Position bias
Position bias is the tendency of language models to use information at the start or end of what they read better than information in the middle. For a web page, it means the facts you put first, such as an answer box, are the facts an AI tool is most likely to use.
Why it matters
Researchers at Stanford and elsewhere tested how language models use long inputs. Performance was often highest when the relevant text sat at the beginning or the end, and dropped when the model had to use text from the middle. That held even for models built for long contexts.1
People skim in a similar way. Nielsen Norman Group's 2018 eye-tracking study found 57% of viewing time on the first screen of a page.2 An answer near the top serves both readers.
In practice
- Open each page with a short, bold answer to its main question.
- Put the facts buyers compare, such as price logic and area, high on the page.
- Keep one point per paragraph, so a key fact never hides mid-block.
- Reorder a page to bring key facts up, rather than repeating them.
- End long pages with a short FAQ that restates the key answers.
Go deeper
How to structure a first screen is in homepage design. The answer-box pattern is in answer engine optimization.