← The Clair Blog

    Why does my AI sound nothing like me?

    Your AI sounds robotic because it's never been told how you write. Here's what's actually going wrong, and the one setting that fixes it.

    Because it's never met you.

    Every time you open a new chat, your AI starts from zero. It doesn't know your job title, your industry, the tone you use with clients versus the tone you use with your team, or the fact that you'd rather delete the whole email than start it with "I hope this finds you well." It doesn't know any of that because nobody told it, and the result is output that sounds like it was written by a polite stranger applying for a job at a council.

    This isn't a prompting problem. You don't need better prompts, more detailed instructions, or a 47-step framework from someone on LinkedIn. You need to do the thing that most people skip entirely: actually tell your AI who you are.

    The setting most people have never touched

    Every major AI tool has a place where you can set permanent instructions that apply to every conversation. ChatGPT calls it Custom Instructions (Settings, Personalisation, Custom Instructions). Claude calls it Styles and Profile (Settings, General). The names vary, the function is the same: you tell the AI who you are, what you do, how you write, and what you never want to see, and it remembers that across every chat from that point on. Not sure which to set up first? Here's how to choose between them.

    Most people have never opened this. The field has been sitting there blank since the day they signed up, which means every single chat starts with the AI guessing who it's talking to, and it guesses wrong, every time, in the same predictable ways.

    Why default AI output sounds the way it does

    Without any context about you, your AI defaults to what it was trained on: a blend of internet writing, corporate communication, and American English that sounds like a slightly nervous LinkedIn post. That's why the output is always too formal, too long, too eager to please, and stuffed with words like "delve" and "leverage" and "I'd be happy to help."

    It's not that the AI is bad at writing, it's that it's writing for a person that doesn't exist. A generic professional with no preferences, no opinions, and no personality. Of course it sounds robotic, it's writing for a robot.

    What changes when you set it up

    The difference is genuinely dramatic. The same prompt that produces a stiff, three-paragraph corporate email with a "please don't hesitate to reach out" sign-off will produce something that actually sounds like you wrote it, once the AI knows your tone, your role, and your pet hates.

    People who've done this describe the shift in the same way: "it went from sounding like a stranger to sounding like me on a good day." That's the bar. Not perfect, not a replacement for your judgement, but a first draft you can actually send without rewriting the whole thing.

    What to actually put in there

    The details matter. Telling your AI "be professional but friendly" gives it almost nothing to work with, which is like briefing a new hire by saying "just be good at your job." It needs specifics:

    • Your role. Your job title and what you actually do day-to-day
    • Your audience. Who you write for, clients, colleagues, or senior stakeholders
    • Your tone. Direct, warm, no waffle, no exclamation marks
    • Your bans. The words and phrases you never want to see
    • Your format. Preferences like short paragraphs or no bullet points

    The more specific you are, the less editing you do afterwards. People who spend 10 minutes getting this right typically say it changes every conversation after that point.

    It's not just about writing

    Custom instructions affect everything your AI does, not just emails. Summaries come back in your voice, brainstorming sessions start from your actual context instead of generic suggestions, and research is framed around your industry and role. Meeting prep reflects the way you actually run meetings. The AI stops being a general-purpose chatbot and starts being something closer to an assistant who's been working with you for months.

    The thing nobody says out loud

    Most people haven't set this up because nobody told them it existed, or because the blank field felt intimidating and they didn't know what to write. That's not a skills gap, it's an onboarding gap. You wouldn't expect a new hire to perform well on day one with no brief, no context, and no idea what the company does, and your AI is in exactly that position right now.

    If you want the practical follow-up, this guide walks through what to put in the custom instructions field.

    Common questions

    Does custom instructions work on free ChatGPT and Claude?

    Yes. Custom instructions and profile settings are available on free accounts for both ChatGPT and Claude. You don't need a paid subscription to set this up.

    How long does it take to set up custom instructions?

    About 10-20 minutes. The time is in thinking about your tone and preferences, not in the technical steps. The technical part is just filling in a text box in your AI tool's settings.

    Will my AI remember custom instructions across every chat?

    Yes. Once you set custom instructions, they apply to every new conversation automatically. You set it once and it carries forward. You can update them whenever you want.

    Can I set up custom instructions if I use more than one AI tool?

    Each tool has its own version of this setting, so you would need to set it up in each one. The information is the same. Your role, tone, preferences, and rules, so adapting it for a second tool takes a few minutes.

    The AI Starter Kit

    Set up ChatGPT and Claude properly. Once.

    The workflows, rules and habits that hand the boring recurring work to your AI and give you your evenings back.

    Get the Kit →