How I Prompt: Nov 13, 2025

Methodology

How It Works

Data Collection

Reads JSONL conversation logs from Claude Code (~/.claude/projects/) and parses the conversations.json export from Claude.ai.

Message Extraction

Filters to human-authored prompts only. Excludes system commands, tool outputs, and empty messages. Preserves timestamps for temporal analysis.

Metric Computation

Runs regex pattern matching across all prompts to count politeness markers, backtrack phrases, questions, and command patterns.

Persona Classification

Computes two composite scores (Engagement and Politeness) and maps them onto a 2×2 matrix to classify your prompting style.

All Metrics

Volume

Total Prompts — your human messages

Conversations — unique chat sessions

Words Typed — total word count

Avg Words/Prompt — verbosity measure

Conversation

Avg Turns — messages per conversation

Longest Session — marathon conversation

Quick Asks — 1-3 turn convos

Working Sessions — 4-10 turn convos

Deep Dives — 11+ turn convos

Response Ratio — Claude words ÷ your words

Temporal

Peak Hour — most active hour

Peak Day — most active weekday

Night Owl % — 11pm-4am activity

Style

Politeness Index — "please", "thanks", "sorry" per 100

Backtrack Index — "actually", "wait" per 100

Question Rate — % ending with "?"

Command Rate — % starting with action verbs

Signature

"You're absolutely right" — Claude's agreement count

Composite Scores:

Engagement = (Question + Backtrack) ÷ 2

Politeness = Index − (Command × 0.5)

The 4 Personas

Based on where you fall on the Engagement (x-axis) and Politeness (y-axis) scales:

High Politeness

Low Politeness

High
Engage

Collaborator

You ask questions politely. AI is your partner, not your tool.

Explorer

You question, iterate, and dig deeper. Thinking out loud.

Low
Engage

Efficient

Polite but focused. You know what you want and ask nicely.

Pragmatist

Balanced and practical. No frills, just results.

Thresholds: Engagement ≥ 12, Politeness Score ≥ 4.5

The Numbers

4,887

prompts sent

1,091

conversations

416K

words typed

14.2

avg turns

2.5x

response ratio

455

longest session

223

deep dives

"That's a lot of 'actually, wait...'"

Prompt Style

Politeness

8.4

per 100 prompts

"please"

352

"sorry"

"thanks"

Says "please" but rarely "thanks"

Backtrack

7.6

per 100 prompts

"actually"

299

"wait"

Course-corrects frequently

Questions

17.1%

of prompts

834 questions of 4,887

Sometimes asks, mostly tells

Commands

7.3%

action verbs

359 commands fix, add...

Prefers context over commands

"The robots will remember this."

How You Use AI

740

Quick Asks

1-3 turns

128

Working Sessions

4-10 turns

223

Deep Dives

11+ turns

Conversation Depth

14.2

avg turns

Longest marathon: 455 turns

Response Ratio

2.5x

Claude's output

For every word you type, Claude writes 2.5

"More collaborator than commander."

Diagnosis

Based on your metrics, you are...

The Collaborator

You ask questions politely. AI is your partner, not your tool.

InquisitiveCourteousPartnership-oriented

8.4

Politeness

7.6

Backtrack

17.1%

Questions

7.3%

Commands

"Efficient. Almost suspiciously so."

HOW I PROMPT

NOV 13, 2025 – FEB 08, 2026 SUMMARY

Prompts4,887

Conversations1091

Words Typed416K

Avg Turns14.2

Longest455 turns

Deep Dives223

Peak Hour11pm

Peak DaySunday

Night Owl26.0%

"You're absolutely right"304x

PERSONA: THE COLLABORATOR

Inquisitive • Courteous • Partnership-oriented

Generated Feb 08, 2026

eeshans.com

Personal project · MIT License · Non-commercial

Source