· Visible thinking / preamble · single-tell paste-in
Stop ChatGPT from restating your question.
LLMs default to long preambles: restating your question, announcing what they're going to do, explaining their approach before doing it. Paste the prompt below and the model starts with the actual answer.
· the paste-in
Apply this rule to every reply you write for me. It removes a specific pattern that flags content as AI-written. Rule: Skip the preamble. Don't restate the question. Don't explain what you're about to do. Don't announce your plan unless explicitly asked. Start with the answer or the first concrete step. Self-review: Read the first paragraph. If it restates the question or describes what's about to happen, delete it. Before sending any reply, run through the self-review. If the rule was violated, rewrite the affected sentence from scratch rather than character-swapping.
What this tell is
Preamble is the boilerplate at the start of LLM replies that doesn't move the answer forward: 'Let me break this down for you...', 'I'll address each of your questions in turn...', 'Before I answer, let me clarify what you're asking...', 'That's an interesting problem. The first thing to consider is...'. Sometimes it shows up as a numbered outline of the response that the model is about to give in full anyway.
Why removing it matters
Preamble wastes tokens (real money on the API, slower replies in the UI) and reader attention. For programmatic use (function-calling, structured output, code assistants) it actively breaks parsing because consumers expect the answer at byte zero, not after 80 tokens of throat-clearing. Removing preamble makes the model's first sentence land on the substance.
Per-model notes
ChatGPT GPT-5.5 has the strongest preamble habit because the reasoning models added explicit 'thinking' phases that often leak into responses. Claude 4.7 is cleaner by default but still introduces tool calls with verbose announcements ('I'll search the web for that...'). Gemini 3.1 Pro varies by query type.
· want the full set?
This page covers visible thinking / preamble only. To kill 20 other 2026 AI tells (em dashes, the word "comprehensive", AI vocabulary cluster, templated transitions, sycophancy, hallucination, rule-of-three lists, over-bolding, and more) in one paste-in prompt, plus a Claude Code rules block and a Node.js lint script: use the AI Tell Killer main tool.
FAQ
Will this break Claude Code or other coding assistants?
No. The rule applies to prose preamble (restating the request, announcing the plan). Coding assistants still produce structured outputs as expected; they just skip the 'I'll start by looking at...' sentences that used to precede the actual code. If you use a coding assistant that NEEDS to announce its plan first (some agent loops require this for orchestration), drop this rule before generating in those contexts.
What about Claude / GPT reasoning modes that show their thinking?
Reasoning modes (Claude extended thinking, GPT o-series reasoning) have a dedicated 'thinking' channel that's separate from the visible reply. This rule applies to the visible reply, not the thinking. The model can still reason internally; it just doesn't expose 'Let me think about this...' in the user-facing answer.
Related: the full AI Tell Killer (all 21 tells, three output formats), the AI Output Linter (paste AI text and scan for tells in real time), and the em dashes guide (deep-dive on the single most-checked tell).