How to Train ChatGPT on Your Writing Style

To train ChatGPT on your writing style, collect two or three pieces of your real writing, extract the specific patterns that make them sound like you, and write those patterns as instructions ChatGPT can follow. Once you have a style brief that works, store it somewhere you can load it quickly—either in ChatGPT's memory and custom instructions, or in a shared tool like fenbs that any connected AI can read at session start.

The mistake most people make is telling ChatGPT "write like me" without any samples or rules. ChatGPT then produces something generic or guesses at a style. The work is in the extraction: turning an instinct about your writing into instructions specific enough for an AI to follow.

Collect the right writing samples

Pick two or three pieces you are satisfied with and that represent the kind of writing you want ChatGPT to produce. An email style brief does not help you write blog posts, and vice versa. Use samples from the same context as your intended output.

Longer samples work better. A 400-word piece gives ChatGPT enough text to detect patterns across multiple sentences and paragraphs. A single paragraph is too short to reveal consistent habits. Choose pieces written under normal conditions, not ones where unusual constraints changed how you wrote.

Extract the patterns from your samples

Read your samples and note what repeats. Look at these dimensions:

• Sentence length: short and direct, long and flowing, or a deliberate mix?
• Paragraph length: two sentences, five sentences, variable?
• Opening structure: do you start with the main point or build toward it?
• Vocabulary: plain words or precise technical ones? Contractions? Words you use often?
• Tone: formal, conversational, warm, dry?
• Things you avoid: passive voice, jargon, hedge words, exclamation marks?

Write down what appears in both samples. Single-sample quirks are less reliable than patterns that show up consistently. What you find across two pieces is much more likely to be your real style than what you remember about your writing.

Write your style rules

Turn each pattern into a rule ChatGPT can test against. Vague instructions produce inconsistent results. "Sound natural" is not useful. "Use contractions; keep sentences under 18 words; start each paragraph with the main point" is.

Aim for six to twelve rules that describe the specific differences between your writing and something generic. Put the most important ones first. Add a "what to avoid" section—words, structures or habits you dislike—since negative constraints are often easier for a model to follow reliably.

Pair the rules with your samples in the prompt: "Here are two pieces written in the style to match. The rules below describe it explicitly. Follow both." The combination of example and description anchors the target more firmly than either alone.

Test, then revise one rule at a time

Give ChatGPT a real writing task with your style brief active. Read the output and compare it to your samples. Mark every place it drifts—wrong vocabulary, too long, too formal—and trace each problem to the missing or vague rule that caused it.

Change one rule per revision. If you adjust several at once, you cannot tell which one fixed the problem. After two or three revision cycles, your rule set usually stabilizes. The corrections become rare and predictable rather than constant.

What to do when ChatGPT reverts to a generic voice

Even with Custom Instructions active, ChatGPT sometimes produces generic output. The most common reasons: the style rule is too vague, the instruction is being overridden by something in the message, or the conversation has become long enough that the instruction carries less weight.

Start a new conversation and check whether the problem disappears. If it does, the issue is a long-context drift, not the rules themselves. Add a short reminder at the end of your style instructions: "Apply these rules to every response, including in long conversations." If the problem persists in a fresh conversation, the rules need to be more specific.

Make your style rules persistent

ChatGPT has two places for persistent instructions: Custom Instructions (profile-level preferences that load for every new conversation) and memory (facts ChatGPT has saved about you over past sessions). Both can hold your style brief, and they work together.

Custom Instructions are more reliable for a precise style brief: you write them explicitly, you can read them back and you can update them when your rules improve. Add your style brief to the "How would you like ChatGPT to respond?" field in Custom Instructions.

If you want the same voice rules to work in Claude, Cursor or other AI apps, an external record is more portable. fenbs stores your style brief in Skills and prompts, and any connected assistant reads it at the start of each session over MCP. For the Claude-specific version of this process, see how to train Claude to write like you.

How to do this with fenbs

1. Create a free account and add a project for your writing or content work.
2. Paste your finished style brief into Skills and prompts. Name it by context: "Blog post voice," "Client email style," "Report tone."
3. Add each standing rule to Decisions and rules—state the rule, the reason and when it applies.
4. Connect your AI apps to fenbs from Integrations. At the start of each session, ask the assistant to read the project context and your style prompt before drafting.
5. When a revision improves a rule, update it in fenbs. Every connected assistant picks it up from the next session.