Context stuffing
Cramming as much text as possible into the prompt and hoping the model sorts it out.
Think of it like
Like dumping every file on your desk in front of a colleague and saying “the answer’s in there somewhere.”
Example
Instead of retrieving the right three pages, a team pastes the entire 90-page manual into context — and answers get slower and vaguer.
How it actually works
Big context windows tempt you to skip retrieval and just include everything. It sometimes works, but it costs tokens, adds latency, and runs into “lost in the middle,” where models attend poorly to material buried between the ends. Curated, relevant context usually beats a firehose.
For product teams
A tempting shortcut that quietly raises cost and lowers quality; selective beats exhaustive.
For engineers
Filling the window with unfiltered content instead of retrieval; degrades via cost, latency, and mid-context attention loss.
Related
- Lost in the middle — The failure it triggers.
- RAG — The disciplined alternative.
- Token limit — The budget it burns.
Read anything AI without the jargon
Look up any term in plain English, or save terms as you read with the free Chrome extension.
Open DecoderAdd to Chrome