LLMInference & InterfacesUpdated 2026.04.28

Lost in the Middle

Also known asLITM긴 문서 중간 누락

In one line

Lost in the Middle is the well-documented effect where information sitting in the middle of a long context is used much less reliably than what sits at the beginning or end — the empirical basis for 'put the important stuff at the front or back'.

Going deeper

Lost in the Middle is a 2023 finding from a Stanford-led study: LLMs use information at the start and end of a long context far more reliably than information in the middle, producing a 'U-shaped' accuracy curve. The size of the effect varies by model and version, but the shape is consistent.

Two practical takeaways for marketers. First, when feeding long internal documents into an AI assistant, push key conclusions, policies and numbers to the beginning or end — never bury them in the middle. Second, the same logic applies to GEO content: do not hide your headline answer in a middle paragraph.

Teams hit this constantly in RAG, document summarisation and long-PDF analysis. The standard workarounds are reordering chunks, lifting critical facts to the top, and re-emphasising them at the close.

Sources

Related terms

How does your brand show up in AI answers?

Villion measures how your brand appears across ChatGPT, Perplexity and AI Overviews, then automates the work that lifts citation rate and share of voice.

Get a free audit