Jev SEO Audit: What an 85-Cent Audit Got Right and Wrong

Jev gave me a useful first pass across four websites for 85 cents, but several confident-looking findings were wrong. I checked its top recommendations against the actual pages and Google Search Console before changing anything.

What is Jev, and how can it audit a website?

Jev is a text-only AI model from TypeSafe. It does not write a free-form report. Instead, software asks it fixed questions and receives a choice, a score or a yes-or-no probability. The jev-seo project on GitHub combines those judgements with rule-based checks, Google's speed data and optional keyword data from DataForSEO.

You provide a website address. The tool checks pages and produces a PDF summary, an Excel list of recommended fixes and a plain-text report. When I used it in September 2026, the project was only days old and at version 0.1.1. Treat the output as a starting point, especially if you are working with a young site.

What did the 85-cent test actually find?

I audited four of my websites. The external services used 85 cents of credit in total, so I paid nothing extra beyond my Claude subscription. That is the cost of my test, not a promised price for every website. Page count, optional ranking data and service pricing can change it.

One useful finding came from a travel guide. Jev scored its content highly but gave its trust signals a low score because readers could not easily see who wrote it or when it was updated. An author box and a visible update date are practical fixes I can check myself.

The tool also surfaced a genuinely slow mobile page. Its rule-based checks and Google's PageSpeed results were the parts I trusted most. I could trace each of those findings back to something observable.

Where did the automated audit get it wrong?

The headline score hid several bad recommendations. It counted a 630-word homepage as 46 words because it read only the first of five content cards. It then made that supposed lack of content its top fix.

It labelled deliberately short email signup pages unhelpful, suggested a local-business label for an online course platform, and produced irrelevant keyword ideas from brand names. One PDF build failed when a site had no ranking data. Claude fixed that error, but the report would have been incomplete if I had accepted it at face value.

The ranking report said one site had zero search visibility. Google Search Console showed real impressions and clicks for that site. DataForSEO's ranking sample is an estimate; Search Console shows what Google recorded for my property. When those two disagree, I start with Search Console.

How should you check an AI SEO audit before making changes?

Open the top three recommended pages yourself. Check whether the tool read the whole page and whether the proposed fix fits that page's purpose. Then compare traffic and search claims with Search Console, and speed claims with Google's PageSpeed data.

Keep a simple decision list: confirmed issue, needs checking, or false positive. This matters more than the letter grade. A technically tidy new site can still have little search traffic, while an older site with an untidy page can already receive clicks.

For my four sites, the useful next steps were specific technical fixes, pages that match real searches, and stronger evidence of who created the content. None followed automatically from an A or B score.

Is a Jev SEO audit worth doing?

If Google search is meant to bring customers to your site, a low-cost audit can give you a workable first list. It is especially useful when you have time to verify the suggestions yourself. If nearly all your clients come from referrals, events or LinkedIn, your next afternoon may be better spent there.

I did the technical part with Claude rather than working in a terminal. I supplied the project link, handled the account decisions and checked the final findings. Never paste API keys into a chat or a public file.

The full Substack edition includes the visual breakdown, the exact questions Jev asks and my subscriber prompt sheet. Start with one confirmed finding and fix that before chasing a score.