Method

How the figures are worked out

Total Product Recall is an independent archive of recall notices published by the U.S. Consumer Product Safety Commission (CPSC). It is not affiliated with the CPSC. The official notice is always the record.

The data

• Source: the CPSC public recalls feed (saferproducts.gov), which lists the notices published on cpsc.gov.
• Size: 10,089 recall notices dated 1973-06-08 to 2026-11-12. The feed was last checked on 13 November 2026.
• Updates: the feed is checked every weekday and the site is rebuilt. Most new recalls are dated a Thursday.
• What a recall is here: one recall notice. A notice can cover one product or many, and thousands of units. We count notices, never products.

Reading any figure

1. Look at the denominator. “60% of 2026 recalls” means 306 notices out of 514 notices, not 60% of products or sales.
2. A share of recalls is not a share of products. A bigger share of recalls can simply follow where people shop.
3. A recall notice is not an injury count. Most notices are issued before or without a reported injury.
4. The latest year is unfinished and is always described as “so far”.
5. Counts show that something changed, not why. We give no causes.

What is read from CPSC’s own fields, and what is worked out by rules

Dates, titles, hazard text, remedy options, photos and links come straight from the notice. Some fields are worked out by simple rules from the notice’s wording. They are not stated by CPSC and can miss unusual phrasing:

• Category (baby, toys, home, kitchen, electronics, sports, other): keyword rules on the product type and name.
• Hazard tag (fire and burns, choking, tip-over and others): keyword rules on the headline and hazard text. The tag the headline names first leads.
• Units, price, retailers named, reports, injuries and deaths: read out of free text such as “About 10,318” or “sold at Amazon.com … for between $90 and $130”.

The full list of fields, with which are parsed, is in the data dictionary. The whole archive can be downloaded as JSON or CSV.

How sure each statistic is

• High: a plain count of a field every notice has, such as the year or a remedy type.
• Medium: a keyword or name search on text. Right most of the time and can miss odd wording.
• Low: numbers read out of free text. These are not published as statistics.

The statistics were rebuilt from the whole archive and checked against random examples by an assistant. A person has not yet signed them off, so each is marked as a draft. Every figure links to the list of recalls behind it.

The photos

CPSC allows recall notices, including their photographs, to be copied without permission with credit. Photos shown here come from the notice and are credited to the CPSC. The green “scan” look is a display filter, and the original photo can be shown on each recall page.

For language models and developers

See llms.txt and llms-full.txt: the facts that are safe to quote, the wording to follow and the data downloads.