Methodology

How We Source Prices

Prices come from three retailers, pulled through each retailer's official product API rather than scraped from HTML: Amazon via the Amazon Creators API, Best Buy via the Best Buy Products API, and Newegg via the Rakuten Advertising API. A background job syncs every tracked listing twice a day, at 6:00 AM and 6:00 PM UTC.

Each listing is keyed to a stable retailer identifier — the Amazon ASIN, or the equivalent SKU for Best Buy and Newegg — so a resync updates the existing row instead of creating a duplicate. When a retailer stops returning a listing, or reports it out of stock, we remove it: Amazon listings are deleted the sync after they disappear from the API response, while Best Buy and Newegg listings get a short grace period (to absorb transient API hiccups) before being swept. Either way, a product that no one can actually buy does not stay ranked.

Listing pages themselves refresh on a 5-minute cache, and individual product rows display a “checked Xh ago” label driven by that listing's last sync timestamp — so you can see exactly how current a given price is, not just trust that the page is fresh.

How Price-Per-Unit Rankings Work

Every product gets a single normalized number: current price divided by capacity or yield. Storage drives, RAM, and flash media are ranked by $/GB. TVs reuse the same underlying field for $/inch (screen diagonal in place of capacity). Printer ink and toner are ranked by $/page, using the manufacturer's stated page yield.

Every listing table on the site — the homepage, category pages, brand and capacity filters, comparison pages, and product detail pages — sorts on that field ascending. There is no separate ranking algorithm that factors in commission rate, retailer relationship, or anything else: cheapest per unit is listed first, full stop.

How Our Articles Are Produced and Fact-Checked

Buying-guide and comparison articles on this site are drafted by an LLM writer, not a human writer. We think that's important to say plainly rather than imply otherwise. What makes the output trustworthy is not who typed it, but what has to survive before it's allowed to publish.

Before an article goes live, it passes through a sequence of automated checks:

• Price-fidelity audit: every product-and-price mention in the draft is cross-checked against the live product database. Any figure that disagrees with the closest database match by more than 10% triggers an automatic rewrite pass.
• Independent claim judge: a separate model, from a different model family than the writer, reads the article section by section against an evidence block built from research sources and the tracked-listing database, and flags claims it cannot support. The gate is fail-closed: if any section goes unjudged, or any claim is flagged high-severity after two automated rewrite passes, the article does not publish.
• Link check: every affiliate URL in the draft is probed before publish; a broken affiliate link blocks publication outright.
• Banned-phrase and sanity checks: drafts are also screened for unverifiable superlatives, guarantee-style language, and capacity or unit figures that don't make physical sense.

If a claim can't be substantiated after the automated rewrite passes, the article is held back rather than published with the unverified claim intact. There is no manual editorial sign-off step in this pipeline — the fact-checking is the gate.

How We Make Money

BuyPerUnit is an affiliate site. When you click through to a retailer and make a qualifying purchase, we may earn a commission — at no additional cost to you. This is disclosed in the site footer on every page.

Commission rates differ by retailer and product, but they play no role in how products are ordered: as described above, every ranking is computed purely from price per unit. A higher-commission listing gets no boost, and a lower-commission listing gets no penalty.

Corrections

Automated sourcing and automated fact-checking still make mistakes — a retailer API can misreport stock, a sync can land between price changes, a model can misjudge a claim. Product and listing pages carry an on-page reporting tool for flagging a wrong price, a dead link, a listing that shows in stock but isn't, or incorrect product details. Reports are reviewed and confirmed errors are corrected in the underlying data, which then flows through on the next sync.