How it works

Every verdict follows the same transparent pipeline. Here’s exactly what happens between pasting a link and getting an answer.

1. Collect the product

Pick from the catalog, paste a URL, or enter details manually. For URLs we read only permitted public metadata (JSON-LD, Open Graph, standard meta tags). We never bypass anti-bot protections or site terms, and we block private/internal addresses. If a page can’t be read, you’re routed to a quick manual-entry form and can edit anything we did find.

2. Understand your needs

We ask category-specific questions — budget, country, use case, must-haves, deal-breakers, and how much price, performance, reliability, battery, display, and support each matter.

3. Score deterministically

A weighted engine scores five dimensions: requirement fit (30%), value for money (25%), quality (20%), reliability (15%), and support (10%). Your priority sliders re-weight these. Missing data lowers confidence — it doesn’t silently penalize a product.

4. Explain with AI (guarded)

The AI writes the reasoning, evidence, pros/cons, and alternatives around those fixed numbers. Product text and reviews are treated as untrusted data — the model never follows instructions hidden inside a page, and cites evidence for important claims. Output is validated against a strict schema before you ever see it. When no AI provider is configured, a deterministic demo analyzer produces the same structured result.

5. Show the verdict

You get a Buy, Consider, Wait, or Skip with scores, a fair target price, deal-breakers, better alternatives, evidence, and “what could change this verdict” — plus the confidence level and anything we’re missing.