Your chats · your scope · your analysis

What do your AI chats reveal about how you work?

Give your agent only the conversations you want included. It can surface recurring themes, open loops, context switching, and blockers — without turning them into a diagnosis.

Copy the prompt
01 · Choose the scope

You decide what the agent may see

A

This chat only

The fastest start. The agent reviews only the current conversation.

B

Selected chats

You choose individual conversations, projects, or a specific date range.

C

Several AI tools

You deliberately add imported or uploaded chats from ChatGPT, Codex, Claude, Gemini, Cursor, or other tools.

Important: An agent can only analyze material already in its context or content you explicitly upload, import, or share through a connected source. It should never silently search “all chats.”

02 · Insights

What the analysis can surface

  • Goals and open loopsWhat you repeatedly start, continue, or leave unfinished.
  • Recurring blockersWhere tools, missing context, or decisions slow you down.
  • Context switchingWhich projects and topics you frequently move between.
  • Work and communication patternsHow you frame tasks, make decisions, and improve outputs.
  • Change over timeWhich patterns persist and where counterexamples or gaps remain.
03 · Start now

Give this prompt to your agent

Analyze how I use AI assistants to identify recurring work and usage patterns.

Before you begin:
1. List the chats, files, or connected sources you can actually access. Do not claim access you do not have.
2. Ask which scope I approve:
   A) only this current chat,
   B) chats I select in this assistant,
   C) additional chats I import or upload from ChatGPT/Codex, Claude, Gemini, Cursor, or other tools.
3. Ask for the date range, analysis goal, and any topics I want excluded.
4. Remind me to remove third-party names, secrets, and credentials before uploading anything.
5. Start only after my explicit confirmation.

Clearly separate observation, interpretation, and uncertainty. Look for recurring goals, topics, open loops, context switching, repeated blockers, decision and communication patterns, and changes over time. Support claims with short excerpts or chat/date references. Include counterexamples and data gaps.

Do not diagnose me or infer sensitive attributes that are not clearly supported by the approved material. End with the most important patterns and five small, concrete improvements. If an evidence-based self-screening on medtests.de might be useful, suggest no more than one appropriate screening and ask again before starting it.

04 · Guardrails

The analysis stays within your approval

  • Confirm source, date range, and exclusions first.
  • Keep observations, interpretations, and uncertainty separate.
  • No diagnosis and no unsupported sensitive inferences.
  • medtests.de receives no chat content through this prompt.
  • Remove third-party names, secrets, and credentials before uploading.

Optional next step

When a pattern deserves a closer look

The agent may suggest no more than one relevant, evidence-based self-screening. It starts only after fresh consent. Results are for orientation and are not a diagnosis.

Browse self-screeningsWebMCP for agents