Large language models and humans share the same cognitive bias: the first and the last information carry the most weight, while the middle fades. In LLMs this is called lost in the middle, in psychology the primacy and recency effects. For pitches, design the opening as a frame, the ending as an anchor, and compress the middle.
Give a Large Language Model a very long document.
Even with a 100k+ token context window, something predictable happens:
- Early instructions receive disproportionate weight.
- The most recent prompt receives disproportionate weight.
- Information buried in the middle weakens.
The tokens are still there.
Their influence decays.
In LLM research, this is often described as “lost in the middle.”
As context grows, meaning in the center of long documents erodes – what some call context rot.
Now compare this to human cognition.
Psychology has documented this pattern for decades:
- Primacy Effect – what comes first shapes perception.
- Recency Effect – what comes last dominates recall.
In a 30-minute pitch, investors remember the opening narrative.
They remember the final takeaway.
Slide 14 rarely survives.
This is not accidental.
Both human cognition and machine cognition operate under the same structural constraints:
- Limited attention
- Competitive signal weighting
- Increasing noise as input length grows
As input expands, signal competes with signal.
Edges become anchors.
The implication for communication is direct.
Most presentations fail in the middle.
Founders add detail to increase certainty.
Executives extend decks to increase credibility.
Cognitive load rises. Retention drops
The structural lesson
1. Design the opening as a framing device.
2. Design the ending as a memory anchor.
3. Compress the middle until only essential signal remains.
Length does not increase persuasion.
Structure increases persuasion.
Humans.
Investors.
Boards.
Large Language Models.
Across biological and artificial species alike, attention has edges.
Design for them.
Frequently asked questions
What does lost in the middle mean for LLMs?
In long documents, early instructions and the most recent prompt receive disproportionate weight, while information in the middle weakens. The tokens are still there, but their influence decays.
What are the primacy and recency effects?
The primacy effect means what comes first shapes perception. The recency effect means what comes last dominates recall.
Why do most presentations fail in the middle?
Founders add detail to increase certainty and executives extend decks to increase credibility. Cognitive load rises and retention drops.
How should I structure a pitch for better retention?
Design the opening as a framing device, design the ending as a memory anchor, and compress the middle until only essential signal remains.
