Jev SEO internal-link analysis: from relevance to human review
Understand the Internal Link Dealer demo, Jev’s role in relevance decisions, a small reproduction plan, and why link suggestions are not ranking guarantees.
What does the video show?
Borja Fatás’s original post shows the Internal Link Dealer SEO interface, with Jev and Opus displayed side by side. This video is part of our manually reviewed case library.
The original video loads when you scroll here
The creator mentions processing 586 pages. That is a creator-reported scale, not our benchmark. The demo does not provide complete source code, an evaluation set, or subsequent GSC results. It does not establish superior accuracy, guaranteed lower costs, or ranking improvements from adding links.
Jev judges; the application does the rest
An internal-link tool needs page text and existing links before identifying useful relationships. Given the interface and Jev’s text-input capabilities, a plausible design uses Jev to evaluate candidate page relationships. The workflow below is a reproduction proposal, not a reconstruction of the creator’s private implementation.
- Ordinary code collects and cleans pages, extracts links, and builds candidate pairs.
- Jev can judge topic relevance and whether a destination helps the source page’s reader.
- Ordinary code validates URLs, excludes existing links, and presents suggestions.
- An editor or reviewed publishing workflow chooses a natural placement and publishes.
Jev does not inspect this video or automatically edit a site when it returns a high score. The official model documentation specifies text input: provide extracted titles, summaries, or passages first.
Reproduce the idea with 10 pages
- Choose 10 public pages you are allowed to process. Store URL, title, summary, and existing internal links. Exclude credentials, personal information, and private admin pages.
- Prefilter candidates by topic or keywords rather than immediately comparing every page with every other page.
- Build a text state containing one source and destination: for example, “Jev starter credit” and “Jev cost calculator,” each with its title and summary.
- Define a Noul question: “Does the destination directly help the source page’s reader estimate API costs?” Adapt the request in our existing API tutorial with this state and question. This is not the creator’s prompt.
- Compare results with human-labeled relevant and irrelevant pairs before choosing thresholds. Output suggestions only; probability is not a search ranking signal.
- Have an editor check that the destination is accessible, the link is not redundant, and the placement is natural. Publish a small batch, then monitor GSC trends.
On this site, a useful example is linking a paragraph explaining token costs in the starter-credit article to the pricing and cost calculator, rather than repeating the same keyword link in every paragraph.
How can you tell whether it helps?
Start with human acceptance rate: relevance, redundancy, and appropriate placement. Record published links and dates, then observe impressions, clicks, and crawl changes for the affected pages. Content updates, competitors, and search demand also affect GSC, so one increase is not proof that internal links caused it.
Short titles and summaries can omit important meaning; evaluate Chinese content separately. Larger sites produce more candidate combinations. Estimate costs from actual input tokens and requests, not page count alone. Official pricing and limits can change; consult our cost guide.
Further reading
Explore content-classification videos, Jev’s limitations, and the official API reference. This article uses the public demo and official documentation checked September 26, 2026. We have not run the creator’s tool or verified its ranking impact.