How PriceSniff works

Price-tracking sites ask you to trust their numbers, so you deserve to know exactly how the numbers are made. This page is the full methodology — nothing on the site works differently than described here.

Where the prices come from

We collect product listings through the official product APIs of our affiliate-partner retailers, on a schedule — every listing is refreshed at least twice a day, within each program's rate limits. Each collection run stores a raw snapshot of what the API returned, and every observed price is appended to that offer's history with its timestamp.

History is append-only. We never edit, smooth, or backfill it. If we didn't observe a price, the chart has a gap.
Only partner retailers are tracked. Products with no partner offer aren't in the catalog at all — see the about page for why. Walmart is our only fully live source today; check any Walmart price on the Walmart price tracker.
Prices can lag the store. A price shown here is the most recent one we observed, not a live quote — the retailer may have changed it since. Click through to see the current price.
When a listing disappears from a retailer, we mark the offer delisted and keep its history rather than pretending it never existed.

The 0–100 product score

Every scored product shows one number and a written rationale. The number is a weighted blend of three components — two computed by fixed formulas that cannot hallucinate, one contributed by an AI model:

Deal quality — 40%. Where the current best price sits inside that product's own 90-day range: at the 90-day low scores full marks, at the high scores zero. No comparison to other products, no "list price" theater — only the product's own recorded history. With no meaningful history yet (new or flat-priced products), this component is neutral.
Data completeness — 20%. How much we actually know: brand, model number, category, an image, normalized attributes, and whether a second store confirms the product. Cross-store confirmation is the strongest signal we have that our data is right.
Value assessment — 40%. An AI model's judgment of the product's value, grounded only in the attributes we collected — it is never given the ability to invent specs or prices. It also writes the rationale you can read by clicking "Why this rating?".

Affiliate commissions play no part in any component. There is no mechanism — human or automated — for a retailer to influence a score.

Matching the same product across stores

When two retailers sell the same item, we merge them into one product page with combined history. Candidates are found by deterministic signals (model numbers, brands, normalized titles), then an AI model confirms or rejects each match. Merges are reversible, and uncertain cases go to human review instead of being guessed. A "N stores" badge marks merged products.

Where AI is involved — and where it isn't

AI models categorize products, normalize attributes, verify cross-store matches, contribute the value-assessment score component, and write the short product summaries. All of it is grounded in data we collected, and AI-written text is labeled as such on the page. AI is never involved in recording prices, drawing charts, or the deal-quality and completeness math — those are fixed code.

What we don't do

No invented prices, specs, or availability — every fact traces to a collected record.
No paid placements, no sponsored rankings, no advertising.
No copying of retailer prose, images, or customer reviews.
No sale of personal data; accounts hold an OAuth identity and your watchlist, nothing more — see the privacy policy.