![\"\"](\"https:\/\/railsware.com\/blog\/wp-content\/uploads\/2026\/09\/fixing-AI-content_image-1024x538.png\")
<\/figure>\n\n\n\nTL;DR<\/strong><\/h2>\n\n\n\n\n- Key approach to writing with AI without sounding like AI: feed the model with your own material first, not a smarter prompt.\u00a0<\/li>\n\n\n\n
- Four habits, not a system: capture your own voice, mine your own notes for topics, draft and argue, run a loop light enough to survive past week three.<\/li>\n\n\n\n
- Voice extraction: upload writing you actually recognize as yours to an LLM, then run one prompt to get a reusable voice brief tied to your actual patterns, not generic advice.<\/li>\n\n\n\n
- Topic mining: build a knowledge base (with meeting notes, recurring client questions, and the odd report that caught your eye) that turns into a pattern generator with one prompt, and any pattern turns into several post angles with a second.<\/li>\n\n\n\n
- Drafting: never stop at version 1.0. Argue with the machine and teach it to write better.<\/li>\n\n\n\n
- Guardrails matter: no sensitive data in any notebook, ever.<\/li>\n<\/ul>\n\n\n\n
***<\/h2>\n\n\n\n
Decades ago, when I was an editor in business media, I read lots of op-ed pieces by entrepreneurs and C-level people. Most were polished, well-structured, stripped of emotions and personality. You could swap the author\u2019s name and hardly anybody would notice. The ones that stuck in memory had a raw nerve in them, a real person shining through words, recognizable style and strong opinions.<\/p>\n\n\n\n
Surely, back then, most of those texts were written by ghostwriters and comms professionals. Now, this role is outsourced to AI. But the problem remains: a lot of content we see online is generic, bland, and almost sterile. Not bad, exactly. Just nobody’s. AI assistants just scaled it up. <\/p>\n\n\n\n
That’s the gap our Comms team spent months testing tools against: how to actually speed up content production without flattening the person writing it. We\u2019ve come up with a fairly easy setup everyone can embed into their AI-powered workflows. So here\u2019s what actually held up, because most advice on this topic is either “AI is ruining writing” or “here are 47 prompts,” and neither one is true or useful.<\/p>\n\n\n\n
The actual fix<\/h2>\n\n\n\n
An AI can only sound like you if you’ve given it something of yours to sound like. Not instructions about your tone. Actual writing. The rest \u2014 the clever prompt phrasing, the persona instructions, the “write in a conversational but authoritative voice” \u2014 is doing maybe five percent of the work people think it’s doing.<\/p>\n\n\n\n
I built the workflow around four things instead: a way to capture your voice, a way to never run dry on topics, a drafting-and-editing step, and a loop simple enough to actually keep doing.<\/p>\n\n\n\n
1. Feed it your own words before you ask for anything<\/h3>\n\n\n\n
Before touching any drafting prompt, create a notebook in Gemini Notebook<\/a>, a Claude Project<\/a>, whatever holds documents for you. Upload writing you actually recognize as yours, for instance:<\/p>\n\n\n\n\n- Old posts you’re proud of.\u00a0<\/li>\n\n\n\n
- Emails you\u2019ve written yourself\u00a0<\/li>\n\n\n\n
- Slack\/any other messenger discussion where you sounded like yourself, not like you were performing professionalism<\/li>\n\n\n\n
- Transcript from meetings where you spoke at lengths, etc.<\/li>\n<\/ul>\n\n\n\n
Once again, the more you give to the machine, the better it will understand your writing persona. <\/p>\n\n\n\n
Then run one prompt that could sound like this. (And yes, it looks like more prompt engineering than the thesis above admits to caring about, and it is \u2014 but notice it’s still not doing anything without your uploaded material. Take away the sources, and this prompt produces the same generic nothing as any other):<\/p>\n\n\n\n\n note<\/span>\n \n Role<\/i>:<\/strong> You are an expert linguist, copywriter, brand voice strategist, and \r\nAI prompt engineer.<\/span>\r\n\r\nTask<\/i><\/strong>:<\/strong> Based only on the sources I've uploaded, analyze how I write \u2014 not what \r\nI write about. Identify: sentence rhythm and length, words and phrases I repeat,\r\nwords I'd never use, how I typically open and close a message, where I'm blunt\r\nversus diplomatic, and any tics (em dashes, lists, self-corrections). \r\nBased on your analysis, create a comprehensive set of system instructions that \r\nI can use to program a custom AI assistant (a Gem or a Claude skill) to write \r\nexactly like me.<\/span>\r\n\r\nAnalysis Criteria<\/i>:<\/strong> Please evaluate the documents and extract rules for the \r\nfollowing:<\/span>\r\n\r\nTone and Energy<\/span>:<\/b> Is the writing formal, playful, highly empathetic, direct, or \r\nauthoritative? What is the emotional resonance?<\/span>\r\n\r\nVocabulary and Lexicon<\/span>:<\/b> What specific phrases, transition words, industry jargon,\r\nor colloquialisms do I frequently use? Are there words I strictly avoid?<\/span>\r\n<\/span>Syntax and Structure: Do I use short, punchy sentences or longer, flowing\r\nexplanations? How do I typically open and close my messages?<\/span>\r\n<\/span>Formatting Habits: Do I heavily rely on bullet points, bold text for emphasis, \r\nspecific emojis, or distinct paragraph spacing?<\/span>\r\n\r\nOutput Format: Do not just describe my voice. Output a highly structured, \r\nready-to-use \u201cVoice brief\u201d that I can copy and paste into a Gem. \r\nUse clear headings within the prompt (e.g., <\/span>Your Persona, Tone Guidelines,\r\nVocabulary Rules, Formatting Instructions<\/span><\/i>) telling the AI exactly how to \r\nemulate me without sounding AI-generated.\"<\/span><\/pre>\n <\/div>\n <\/section>\n\n\nThe output is a voice brief that describes your actual writing patterns, not “professional LinkedIn tone.” Paste it into a Gemini Gem or a Claude Project’s custom instructions (or even create a skill) once, and every draft from then on starts from your voice instead of a blank one.<\/p>\n\n\n\n
Two things not to upload, in case it needs saying out loud: anything with sensitive business data, and anything under an NDA. Voice extraction doesn’t need real names or real numbers to work, just your sentences.<\/p>\n\n\n\n
2. Mine what you already have instead of staring at a blank page<\/h3>\n\n\n\n
You don\u2019t run out of ideas because you lack ideas. You run out because the good ones are buried in notes you never open again. That\u2019s why build your knowledge base. Hold a second notebook (call it “Market”, \u201cIntel\u201d, or whatever fits) with meeting transcripts, recurring client questions, retros, and the odd industry report you found interesting. (And later on, you could design a regular news brief to be delivered by AI, so you could pick from there directly). <\/p>\n\n\n\n
Turn it into a topic generator with one prompt like this:<\/p>\n\n\n\n\n note<\/span>\n \n Scan these sources for patterns that show up more than once \u2014 the same objection,\r\nthe same misunderstanding, the same moment of surprise. List 5\u201310 patterns. \r\nFor each, note what triggered it and why it would surprise someone outside my \r\nfield.<\/pre>\n <\/div>\n <\/section>\n\n\nTake one pattern and hand it back to the same AI with a second prompt like:<\/p>\n\n\n\n\n note<\/span>\n \n Here's a pattern I noticed: [paste it<\/em>]. Pitch me 3-5 different post angles in \r\ntwo sentences each \u2014 don't write the post yet.<\/pre>\n <\/div>\n <\/section>\n\n\nThe angles you\u2019ll see depend on the pattern. It might be a contrarian take pushing back on standard advice in your field. Or a personal-story version built around one specific moment. A how-to, if the pattern is really just you explaining the same fix for the third time this month. A data angle, if you’ve actually got numbers behind it. None of these are mandatory, and none of them are the only options \u2014 you’re not committing to write anything yet. You’re just seeing what’s there.<\/p>\n\n\n\n
3. Draft it and argue with it<\/h3>\n\n\n\n
Pick one angle and hand it back to your Voice Gem or skill with a third prompt like:<\/p>\n\n\n\n\n note<\/span>\n \n Using my voice brief and this angle: [paste chosen angle], draft a short post. \r\nKeep it under 150 words. Open with the specific moment, not a summary. \r\nNo emojis, no rhetorical questions, no 'here's the thing' setups, \r\nno three-part lists dressed up as insight.\r\nEnd when the point is made \u2014 don't add a wrap-up sentence. \r\nIf anything sounds like generic LinkedIn voice rather than mine, \r\nflag it instead of smoothing it over.<\/pre>\n <\/div>\n <\/section>\n\n\nWhat comes back is a draft, not a finished post. Read it the way you’d read a colleague’s first pass: something in it is probably off, and your job is to notice what and say so, not to proofread lightly and hit publish. (And yes, you can add emojis to your liking \u2013 don\u2019t outsource this to an AI\ud83e\udd13)<\/p>\n\n\n\n
Most people stop there. Fix one word or a line, post, move on. Don’t. When you catch the tool doing something that isn’t you, tell it what was wrong, in the same chat:<\/p>\n\n\n\n\n note<\/span>\n \n That closing line reads like a generic LinkedIn wrap-up. \r\nI don't write tidy summary sentences \u2014 I stop when the point's made. \r\nAdd this as a rule to my voice brief.<\/span><\/pre>\n <\/div>\n <\/section>\n\n\nA voice brief you write once and never touch again drifts back toward generic on its own \u2014 every model update, every new chat, is a small reset. Correct it enough times, give it more samples and feedback, and it stops being a static document. It starts catching your own tics before you do.<\/p>\n\n\n\n
4. Build a loop, not a system<\/h3>\n\n\n\n
The failure mode isn’t lack of tools. It’s building an elaborate system in week one and abandoning it in week three. The loop that survives is boring on purpose: capture two or three lines after any meeting that surprised you, drop new notes into the notebook once a week, ask for angles, write a draft, argue with the draft instead of politely accepting it, post. <\/p>\n\n\n\n
Once you\u2019ve created the pillars \u2014 the knowledge base and the voice brief exist \u2014 the whole thing takes about twenty minutes a week. Most of that twenty minutes is the capture step, and capture is free. It’s just remembering to jot something down.<\/p>\n\n\n\n
The part that convinced the room<\/h2>\n\n\n\n
The moment that landed hardest in the workshop wasn’t a slide. It was a side-by-side. I asked an AI, cold, to write a short LinkedIn post about the workshop itself. <\/p>\n\n\n\n![\"\"](\"https:\/\/railsware.com\/blog\/wp-content\/uploads\/2026\/09\/image-2-1024x564.png\" "\\"\u0417\u043d\u0456\u043c\u043e\u043a")
<\/figure>\n\n\n\nIt came back with an exclamation point, a “just wrapped up” opening, and literally the \u201crobotic fluff\u201d I wanted to avoid. Then I asked the same question through a Gem carrying my voice brief. The second draft opened with “most people feed an LLM three bullet points and end up sounding like a generic McKinsey slide deck with a pulse” \u2014 which is, frankly, closer to the level of irony I have in my voice (probably, more than I’d like to admit).<\/p>\n\n\n\n![\"\"](\"https:\/\/railsware.com\/blog\/wp-content\/uploads\/2026\/09\/image-3-1024x672.png\")
<\/figure>\n\n\n\nThe output is far from perfect, I admit. But it needs far less rework from my side to be \u201cmine\u201d than the first option.
Same model. Same question. The only variable that changed was whether it had anything of mine to work from.<\/p>\n\n\n\n
What to actually do with this<\/h2>\n\n\n\n
This article has been about helping LLMs understand a personal voice. However, you can apply the same logic to other content streams that have a well-defined tone of voice, such as video scripts, blog posts, media comments, etc. Start with this basic setup, and then grow and update it with more context and feedback. This very approach helped us grow our own \u201cRailsware influencers\u201d with LinkedIn impressions reaching tens of thousands per month.\u00a0\u00a0<\/p>\n\n\n\n
The tools keep changing. Gemini, Claude, whatever comes next \u2014 the underlying move doesn’t. Give the model your actual material before you ask it for your voice back. Skip that step, and you’re right back to competent, forgettable, and nobody’s \u2014 just faster.<\/p>\n\n\n\n
FAQ<\/h2>\n\n\n\nWhat’s the fastest way to start writing with AI without sounding like AI?<\/h3>\nIf you want to try it, the easiest path is not reading the rest of this post twice. Instead, open your LLM right now, upload three to five things you’ve written that you’d stand behind, and run the voice-extraction prompt above. Most people don’t have samples sitting around, so check your Sent folder. Nobody proofreads an email to a colleague the way they proofread a public post, which is exactly why it’s the most honest sample of your writing you have. If you\u2019re not an email person, look through your Slack or any other messenger to find a thread where you discussed some heated topic.\u00a0<\/p>\n<\/div><\/div>\n\n\n\n
How much of my own writing do I need to upload?<\/h3>\nThree to five pieces is enough to start. Adding more source material produces a more accurate voice brief; too little and the output defaults back to generic advice.<\/p>\n<\/div><\/div>\n\n\n\n
Is it safe to upload work material to build a voice brief?<\/h3>\nNever upload sensitive business data or anything under NDA. Voice extraction only needs sentence-level patterns \u2014 it doesn’t require real names, numbers, or client details to work.<\/p>\n<\/div><\/div>\n\n\n\n
Does this method only work for LinkedIn posts?<\/h3>\nNo. The same voice-extraction and drafting process applies to any content with a consistent tone: video scripts, blog posts, media comments, internal memos.<\/p>\n<\/div><\/div>\n\n\n\n
What should I do if an AI draft still sounds generic?<\/h3>\nCorrect it in the same chat rather than editing it silently. Tell the model exactly what read as generic (for example, a “wrap-up” closing line) and ask it to add that correction as a permanent rule to the voice brief, so future drafts avoid the same tic.<\/p>\n<\/div><\/div>\n\n\n\n
\n Additional sources<\/span>\n \n \n \n - \n \n \n How to AI: using Claude Design when you\u2019re not a designer <\/span>\n <\/i>\n <\/a>\n <\/li>\n \n \n
- \n \n \n How to AI: co-designing your work with AI in 5 modes <\/span>\n <\/i>\n <\/a>\n <\/li>\n \n <\/ul>\n<\/div>\n\n\n
\n \n 
<\/div>\n\n \n Article by<\/span>\n \n Stas Sokolov <\/span>\n \n As Comms Lead at Railsware, Stas leverages his 16 years of experience to lead the company\u2019s messaging strategy. A writer at heart and a strategist by trade, he\u2019s obsessed with clarity and brand consistency. His “off-screen” life is fueled by science fiction, long dog walks, and an endless queue of history podcasts. Linkedin<\/a> <\/div>\n \n <\/div>\n<\/section>","protected":false},"excerpt":{"rendered":"Everyone has a hot take on AI and content, often offering a single prompt as a universal fix. Far fewer will show you what actually happens behind the curtain.\u00a0 At Railsware, we’d rather show how to build a system and processes. Prompts can be great. But using prompts on top of a thought-through foundation is…<\/p>\n","protected":false},"author":104,"featured_media":19436,"comment_status":"closed","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"_acf_changed":false,"inline_featured_image":false,"footnotes":""},"categories":[365,738],"tags":[],"coauthors":["Stas Sokolov"],"class_list":["post-19432","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-insights","category-marketing"],"acf":[],"aioseo_notices":[],"aioseo_head":"\n\t\t\n\t\n\t\n\t\n\t\n\t\n\t\t\n\t\t\n\t\t\n\t\t\n\t\t\n\t\t\n\t\t\n\t\t\n\t\t\n\t\t\n\t\t\n\t\t\n\t\t
***<\/h2>\n\n\n\n
Decades ago, when I was an editor in business media, I read lots of op-ed pieces by entrepreneurs and C-level people. Most were polished, well-structured, stripped of emotions and personality. You could swap the author\u2019s name and hardly anybody would notice. The ones that stuck in memory had a raw nerve in them, a real person shining through words, recognizable style and strong opinions.<\/p>\n\n\n\n
Surely, back then, most of those texts were written by ghostwriters and comms professionals. Now, this role is outsourced to AI. But the problem remains: a lot of content we see online is generic, bland, and almost sterile. Not bad, exactly. Just nobody’s. AI assistants just scaled it up. <\/p>\n\n\n\n
That’s the gap our Comms team spent months testing tools against: how to actually speed up content production without flattening the person writing it. We\u2019ve come up with a fairly easy setup everyone can embed into their AI-powered workflows. So here\u2019s what actually held up, because most advice on this topic is either “AI is ruining writing” or “here are 47 prompts,” and neither one is true or useful.<\/p>\n\n\n\n
The actual fix<\/h2>\n\n\n\n
An AI can only sound like you if you’ve given it something of yours to sound like. Not instructions about your tone. Actual writing. The rest \u2014 the clever prompt phrasing, the persona instructions, the “write in a conversational but authoritative voice” \u2014 is doing maybe five percent of the work people think it’s doing.<\/p>\n\n\n\n
I built the workflow around four things instead: a way to capture your voice, a way to never run dry on topics, a drafting-and-editing step, and a loop simple enough to actually keep doing.<\/p>\n\n\n\n
1. Feed it your own words before you ask for anything<\/h3>\n\n\n\n
Before touching any drafting prompt, create a notebook in Gemini Notebook<\/a>, a Claude Project<\/a>, whatever holds documents for you. Upload writing you actually recognize as yours, for instance:<\/p>\n\n\n\n Once again, the more you give to the machine, the better it will understand your writing persona. <\/p>\n\n\n\n Then run one prompt that could sound like this. (And yes, it looks like more prompt engineering than the thesis above admits to caring about, and it is \u2014 but notice it’s still not doing anything without your uploaded material. Take away the sources, and this prompt produces the same generic nothing as any other):<\/p>\n\n\n\n The output is a voice brief that describes your actual writing patterns, not “professional LinkedIn tone.” Paste it into a Gemini Gem or a Claude Project’s custom instructions (or even create a skill) once, and every draft from then on starts from your voice instead of a blank one.<\/p>\n\n\n\n Two things not to upload, in case it needs saying out loud: anything with sensitive business data, and anything under an NDA. Voice extraction doesn’t need real names or real numbers to work, just your sentences.<\/p>\n\n\n\n You don\u2019t run out of ideas because you lack ideas. You run out because the good ones are buried in notes you never open again. That\u2019s why build your knowledge base. Hold a second notebook (call it “Market”, \u201cIntel\u201d, or whatever fits) with meeting transcripts, recurring client questions, retros, and the odd industry report you found interesting. (And later on, you could design a regular news brief to be delivered by AI, so you could pick from there directly). <\/p>\n\n\n\n Turn it into a topic generator with one prompt like this:<\/p>\n\n\n\n Take one pattern and hand it back to the same AI with a second prompt like:<\/p>\n\n\n\n The angles you\u2019ll see depend on the pattern. It might be a contrarian take pushing back on standard advice in your field. Or a personal-story version built around one specific moment. A how-to, if the pattern is really just you explaining the same fix for the third time this month. A data angle, if you’ve actually got numbers behind it. None of these are mandatory, and none of them are the only options \u2014 you’re not committing to write anything yet. You’re just seeing what’s there.<\/p>\n\n\n\n Pick one angle and hand it back to your Voice Gem or skill with a third prompt like:<\/p>\n\n\n\n What comes back is a draft, not a finished post. Read it the way you’d read a colleague’s first pass: something in it is probably off, and your job is to notice what and say so, not to proofread lightly and hit publish. (And yes, you can add emojis to your liking \u2013 don\u2019t outsource this to an AI\ud83e\udd13)<\/p>\n\n\n\n Most people stop there. Fix one word or a line, post, move on. Don’t. When you catch the tool doing something that isn’t you, tell it what was wrong, in the same chat:<\/p>\n\n\n\n A voice brief you write once and never touch again drifts back toward generic on its own \u2014 every model update, every new chat, is a small reset. Correct it enough times, give it more samples and feedback, and it stops being a static document. It starts catching your own tics before you do.<\/p>\n\n\n\n The failure mode isn’t lack of tools. It’s building an elaborate system in week one and abandoning it in week three. The loop that survives is boring on purpose: capture two or three lines after any meeting that surprised you, drop new notes into the notebook once a week, ask for angles, write a draft, argue with the draft instead of politely accepting it, post. <\/p>\n\n\n\n Once you\u2019ve created the pillars \u2014 the knowledge base and the voice brief exist \u2014 the whole thing takes about twenty minutes a week. Most of that twenty minutes is the capture step, and capture is free. It’s just remembering to jot something down.<\/p>\n\n\n\n The moment that landed hardest in the workshop wasn’t a slide. It was a side-by-side. I asked an AI, cold, to write a short LinkedIn post about the workshop itself. <\/p>\n\n\n\n It came back with an exclamation point, a “just wrapped up” opening, and literally the \u201crobotic fluff\u201d I wanted to avoid. Then I asked the same question through a Gem carrying my voice brief. The second draft opened with “most people feed an LLM three bullet points and end up sounding like a generic McKinsey slide deck with a pulse” \u2014 which is, frankly, closer to the level of irony I have in my voice (probably, more than I’d like to admit).<\/p>\n\n\n\n The output is far from perfect, I admit. But it needs far less rework from my side to be \u201cmine\u201d than the first option. This article has been about helping LLMs understand a personal voice. However, you can apply the same logic to other content streams that have a well-defined tone of voice, such as video scripts, blog posts, media comments, etc. Start with this basic setup, and then grow and update it with more context and feedback. This very approach helped us grow our own \u201cRailsware influencers\u201d with LinkedIn impressions reaching tens of thousands per month.\u00a0\u00a0<\/p>\n\n\n\n The tools keep changing. Gemini, Claude, whatever comes next \u2014 the underlying move doesn’t. Give the model your actual material before you ask it for your voice back. Skip that step, and you’re right back to competent, forgettable, and nobody’s \u2014 just faster.<\/p>\n\n\n\n If you want to try it, the easiest path is not reading the rest of this post twice. Instead, open your LLM right now, upload three to five things you’ve written that you’d stand behind, and run the voice-extraction prompt above. Most people don’t have samples sitting around, so check your Sent folder. Nobody proofreads an email to a colleague the way they proofread a public post, which is exactly why it’s the most honest sample of your writing you have. If you\u2019re not an email person, look through your Slack or any other messenger to find a thread where you discussed some heated topic.\u00a0<\/p>\n<\/div><\/div>\n\n\n\n Three to five pieces is enough to start. Adding more source material produces a more accurate voice brief; too little and the output defaults back to generic advice.<\/p>\n<\/div><\/div>\n\n\n\n Never upload sensitive business data or anything under NDA. Voice extraction only needs sentence-level patterns \u2014 it doesn’t require real names, numbers, or client details to work.<\/p>\n<\/div><\/div>\n\n\n\n No. The same voice-extraction and drafting process applies to any content with a consistent tone: video scripts, blog posts, media comments, internal memos.<\/p>\n<\/div><\/div>\n\n\n\n Correct it in the same chat rather than editing it silently. Tell the model exactly what read as generic (for example, a “wrap-up” closing line) and ask it to add that correction as a permanent rule to the voice brief, so future drafts avoid the same tic.<\/p>\n<\/div><\/div>\n\n\n\n Everyone has a hot take on AI and content, often offering a single prompt as a universal fix. Far fewer will show you what actually happens behind the curtain.\u00a0 At Railsware, we’d rather show how to build a system and processes. Prompts can be great. But using prompts on top of a thought-through foundation is…<\/p>\n","protected":false},"author":104,"featured_media":19436,"comment_status":"closed","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"_acf_changed":false,"inline_featured_image":false,"footnotes":""},"categories":[365,738],"tags":[],"coauthors":["Stas Sokolov"],"class_list":["post-19432","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-insights","category-marketing"],"acf":[],"aioseo_notices":[],"aioseo_head":"\n\t\t\n\t\n\t\n\t\n\t\n\t\n\t\t\n\t\t\n\t\t\n\t\t\n\t\t\n\t\t\n\t\t\n\t\t\n\t\t\n\t\t\n\t\t\n\t\t\n\t\t\n
Role<\/i>:<\/strong> You are an expert linguist, copywriter, brand voice strategist, and \r\nAI prompt engineer.<\/span>\r\n\r\nTask<\/i><\/strong>:<\/strong> Based only on the sources I've uploaded, analyze how I write \u2014 not what \r\nI write about. Identify: sentence rhythm and length, words and phrases I repeat,\r\nwords I'd never use, how I typically open and close a message, where I'm blunt\r\nversus diplomatic, and any tics (em dashes, lists, self-corrections). \r\nBased on your analysis, create a comprehensive set of system instructions that \r\nI can use to program a custom AI assistant (a Gem or a Claude skill) to write \r\nexactly like me.<\/span>\r\n\r\nAnalysis Criteria<\/i>:<\/strong> Please evaluate the documents and extract rules for the \r\nfollowing:<\/span>\r\n\r\nTone and Energy<\/span>:<\/b> Is the writing formal, playful, highly empathetic, direct, or \r\nauthoritative? What is the emotional resonance?<\/span>\r\n\r\nVocabulary and Lexicon<\/span>:<\/b> What specific phrases, transition words, industry jargon,\r\nor colloquialisms do I frequently use? Are there words I strictly avoid?<\/span>\r\n<\/span>Syntax and Structure: Do I use short, punchy sentences or longer, flowing\r\nexplanations? How do I typically open and close my messages?<\/span>\r\n<\/span>Formatting Habits: Do I heavily rely on bullet points, bold text for emphasis, \r\nspecific emojis, or distinct paragraph spacing?<\/span>\r\n\r\nOutput Format: Do not just describe my voice. Output a highly structured, \r\nready-to-use \u201cVoice brief\u201d that I can copy and paste into a Gem. \r\nUse clear headings within the prompt (e.g., <\/span>Your Persona, Tone Guidelines,\r\nVocabulary Rules, Formatting Instructions<\/span><\/i>) telling the AI exactly how to \r\nemulate me without sounding AI-generated.\"<\/span><\/pre>\n <\/div>\n <\/section>\n\n\n2. Mine what you already have instead of staring at a blank page<\/h3>\n\n\n\n
Scan these sources for patterns that show up more than once \u2014 the same objection,\r\nthe same misunderstanding, the same moment of surprise. List 5\u201310 patterns. \r\nFor each, note what triggered it and why it would surprise someone outside my \r\nfield.<\/pre>\n <\/div>\n <\/section>\n\n\n
Here's a pattern I noticed: [paste it<\/em>]. Pitch me 3-5 different post angles in \r\ntwo sentences each \u2014 don't write the post yet.<\/pre>\n <\/div>\n <\/section>\n\n\n
3. Draft it and argue with it<\/h3>\n\n\n\n
Using my voice brief and this angle: [paste chosen angle], draft a short post. \r\nKeep it under 150 words. Open with the specific moment, not a summary. \r\nNo emojis, no rhetorical questions, no 'here's the thing' setups, \r\nno three-part lists dressed up as insight.\r\nEnd when the point is made \u2014 don't add a wrap-up sentence. \r\nIf anything sounds like generic LinkedIn voice rather than mine, \r\nflag it instead of smoothing it over.<\/pre>\n <\/div>\n <\/section>\n\n\n
That closing line reads like a generic LinkedIn wrap-up. \r\nI don't write tidy summary sentences \u2014 I stop when the point's made. \r\nAdd this as a rule to my voice brief.<\/span><\/pre>\n <\/div>\n <\/section>\n\n\n4. Build a loop, not a system<\/h3>\n\n\n\n
The part that convinced the room<\/h2>\n\n\n\n
<\/figure>\n\n\n\n<\/figure>\n\n\n\n
Same model. Same question. The only variable that changed was whether it had anything of mine to work from.<\/p>\n\n\n\nWhat to actually do with this<\/h2>\n\n\n\n
FAQ<\/h2>\n\n\n\n
What’s the fastest way to start writing with AI without sounding like AI?<\/h3>
How much of my own writing do I need to upload?<\/h3>
Is it safe to upload work material to build a voice brief?<\/h3>
Does this method only work for LinkedIn posts?<\/h3>
What should I do if an AI draft still sounds generic?<\/h3>
\n \n
<\/div>\n\n