How to Make Gemini Write Like Claude (2026)

If you have used both, you know the feeling. You give Claude a paragraph of context and get back something that reads like a person wrote it. You give Gemini the same thing and get a competent block of text that could have come from anywhere.

That gap is real, and it is fixable. It is also not what most people should be aiming at.

What actually differs between the two

The interesting part is that the difference is not intelligence. Ask both models a hard question and you will get comparable reasoning. The difference shows up in the shape of the output, and it comes from how each model was tuned.

Claude's defaults, in practice:

• Longer paragraphs that connect ideas instead of listing them
• A willingness to state a position without stacking qualifiers
• Fewer headers and bullets unless you ask for structure
• Transitions that carry an argument forward

Gemini's defaults, in practice:

• A pull toward summarizing rather than arguing
• Structure by reflex: headers, bullets, bold labels
• More hedging ("it's worth noting", "generally speaking")
• Shorter paragraphs that each restate their own topic

Read those lists again and notice something: every single item is a behavior you can describe. None of them requires access to Anthropic's weights. That is why "make Gemini write like Claude" is a solvable problem, and it is also why the solution has nothing to do with mentioning Claude.

Fix 1: describe the behavior, never the model

The instruction that fails is "write like Claude". Gemini has no reliable internal representation of how a competitor's model writes, so it does what any model does with a vague instruction: it guesses, then drifts back to its defaults within two exchanges.

The instruction that works names the mechanics:

Write in connected paragraphs of four to six sentences. Do not use bullet points unless I ask for a list. State your conclusion in the first sentence of each section, then support it. Do not open sentences with "it's worth noting", "generally speaking", or "in conclusion". If a claim is uncertain, say so once in plain words rather than hedging throughout.

That is the same request, made legible. Every clause is checkable, which is exactly what makes it followable.

Fix 2: put the rules where they persist

A prompt fixes one response. Gemini forgets it by the next conversation, which is why people end up re-pasting the same instructions every morning.

Two places hold rules persistently:

Gems. Create one, put your style rules in the instructions field, and use it as your default writing chat. Everything inside that Gem inherits the rules. This is the closest equivalent to a Claude Project.

Saved info. Under your Gemini settings you can store standing facts and preferences that apply across chats. Keep this short and factual, and let the Gem carry the detailed style rules. Splitting them this way avoids the two sets of instructions contradicting each other.

The failure mode here is worth naming: people write thirty lines of style rules, save them, and get inconsistent results anyway. Almost always the rules contain a contradiction ("be concise" plus "write four to six sentence paragraphs") or an unmeasurable adjective ("be engaging"). Strip anything a stranger could not verify by reading the output.

Fix 3: show, then constrain

Rules cover what you can articulate. They miss everything you do without noticing, and that is a larger share of your style than most people expect.

Give Gemini two or three samples of writing you want it to match, and pair them with a single instruction: match the sentence rhythm and paragraph length of these samples, not their topic. Without that last clause, models tend to copy subject matter instead of structure.

This gets you noticeably closer. It also runs into a ceiling, and the ceiling is the interesting part.

The ceiling, and why the question is usually the wrong one

Do all three fixes and Gemini will stop sounding like stock Gemini. It will start sounding like a competent, structured, slightly formal writer. Which is to say: it will sound like a different AI default rather than no AI default.

That is fine if you are drafting internal notes. It is a problem the moment your name goes on the output, because you have not removed the tell. You have swapped it.

Look at the search behavior around this and the intent gives itself away. People search "how to make Gemini write like Claude" and, in the same breath, "how to train Gemini to write like me". Those are not two questions. They are one question asked twice, and the second version is the honest one. Claude is standing in for "the version of this that does not sound machine-made".

If that is the actual goal, then the target should be your writing, not another model's.

What targeting your own voice requires

Copying yourself is harder than copying Claude, for a specific reason: you cannot see your own patterns. You write on autopilot. Ask anyone to describe their sentence rhythm, their formality baseline, or how their register shifts between a client email and a Slack reply, and you will get a guess that does not survive contact with their sent-mail folder.

That is the work a Style Profile does. Instead of you introspecting, it analyzes your actual writing samples and extracts the measurable patterns: sentence architecture, vocabulary register, structural habits, and the shifts you make between audiences. The output is a written specification of how you write, which is precisely the format every one of these models follows best.

The practical payoff is portability. The same specification goes into a Gemini Gem, a Claude Project, and ChatGPT's custom instructions. All three then aim at the same target, so you stop getting three different voices depending on which tab you happen to have open. The cross-model consistency people were trying to buy by imitating Claude comes free, because the target no longer lives inside any one model.

The short version

Make Gemini write like Claude by describing Claude's behaviors as explicit, checkable rules, storing them in a Gem so they persist, and showing samples for what rules cannot capture. That closes most of the gap.

Then ask whether you wanted the gap closed in that direction. If the writing carries your name, the better target was always your own voice, and the same three mechanics work for it.

My Writing Twin builds a Style Profile from your actual writing samples, then deploys it across Gemini, Claude, and ChatGPT, so every draft aims at your voice instead of a model's defaults.

Get Your Style Profile →

Or read the setup guide for making Gemini write like you first.