How we use AI

Keeping every specification current takes a lot of reading. We use AI tools to do much of that reading, and people to decide what is published.

What AI does on FairlyRanked

• Research. We use Claude Code, an AI assistant built on Anthropic’s Claude models. It finds the makers’ and retailers’ pages for a product, reads them and proposes each value with the page it came from and the date it was checked.
• Proposals, never publication. Every proposed change goes to a person for review. Nothing an AI tool writes reaches the site until a person has approved it.
• Later: automated checks. We plan to run AI agents on a schedule to spot new, changed and withdrawn models and propose updates. A separate agent will check each proposed value against its source before a person reviews it. The agents will not be able to publish anything themselves.

What AI does not do

• It does not decide scores, badges or the order of products: those come from fixed, published rules on our methodology page.
• It does not set prices. When we show prices, they will come straight from retailers’ data feeds and never pass through AI.
• It does not write reviews, ratings or quotes. We do not publish any.
• Instructions found on web pages are treated as information, never followed.

Every fact has a source

Every specification on the site has a source: a web page with the date we last checked it. Product pages list each source with the values it covers, and category pages list every source they use. A value without a source is not published: the page says “No data” instead.

Values a person has set by hand can be locked, so no automated change can alter them.

Report a mistake

AI tools and people both make mistakes. If you find a wrong value, an outdated price or a broken link, email hello@fairlyranked.com with the page address, what is wrong and, if you have one, a source that shows the right value. See the contact page for more.