I gave Jev three SEO jobs. It couldn’t write a paragraph. For once, that was the selling point.
TypeSafe’s model returns probabilities for fixed choices. Here’s what happened in my September tests.
Internal links: cheap suggestions, human judgement
3,182 decisions cost $0.12. From 147 proposed links, I approved 109 and rejected 38. A shortlist, confidence threshold and sensible anchor rules made the difference. Related doesn’t automatically mean useful.
Cannibalization: spot the problem
Across 64 overlapping searches, Jev flagged 26 where a different page looked better suited. Cost: $0.003. Only three proposed fixes reached 70% confidence. I’m keeping the redirect button away from the robot.
ChatGPT visibility: mentioned isn’t recommended
Of 40 answers, my site appeared in 11 but was recommended in just two. Mention detection matched a plain text search on all 40. That doesn’t independently validate every recommendation judgement.
The useful bit
Keep questions narrow. Let code handle maths and dates, Jev classify meaning, and a writing model handle prose. Human review stays.
Read my original Jev for SEO breakdown for the methodology, costs and three free Claude Code skills.