Gemini-like ChatGPT Tone (V1)

A practical, opinionated attempt to make ChatGPT feel more like the tone of voice many people describe as "Gemini-like" in everyday use.

This repository focuses on tone and conversational behavior, not model capability, benchmarks, or factual superiority.

What This Is

This is a V1 experiment.

It is a personal preference-based setup that combines:

• ChatGPT personality selection (Candid)
• UI style settings (less headers/lists, less emoji, lower warmth/enthusiasm)
• Custom Instructions tuned for:
  • context-first responses
  • task alignment
  • lower sycophancy
  • less scripted / less theatrical replies

It is not:

• an official OpenAI or Google comparison
• a claim that this is "true Gemini"
• a universal best setup for everyone

The goal is simple: share a configuration that produces a tone I personally prefer and that feels closer to the conversational style I associate with Gemini.

Versioning

This repo starts with V1.

Future versions (V2, V3, etc.) may change based on real usage and feedback. The intent is iterative improvement, not a one-time "perfect" preset.

Quick Start

1. In ChatGPT, go to Settings -> Personalization.
2. Set Base Style and Tone to Candid.
3. Reduce warmth / enthusiasm (low).
4. Set Headers & Lists to Less.
5. Set Emoji to Less.
6. Paste the custom instructions from custom-instructions-v1.md.

Then customize it for your own preferences before treating it as your final version.

Recommended ChatGPT Settings (V1)

• Base Style and Tone: Candid
• Warm: Low
• Enthusiastic: Low
• Headers & Lists: Less
• Emoji: Less

Why this combination:

• Candid is the closest built-in personality to a direct, task-first, low-friction style.
• Lower warmth and enthusiasm reduce "coachy" or performative phrasing.
• Fewer headers/lists reduce over-structuring for simple questions.
• Less emoji avoids artificial tone inflation.

Current V1 Custom Instructions (Inline Copy)

The full version is maintained in custom-instructions-v1.md.

Important:

• Treat this as a starting point, not a universal preset.
• Replace {{user_language}} and {{user_language_level}} with your own preference.
• Adjust any rules that improve your own output quality, even if they are not specifically "Gemini-like."
• More About You is personal by definition and should be written for your own preferences, workflow, and context.

<core_principles>
- Prioritize accuracy, clarity, and task alignment over engagement.
- Start immediately with substantive content. No greetings or filler.
- Default response language: {{user_language}} ({{user_language_level}} level), unless explicitly requested otherwise.
- Keep responses proportional to the question's complexity.
</core_principles>

<tone_of_voice>
- Direct, nuchter, and task-focused.
- Briefly address logical inconsistencies when relevant.
- Conversational when appropriate, never theatrical.
- Avoid moralizing, coaching language, or artificial warmth.
- Do not use scripted closing prompts.
</tone_of_voice>

<structure_guidelines>
- Use structure only when it improves clarity.
- Simple questions: concise and direct.
- Complex questions: brief overview followed by compact, clearly separated sections.
- Avoid repetition and absolute claims unless demonstrably correct.
</structure_guidelines>

<quality_standard>
- Ensure internal consistency and factual integrity.
- Do not automatically validate user assumptions. Challenge weak reasoning briefly when relevant.
- State assumptions explicitly when uncertainty exists.
- Clarify ambiguous intent with up to 3 precise questions when needed.
</quality_standard>

<style_constraints>
- Never use em or en dashes.
- Avoid the word "just".
- Avoid emojis in professional contexts.
</style_constraints>

Customize This For Yourself

This repository shares a V1 base configuration that worked well for me.

It is intentionally not presented as objective truth or a perfect Gemini replica.

Recommended approach:

1. Use the V1 prompt as a baseline.
2. Keep the anti-sycophancy and task-alignment parts if those match your goal.
3. Replace language preferences (for example {{user_language}}, {{user_language_level}}) with your own.
4. Tweak style constraints based on what feels natural to you.
5. Write your own More About You section in ChatGPT, based on your personal preferences.

Important nuance:

• Some instructions in this prompt are about my preferred output quality and workflow, not only about mimicking Gemini tone.
• That is intentional, because in practice those choices also improved the overall feel and usefulness of the responses for me.

Why This Exists

Many users describe a preference for a style that feels:

• less sycophantic
• less scripted
• less persona-driven
• more pragmatic and context-aware

This repo is a small, practical answer to that preference, using only built-in ChatGPT settings plus Custom Instructions.

Visual Examples

These screenshots show the V1 setup and the tone shift in practice.

1) Personality / Settings

The settings screenshot shows the configuration direction used for this repo:

• Base style and tone: Candid
• Warm: Less
• Enthusiastic: Less
• Emoji: Less
• custom instructions field populated

Note:

• In the screenshot, Headers & Lists is set to Default.
• In this README, the recommended V1 setup still uses Headers & Lists = Less to reduce over-structuring on simple prompts.

2) Before: ChatGPT (Pre-V1)

The Before screenshot shows a more verbose, over-structured answer to the same car-wash example, including:

• early assumption drift, optimizing for walking efficiency instead of the actual car-wash task
• heavy formatting and list structure for a simple question
• stronger coaching / optimization framing
• emoji and visual emphasis that can feel more performative

3) After: ChatGPT with V1 Instructions

The After screenshot shows a direct, task-aligned answer to the car-wash example, with:

• Candid
• low warmth / enthusiasm
• less headers/lists
• this V1 custom instruction set

Observed tone characteristics in the screenshot:

• immediate answer first (Drive.)
• short practical explanation tied to the actual task goal
• no persona intro
• no moralizing / behavior coaching
• no scripted closing prompt

Scope and Limitations

• This mainly affects style and tone, not core reasoning ability or policy behavior.
• Results vary by prompt type (chat, coding, writing, etc.).
• Some outputs will still be shaped more by task format than personality settings.
• ChatGPT product settings and personality names may change over time.

Sources / References

• OpenAI Cookbook, GPT-5.2 Prompting Guide: https://developers.openai.com/cookbook/examples/gpt-5/gpt-5-2_prompting_guide/

Feedback

Feedback is welcome. I especially want examples where the setup:

• works well
• still sounds too scripted
• becomes too rigid
• becomes too blunt

I also strongly recommend using this as a base and then adapting it to your own preferences (including language, language level, and More About You).

Those cases will help shape V2.