Five habits that define heavy AI users

Executive overview

Most people type short queries into a single AI tool. Power users operate differently: they talk, delegate to multiple models, and use AI before major decisions.

Five habits separate heavy AI users from everyone else. Each one shifts how information flows between human and machine.

The core insight: using AI heavily is itself a skill — one that requires deliberate habits, not just access.

Talking instead of typing

• Dictation removes the bottleneck of typing and reduces self-editing mid-thought.
• Speaking to AI enables a flow state that gets more context into the model faster.
• Dictation is a skill — structured verbal communication improves with practice.
• Tools: Whisper Flow and Super Whisper work system-wide; ChatGPT has it built in.
• Advanced voice mode enables back-and-forth conversation, not just input.
• Use cases: ideating on a commute, getting live answers during social conversations.
• Note: ChatGPT Plus degrades voice model quality after ~90 minutes per session.

Using AI as a panel of advisors

Two distinct modes:

Consensus across models

• For high-stakes research, run the same prompt across five or more models simultaneously.
• Each model has different strengths; one will catch what others miss.
• Synthesise the outputs — often using a sixth AI — into a single report.
• For design tasks, compare visual outputs across models to find the one that matches your taste.

Specialists by task

• Broad research quickly: GPT-4.1 with extended reasoning.
• Deep, holistic research: Claude (broader source coverage).
• X/Twitter-specific research: Grok (direct platform access).
• Complex PDF extraction (handwriting, charts, visuals): Gemini.
• Image editing with consistent characters/objects across scenes: Imagen (Google).
• Model strengths shift week to week — keep updating your mental map.

Taking AI input on major decisions

• Provide full context, not keyword queries — include constraints, budget, location, preferences.
• Low-stakes example: ask for three chair options with trade-offs stated in plain English, filtered by local stock.
• High-stakes example: upload a lease, ask the AI to flag abnormalities, then use it to draft amendment requests to the landlord.
• AI writes the revised wording in the contract's own tone, reducing friction for both parties.
• AI informs the decision; the human makes it.

Using AI to write and improve prompts

• Always write a base prompt covering three things: what you want, why you want it, and any additional context.
• Feed the base prompt into a prompt improver (available natively in both OpenAI and Claude).
• The improver injects best practices for that specific model: structure, delimiters, output format, verbosity, stopping conditions.
• Result: high-quality outputs without needing to be a prompt engineer.

Selectively not using AI

• Heavy AI use can degrade the ability to focus for extended periods.
• Reading long-form content via AI summaries is efficient but erodes deep reading as a skill.
• Power users deliberately choose when to read versus when to delegate to AI.
• Protecting the focus muscle is a productivity differentiator — not a rejection of AI.