Knowledge decay
Knowledge decay is when AI platforms keep repeating facts about you that used to be true, like an old price or a product you've retired.
Knowledge decay is when AI platforms keep saying things about your business that were true once and aren't any more. An old price, a product you stopped selling, a shop you closed, a slogan you dropped two years ago.
It happens because a model learns from the web as it was when it was trained, and some of the pages it reads later haven't been updated either. A customer who believes the old fact may never find out it was wrong.
Where the old facts come from
There are two places. One is the model itself: what it learned in training stays the same until the next model comes out. The other is the web: when the platform searches, it may read a review from three years ago, an old directory listing or a news story about your launch price.
For example, Harrow & Pine raised its loaf price from £3.50 to £4.20 last spring. A food blog from 2023 still lists the old price, and a directory still shows the old opening hours. An AI answer that reads both tells people the bread costs £3.50 and the shop opens at 7, when it now opens at 8.
How to fix it
Make the current facts easy to find on your own site, in plain text, with a date where it helps. Then go after the outside pages the platforms read: ask the blog to update its post, and correct your directory listings. When the wrong fact comes from the model alone, the platforms have their own ways to report an answer.
In AnswerOrbit
Claims checks what the answers say about you against your facts every day. Inaccuracy drivers shows the pages behind each wrong claim, or the platform's own knowledge when no page was used. From a claim, Copy evidence gives you everything to send to whoever runs the page.