GOOGLE'S LATEST STUDY REVEALS HOW LLMS RECALL FACTS

Summary from the TLDR Marketing newsletter

Getting recalled correctly by AI models does not require more content - it requires making existing content easier to retrieve. Advanced models store 95% to 98% of the facts they are trained on but reliably recall only 26% to 34% of them. Recall breaks down most on facts asked backward from how they were learned, or on facts that are simply rare. Auditing where AI tools already mention or skip a brand's key facts shows exactly which messages need restating in both directions.