GUIDE · 2026-09-22 · READ TIME 7 MIN

Jev for SEO: engineering guide and benchmarks

Jev for SEO is the practice of using TypeSafe AI's Jev, a System One decision model, to run structured SEO judgments: intent classification, freshness scoring, internal link mapping, and cannibalization checks, at roughly $0.0004 per decision. jev-seo is the free, MIT-licensed Rust client that makes Jev for SEO practical: it scrapes DuckDuckGo for free and reads your local files, then batches Jev questions against that state. No hosted dashboard, no subscription.

What Jev is, and why it matters for SEO

Jev is a decision model, not a writer. Given a blob of state and a question with fixed answer choices, it returns a typed answer: a Choice, a Score, or a Noul probability. That design changes what SEO tools can cost. Traditional LLMs generate tokens, which is expensive at scale. Jev answers questions, which is cheap at scale: about $0.0004 per decision at list price, with output tokens free, at up to 250k tokens per second.

That combination makes workflows practical that a dashboard would bill you hundreds of dollars for: classify every keyword on your Search Console export by intent and funnel stage, score a competitor page for answer depth and freshness, judge whether each page deserves an internal link, and flag duplicate titles that cannibalize each other.

Why hosted dashboards are the wrong tool for this

Semrush and Ahrefs charge over $130 a month and gate their best data behind higher tiers. For a single site, a freelancer, or an agent workflow, that is a license fee for what is mostly batch judgment. Jev-based tooling inverts the economics: the judgment engine costs fractions of a cent per call, and the data layer can be free when you scrape DuckDuckGo and audit local files.

Real benchmarks

Numbers from the current release, v0.1.0, measured on a Windows 11 machine with 8GB RAM and a 2-core i3-1115G4:

• Release binary: 6.7MB executable, 3.2MB zipped download.
• Idle memory: under 30MB RAM, safe for CI runners and low-RAM laptops.
• Audit pass on a 6-file site: reads structure, schema, headings, and copy in one pass, in seconds.
• GEO scoring: direct-answer confidence 0.90 reported on a well-formed landing page; composite score 7/10 after content fixes.
• Search radar: live DuckDuckGo SERP scraping with no API key; ranked the project #1 for its own head term in one shot.
• Jev inference speed at the API: 70 to 500ms per decision, up to 1200 requests per minute.

Full runnable details on the benchmarks page.

Step by step: run a Jev for SEO audit today

1. Install

# cargo route
cargo install jev-seo

# or grab the release binary for Windows, macOS, or Linux:
# https://github.com/AkashPriyadarshii/jev-seo/releases

2. Audit your site

jev-seo audit .

The audit walks local HTML files and reports pass/fail on titles, meta descriptions, canonical links, headings, schema, orphan pages, thin pages, and internal links. On a clean page set you should see pass rates in the 70 to 90 percent range, with each failure pointing at the exact file and rule.

3. Validate schema

jev-seo schema index.html

Parses JSON-LD and reports validity against Schema.org, catching malformed fields before Google does.

4. Score GEO citation likelihood

jev-seo geo --query "jev seo" --target index.html

Scores how likely a generative engine (Gemini Overviews, Perplexity, SearchGPT) is to cite the page for that query, with a confidence value and direct-answer detection. If the score sits below 7/10, the tool tells you which signals are missing, usually direct answers, statistics, or step-by-step structure.

5. Generate a content brief

jev-seo brief "jev for seo"

Reads the live SERP and produces a brief blueprint: suggested title, target word count, search intent, audience, and the differentiator to beat the incumbent, for example a missing step-by-step section.

6. Run the search radar

jev-seo keywords "jev seo"
jev-seo query "jev seo"
jev-seo rank --domain jevseo.vercel.app --query "jev seo"

Keyword intents, live SERP competitors, and weekly rank drift for your own domain, all from DuckDuckGo with no API key.

7. Check robots and sitemap against AI crawlers

jev-seo robots https://jevseo.vercel.app
jev-seo sitemap https://jevseo.vercel.app

Reads your live robots.txt and sitemap.xml, reports which AI crawlers (GPTBot, ClaudeBot, PerplexityBot) are allowed or blocked, and validates sitemap URLs and HTTPS.

8. Wire it into an agent

jev-seo mcp

Starts an MCP server so Claude Code, Cursor, and other agents call every command above as native tools.

9. Keep the tool fresh

jev-seo update

Automatically lifts the latest release binary from GitHub.

When Jev for SEO is the wrong call

Jev does not write articles, generate meta descriptions from scratch, or explain its recommendations in prose. It also cannot crawl on its own; someone has to collect state. If your workflow needs generative copy or a full site crawl with rendered JavaScript, pair Jev with an LLM for writing and a real crawler for fetching, and keep Jev on the judgment layer where it is cheap and typed.

What to do next

Install jev-seo, run jev-seo audit . on your own repo, then jev-seo geo on your money page. Compare prices against your current dashboard for one month. If the numbers hold, cancel the subscription and keep the binary. Full command catalog on the use cases page, and per-command depth on the benchmarks page.

NEXT: benchmarks → · use cases → · home