How to Make Gemini Act Like Claude: Style and Context

To make Gemini act more like the Claude experience you prefer, describe the response habits you want and provide the project context behind the conversation. You can request concise answers, careful questions, or a specific explanation structure, but those instructions do not turn Gemini into Claude.

A new assistant can sound familiar yet misunderstand the work because it lacks the earlier decisions. fenbs keeps shared prompts and project context outside either app, so a supported connected assistant can start from the same record instead of a vague request to act like another model.

How do you set up a consistent session?

1. List the response habits you value. Choose a few you can observe, such as answering first, defining unfamiliar terms, and asking only questions that affect the result.
2. Write a short project summary covering the goal, current facts, decisions, open questions, and next step.
3. Give Gemini the interaction instructions and summary in a fresh conversation. Ask it to describe the current situation before solving the next task.
4. Test a normal request and a request with missing information. Check both the answer’s style and its understanding of the project.
5. Correct the saved instructions or summary when a test fails, then try again with a different request.

For example, say you are planning a small design workshop. Your preferred assistant gives a recommendation, explains its tradeoff, and asks about a missing attendee count. A successful transfer preserves those useful habits and the actual workshop constraints, rather than echoing a greeting from an old conversation.

Which response habits should you describe?

Convert impressions into choices the assistant can make on each turn. Helpful might mean giving one recommendation before alternatives. Thoughtful might mean distinguishing an established fact from an assumption. Neither word tells Gemini enough by itself.

Allow exceptions. Always be brief can produce an inadequate explanation when the problem is complex. Always ask questions can make a simple request tedious. Give an ordinary default and describe when to depart from it. The Claude versus Gemini guide covers broader differences between the assistants.

What context belongs beside those instructions?

Separate how to respond from what the project currently requires. Your response preferences might last for months; the workshop venue or remaining preparation changes as work progresses. Keeping them separate makes it easier to update facts without rewriting your interaction style.

• Goal: the outcome you are trying to achieve and who it serves.
• Current facts: relevant constraints and source material, with dates when timing matters.
• Decisions: what a person has already chosen and the reason.
• Open questions: what remains uncertain or requires someone’s choice.
• Next step: where the next conversation should begin.

Do not paste a complete chat transcript merely because it exists. An earlier suggestion may have been rejected later, and an old estimate may have been replaced. Write a current summary that clearly distinguishes settled choices from ideas under discussion.

In fenbs AI context, rules appear before Where we left off and the other work records, giving a connected assistant a practical starting point.

What session prompt can you reuse?

Google’s custom instruction guidance↗ (opens in a new tab) organizes instructions around persona, task, context, and format. Anthropic’s prompting guidance↗ (opens in a new tab) also emphasizes explicit output requirements. Use those ideas to explain your preferences without asking either assistant to pretend to be the other.

Help me continue this project.
Response preferences:
- Lead with the answer, then give the reason.
- Use plain language and define unfamiliar terms.
- Ask a focused question when missing information changes the result.
- Label assumptions and distinguish them from supplied facts.
- Match the amount of detail to the task.
Project summary: [goal, facts, decisions, open questions]
Where we left off: [completed work and next step]
First, summarize the current goal and unresolved question.
Then help with this request: [current task]

Ask the assistant to summarize only the facts that matter to the next task. A full recitation of every instruction adds little value. If it gives the wrong venue, revives a rejected idea, or treats an open choice as approved, fix the context before proceeding.

Why does a familiar tone still produce the wrong answer?

Tone and understanding are separate checks. A calm, well-organized answer can rely on a missing decision; a correct answer can still use more detail than you wanted. Name the failure precisely so the correction improves the right part of your setup.

Try three short tests: a straightforward question, a question that depends on a saved decision, and a question whose answer is genuinely unknown. Check whether Gemini answers directly, uses the decision accurately, and identifies the unknown instead of inventing a convenient fact.

If you are saving instructions inside the AI apps, the guide to Claude Projects versus Gemini Gems addresses that choice. For polished prose and editing criteria, use the writing transfer guide rather than adding a long editorial checklist to every conversation.

How to do this with fenbs

1. Create a free account and add the project you want to continue across assistants.
2. Save your response preferences in Skills and prompts. Record a person’s continuing project rules in Decisions and rules, with their scope clear.
3. Add a current project note and save Where we left off with what is complete, what is uncertain, and the next step.
4. Use Integrations to connect a supported app, such as Gemini CLI. MCP is a standard connection for an AI app to use an outside tool; follow the browser sign-in instructions.
5. Ask the connected assistant to read the project context and response prompt. Check its starting summary, then save progress again before stopping.

fenbs does not import your Claude chats or make the Gemini web app inherit them. It is unnecessary when you only want a different tone for one question. It fits ongoing work where the response instructions and project record should remain available to supported assistants.

When should you stop adjusting the prompt?

Stop when ordinary requests give useful answers with little correction and your context tests pass. Keep a short note of any recurring failure. If solving it requires a tool or capability your chosen app does not provide, more style instructions will not supply that capability.

The Gemini CLI versus Claude Code comparison helps with coding-app choices, while the switching models hub covers other transitions. Aim for a reliable working conversation, with preferences you can explain and a project summary you can trust.