev?Can you fine-tune Jev?Designing questions for Jev What is RLCD?Jev statistics