CLN Articles · Prompt Engineering
The Prompt Paradox
Author Vadim Shchetinkin Published February 2026 Version CLN v1.2
Why your 500-word prompt gets worse results than 5 CLN lines.
01 / More words, less clarity

You write 500 words of context, caveats, and instructions. The AI responds with 2000 words of confident prose that misses the point. You try again. More words. Worse results. This is the Prompt Paradox: prose introduces ambiguity faster than it adds context.

02 / The Analysis CLN
= paradox: increasing prompt verbosity decreases output precision
= root cause: prose introduces ambiguity faster than it adds context
Prose Prompt
- adjectives create interpretation branches (creative? how creative?)
- nested clauses bury the actual instruction
- context and command are mixed — AI can't tell which is which
! 500 tokens of prose ≈ 200 tokens of signal + 300 tokens of noise
CLN Prompt
+ operators separate context (=) from instructions (.)
+ one fact per line = zero ambiguity per fact
+ hierarchy shows priority: Level 1 = must do, Level 2 = nice to have
+ 80 tokens of CLN ≈ 80 tokens of signal + 0 noise
Result
= CLN prompts are 2–4× shorter AND more precise
* the best prompt is one the AI cannot misinterpret
03 / Side by Side Example
Prose prompt 127 tokens
Please write a comprehensive analysis of remote work productivity. Consider both the advantages and disadvantages. Be thorough but not too long. Include statistics if possible. Think about impact on individual contributors and team dynamics. Format it nicely.
CLN prompt 48 tokens
= TASK: Analyze remote work productivity = focus: individual tasks + team collaboration . include statistics with sources + advantages (with data) - disadvantages (with data) ! risks commonly overlooked = OUTPUT: CLN format, max 30 lines
48 tokens vs 127. The CLN prompt is 62% shorter, costs 60% less, and produces more focused output.
← Tools Semantic Density →