Jev for Marketing: Social and SEO Tools Built on It
One project routes social-media research through Jev, and two unrelated tools named jev-seo audit pages against dozens of rules. What their evidence actually supports, and what it doesn't.
Marketing tooling is one of the more crowded corners of the Awesome Jev radar, and it's a good test of this series' core discipline: separate what a typed decision model actually verifies from what a marketer wants it to mean. A GEO ("generative engine optimization") score is not a measurement of how often ChatGPT or Perplexity actually cites your page — it's a rubric a model applied to your content. Three projects here make that distinction with varying degrees of clarity.
One, jev-social, uses Jev to route a browser-automation research loop across social platforms. The other two are both named jev-seo, built independently by different authors, and both audit websites against a set of SEO and GEO rules — but they differ meaningfully in scope, rule count and what they claim.
Key Takeaways
- Two unrelated projects are both called jev-seo. AkashPriyadarshii's is a Rust CLI with 53 rules and a 13-tool MCP server; AgriciDaniel's is a Python tool with 52 rules and PDF/Excel/Markdown output. Check the author before citing either.
- Neither jev-seo's GEO or citation score measures actual AI-search citations. Both apply a rubric; AkashPriyadarshii's README says uncertain answers are gated behind
[verify]markers rather than printed as fact. - jev-social is read-only by design: search, open a profile or post, read comments — no posting, no engagement actions.
- AkashPriyadarshii's jev-seo hard-caps AI spend via a
--jev-budgetflag, default $0.25 per run, and offers a fully free DuckDuckGo-only path. - All three send data to external services — page content, social post text, or search queries — so none is a private, offline tool.
jev-social: a typed router for social-platform research
jev-social (listing) pairs Jev with socai, a separate browser-automation CLI, to research Instagram, TikTok and LinkedIn through a real Chrome browser. Instead of generating arbitrary commands, Jev picks the next operation from a fixed, observed set: search, open a specific post or profile, read comments. Each step's captured results feed the next decision, so the loop only ever targets things it has actually seen on the page, not coordinates a model guessed at.
It's explicitly read-only — no posting, no liking, no following — and the author frames login or access gates as producing partial results rather than failures. Step counts are bounded (1–30, default 12), and speed depends on which operations get selected and how the live site responds; the one number in the README is a 64-second Instagram run that captured four records across two searches and one post-detail read, which is a single example, not a throughput guarantee. It runs via npx github:socai-io/jev-social, needs Node 20+, an OpenRouter key and a current socai CLI install, and reports cited sources alongside captured text rather than free-text summaries. MIT-licensed, 47 stars as of September 2026.
AkashPriyadarshii/jev-seo: a Rust CLI with confidence gates
jev-seo (AkashPriyadarshii) (listing) is a single Rust binary combining local content audits (53 rules), live crawling that respects robots.txt, a citation-readiness score for answer engines, SQLite-based rank tracking, and CI/CD gating on audit thresholds. It also ships a 13-tool MCP server for agent integration with Claude Code, Gemini CLI and Cursor.
Jev enters as "typed primitives" with confidence gates: only decisive answers print as stated fact, and uncertain ones show [verify] or get withheld entirely — a meaningfully more careful design than presenting every model output as settled. Jev usage is hard-capped by a --jev-budget flag, default $0.25 per run, and a free path using DuckDuckGo alone requires no paid backend. The README reports concrete, reproducible-sounding numbers: 81 unit tests, a 6ms cold start for --help, a local audit of 4 files in about 0.2 seconds, a live crawl of 6–10 pages in 3–11 seconds depending on the server, and score repeatability described as "99 stable; findings ±2 from timing rules only" across three crawls. Large sites are sampled at a default cap of 50 pages, and the author is explicit that heuristic rules — like em-dash density as an "AI slop" marker — reflect editorial convention, not confirmed ranking signals, and that the tool doesn't predict rankings or traffic. Installable via cargo install jev-seo, MIT-licensed, 39 stars as of September 2026.
AgriciDaniel/jev-seo: crawl-to-report, with page-level Jev judgments
jev-seo (AgriciDaniel) (listing) crawls a site starting from a single homepage URL and produces PDF, Excel and Markdown reports, evaluated against 52 rules tied to Google Search Central guidance plus Core Web Vitals measurements. Jev's role here is page-level: it answers 13 page questions and 5 site-level questions per page — covering things like page type, search intent and content helpfulness — with probability scoring, and results are flagged when confidence drops below a decisiveness threshold, echoing the same gate-uncertain-answers pattern as AkashPriyadarshii's tool.
The README reports standard-mode cost of about $0.01 per site, with an optional --full mode adding $0.30 for rankings and backlink data, crawl time of "about a minute," and Jev score consistency of 0.03 average variance across repeated runs — plus a claim that rule facts and crawl details were "all verified; word counts within 5%." Large sites are capped at 60 pages with sampling, and the author is upfront that the tool doesn't track rankings over time or predict traffic, and that PageSpeed scores vary run to run with field data only available for high-traffic sites. It needs Python 3.10+, Pango for PDF rendering, and optional Playwright/poppler for JS rendering and PDF processing. MIT-licensed, with bundled fonts under SIL Open Font License; 45 stars as of September 2026.
Reading GEO scores from either tool correctly
Both jev-seo projects gate uncertain Jev answers rather than printing every model output as fact — a genuinely good practice, worth expecting from any GEO tool going forward. But a "citation-readiness" or "content helpfulness" score from either one is still a rubric applied by a model, not a measurement of whether ChatGPT, Perplexity or Google AI Overviews have actually cited the page. Treat both scores the way you'd treat any other proxy metric: directionally useful for finding gaps, not a substitute for checking real citation logs.
What to do next
- Confirm which jev-seo you're citing before quoting a rule count or star figure — the two share a name but nothing else.
- Set a Jev budget cap (AkashPriyadarshii's tool has one built in) before running either SEO tool against a large site.
- Cross-check jev-social's captured evidence against the source post or profile rather than trusting the summary alone — it's a research aid, not a verified record.
- Read How to Evaluate Jev Projects Honestly before adding any GEO score to a report you'll show a client, and see Jev Guardrails for how other projects handle uncertain answers.