A comparison series that automates article generation while keeping fixed evaluation axes and scoring, testing how much trust automated comparisons can keep.
What AI handles
- Scores each product on fixed axes and writes the comparison article.
What a person decides
- The axes, the scoring method, and the list of products allowed in, all set beforehand. Articles with products outside the list are not published.
Hypothesis
Automated comparison articles can stay trustworthy if the axes and scoring rationale are fixed.
Learnings
- Generated articles contained product codes that do not exist and totals that disagreed with the item scores. Before publishing, products are now checked against an approved list and totals are recomputed with a fixed formula.
- Generation stopped in August 2026 when the API usage limit was reached, and exports kept failing in September. The scheduled run is now off.
Stack
site-engine / Claude API