[{"data":1,"prerenderedAt":636},["ShallowReactive",2],{"blog:\u002Fblog\u002Fai-that-sounds-like-you":3,"blog:related:\u002Fblog\u002Fai-that-sounds-like-you":271},{"id":4,"title":5,"author":6,"body":7,"date":256,"description":257,"draft":258,"extension":259,"image":260,"meta":261,"navigation":262,"path":263,"seo":264,"stem":265,"tags":266,"updated":260,"__hash__":270},"blog\u002Fblog\u002Fai-that-sounds-like-you.md","Getting AI to write in your voice (and why \"write like me\" fails)","SelfSM",{"type":8,"value":9,"toc":245},"minimark",[10,14,17,22,25,49,52,66,75,79,82,88,94,100,103,107,110,116,127,133,139,143,146,149,152,160,164,167,175,182,186,189,200,206,210,216,222,228,239],[11,12,13],"p",{},"You paste three of your best posts into ChatGPT, add \"write in this voice\", and get back something that is unmistakably not you. It has your topic and none of your speech. Add more examples and it gets more average, not more accurate.",[11,15,16],{},"This is not a prompting skill issue. The instruction \"write like me\" asks the model to infer a voice from evidence and then apply it, and the inference step is where it fails. What the model extracts from three posts is not your voice, it is the genre those posts belong to.",[18,19,21],"h2",{"id":20},"what-the-model-actually-learns-from-your-examples","What the model actually learns from your examples",[11,23,24],{},"Give a model five LinkedIn posts and it will reliably pick up:",[26,27,28,37,43],"ul",{},[29,30,31,32,36],"li",{},"The ",[33,34,35],"strong",{},"format",": line breaks, list use, length, whether you open with a one-liner.",[29,38,31,39,42],{},[33,40,41],{},"register",": casual or formal, first person or third.",[29,44,31,45,48],{},[33,46,47],{},"topic domain",": SaaS, recruiting, design.",[11,50,51],{},"Those are the visible layer, and copying them produces something that looks like your posts from a distance. What it will not pick up from five examples:",[26,53,54,57,60,63],{},[29,55,56],{},"Which words you never use.",[29,58,59],{},"Which claims you would refuse to make.",[29,61,62],{},"What you find embarrassing.",[29,64,65],{},"The specific things you have actually done, which are the only source of material that could not have come from anyone else.",[11,67,68,69,74],{},"So the output lands in a predictable place: correct shape, generic substance. It reads like a competent stranger doing an impression, and readers detect it immediately even when they cannot name what is wrong. That gap is ",[70,71,73],"a",{"href":72},"\u002Fblog\u002Fwhy-ai-social-posts-get-ignored","why AI-written posts get ignored"," rather than disliked. Nothing about them is bad. There is just nobody in them.",[18,76,78],{"id":77},"a-voice-is-three-separate-things","A voice is three separate things",[11,80,81],{},"Treating \"voice\" as one setting is what makes it unfixable. It is at least three, and they need different inputs.",[11,83,84,87],{},[33,85,86],{},"Style."," Sentence length, punctuation habits, whether you use questions, how you open and close. This is the layer examples genuinely teach. It is also the least important, because it is the layer readers notice last.",[11,89,90,93],{},[33,91,92],{},"Vocabulary boundaries."," The words that are yours and the words that are disqualifying. Most people cannot list these from memory but recognise them instantly: \"leverage\", \"unlock\", \"in today's landscape\", \"game-changer\". A negative list works better than a positive one, and it is much easier to write.",[11,95,96,99],{},[33,97,98],{},"Substance."," The facts, numbers, opinions and stories you are allowed to draw on. This is the layer that makes a post identifiably yours, and no amount of style transfer produces it. If the model does not have your material, it will invent something plausible in your shape, which is the worst of both.",[11,101,102],{},"Almost everyone tries to solve a substance problem with style instructions. That is why the loop feels endless.",[18,104,106],{"id":105},"what-to-give-a-model-instead","What to give a model instead",[11,108,109],{},"If you want output you would actually publish, give it these four things and stop tuning adjectives.",[11,111,112,115],{},[33,113,114],{},"Ten to twenty of your own posts, not three."," Three is enough to average, not enough to characterise. And use ordinary posts, not only your best: your best posts are your most edited, which means they are the least like how you write.",[11,117,118,121,122,126],{},[33,119,120],{},"A short list of banned words and constructions."," Ten to fifteen entries is plenty. Include the ones that are technically fine but not you. This is the single highest-return input and takes fifteen minutes; if you want a starting list, the ",[70,123,125],{"href":124},"\u002Fblog\u002Fai-writing-tells","common AI tells"," are most of it.",[11,128,129,132],{},[33,130,131],{},"A file of facts you are allowed to cite."," Numbers with the date they were true, customer results with permission status, opinions you hold and would defend. This is boring to assemble and it is the entire difference between a post about your company and a post that could be about any company.",[11,134,135,138],{},[33,136,137],{},"Two or three examples of what you rejected and why."," \"Too salesy\", \"I would never say hustle\", \"this claims something I cannot prove.\" Rejections carry more signal per word than approvals, because approval only says the output was acceptable while a rejection names a boundary.",[18,140,142],{"id":141},"the-test-that-tells-you-whether-it-worked","The test that tells you whether it worked",[11,144,145],{},"Take the draft and remove your name, your company name and the topic. Show it to someone who knows you.",[11,147,148],{},"If they can tell it is yours, the voice landed. If they say \"this could be anyone in your industry\", it did not — regardless of how well it matched your line breaks.",[11,150,151],{},"A cheaper version you can run alone: read it out loud. The places you stumble or feel faintly embarrassed are almost always sentences you would not have written. Voice mismatches are much easier to hear than to see, which is why they survive so many silent re-reads.",[11,153,154,155,159],{},"For the mechanical layer, the constructions that mark text as machine-written regardless of whose voice it was aiming at, you can run a draft through our ",[70,156,158],{"href":157},"\u002Fai-check","AI writing check"," and see the specific spans flagged rather than a score.",[18,161,163],{"id":162},"why-a-chat-window-keeps-losing-your-voice","Why a chat window keeps losing your voice",[11,165,166],{},"Even when you get a good result in ChatGPT, it does not persist. Next week you open a new conversation, the model has none of it, and you re-explain from scratch or paste a prompt block you now maintain by hand.",[11,168,169,170,174],{},"This is the actual daily cost, and it is not about output quality at all. It is that the voice lives in a conversation rather than in the account, so every session starts at zero and every improvement has to be manually carried forward. Anyone who has ",[70,171,173],{"href":172},"\u002Fblog\u002Fchatgpt-claude-for-social-media","used ChatGPT or Claude for social content"," for more than a month has a document somewhere that exists only to re-teach the model things it already learned.",[11,176,177,178,181],{},"Making the voice a stored profile rather than a prompt is most of what ",[70,179,6],{"href":180},"\u002Ffor\u002Ffounders"," does: the samples, the banned list, the facts and the rejections accumulate in one place, and every draft starts from them instead of from a blank context. The interesting part is not that drafts get better. It is that a correction you make once stops being a correction you make weekly.",[18,183,185],{"id":184},"what-will-still-not-work","What will still not work",[11,187,188],{},"Two honest limits, because the promise gets oversold.",[11,190,191,194,195,199],{},[33,192,193],{},"Nothing gives you material."," If you have not done anything worth writing about this month, no voice profile produces a post worth reading. The model can shape what you have; it cannot supply it. This is the same reason ",[70,196,198],{"href":197},"\u002Fblog\u002Flinkedin-ghostwriter-cost","hiring a ghostwriter"," does not remove the founder from the loop.",[11,201,202,205],{},[33,203,204],{},"Voice is not a substitute for judgment."," A well-tuned model will happily write a sentence in your voice that you should not publish. The last read is still yours, and the useful version of that read is checking claims rather than admiring phrasing.",[18,207,209],{"id":208},"faq","FAQ",[11,211,212,215],{},[33,213,214],{},"Why doesn't ChatGPT sound like me even when I give it examples?","\nBecause from a handful of examples it extracts format and register, the visible layer, not your vocabulary boundaries or your material. You get the right shape with generic substance, which reads as an impression. Fix it by supplying more samples, an explicit banned-word list and a file of facts you can cite, rather than more adjectives.",[11,217,218,221],{},[33,219,220],{},"How many writing samples does an AI need to learn a voice?","\nTen to twenty is the useful range for social posts. Three averages rather than characterises. Include ordinary posts, not only your best ones — heavily edited work is the least representative of how you actually write.",[11,223,224,227],{},[33,225,226],{},"Is it better to describe my voice or show examples?","\nBoth, for different layers. Examples teach rhythm and structure, which people describe badly. Explicit rules work better for boundaries: what you never say, what you refuse to claim. Descriptions like \"professional but approachable\" carry almost no information and are the most common wasted instruction.",[11,229,230,233,234,238],{},[33,231,232],{},"Can AI-written posts be detected?","\nDetectors measure statistical patterns and are unreliable on short text, so a confident verdict on a 200-word post is worth little either way. Readers are the real detector, and they respond to the absence of specifics rather than to phrasing. See ",[70,235,237],{"href":236},"\u002Fblog\u002Fwhat-ai-detectors-catch","what AI detectors actually catch",".",[11,240,241,244],{},[33,242,243],{},"What is the fastest thing I can do to make AI drafts sound more like me?","\nWrite a list of ten words and phrases you would never use, and put it in the prompt. It takes fifteen minutes, needs no tooling, and removes most of what makes a draft read as machine-written.",{"title":246,"searchDepth":247,"depth":247,"links":248},"",2,[249,250,251,252,253,254,255],{"id":20,"depth":247,"text":21},{"id":77,"depth":247,"text":78},{"id":105,"depth":247,"text":106},{"id":141,"depth":247,"text":142},{"id":162,"depth":247,"text":163},{"id":184,"depth":247,"text":185},{"id":208,"depth":247,"text":209},"2026-08-25","Prompting an AI to sound like you almost never works, and the reason is mechanical. What a voice is actually made of, what you have to give a model, and how to tell whether it landed.",false,"md",null,{},true,"\u002Fblog\u002Fai-that-sounds-like-you",{"title":5,"description":257},"blog\u002Fai-that-sounds-like-you",[267,268,269],"founders","ai","writing","o_PBouvxI0s03S8G-pPaBBw4lqbGchRi3EAnBrrt11c",[272,410,518],{"id":273,"title":274,"author":6,"body":275,"date":403,"description":404,"draft":258,"extension":259,"image":260,"meta":405,"navigation":262,"path":124,"seo":406,"stem":407,"tags":408,"updated":260,"__hash__":409},"blog\u002Fblog\u002Fai-writing-tells.md","The tells that make writing sound like AI (and how to fix each one)",{"type":8,"value":276,"toc":393},[277,280,284,287,293,297,300,305,309,312,317,321,324,329,333,336,341,345,348,353,357,360,365,369,377],[11,278,279],{},"People can't always explain why a post feels off, but they sense it almost immediately and keep scrolling. What they're reacting to is a set of patterns: habits the models fall into that, once you've seen them a few times, are impossible to unsee. Here's the field guide, with the fix for each.",[18,281,283],{"id":282},"the-em-dash-everywhere","The em-dash everywhere",[11,285,286],{},"Generated posts often overuse the em-dash. One per paragraph, sometimes two, usually in that \"clause — extra clause\" rhythm. Human writers tend to use it less predictably, and with more intention.",[11,288,289,292],{},[33,290,291],{},"Fix:"," Read the post and keep at most one em-dash. Turn the rest into full stops. Short sentences carry more weight anyway.",[18,294,296],{"id":295},"puffery-instead-of-specifics","Puffery instead of specifics",[11,298,299],{},"\"Powerful.\" \"Seamless.\" \"Game-changing.\" \"Robust.\" Adjectives that sound like value but carry none. They're what a sentence reaches for when it doesn't have a fact.",[11,301,302,304],{},[33,303,291],{}," Delete the adjective and put a specific in its place. Not \"a powerful workflow\" but \"cut our drafting time in half.\" If you can't name the specific, the claim probably isn't true yet.",[18,306,308],{"id":307},"the-its-not-x-its-y-reflex","The \"it's not X, it's Y\" reflex",[11,310,311],{},"\"It's not about working harder, it's about working smarter.\" Once you notice this construction, you'll see it in a lot of generated posts. It feels insightful and says almost nothing.",[11,313,314,316],{},[33,315,291],{}," Cut it. State the point directly. If Y is the real idea, just say Y.",[18,318,320],{"id":319},"suspicious-symmetry","Suspicious symmetry",[11,322,323],{},"Three bullet points, each the same length, each starting with the same kind of phrase. Real thinking is lumpy: one idea is big, another is a footnote. Perfectly balanced structure reads like a template.",[11,325,326,328],{},[33,327,291],{}," Make the list uneven on purpose. One long point, one short. Cut the third if it's only there to make three.",[18,330,332],{"id":331},"the-rhetorical-question-opener","The rhetorical-question opener",[11,334,335],{},"\"Ever wondered why some posts go viral?\" It's a stall dressed as a hook. The reader hasn't been given anything yet, just asked to do the work.",[11,337,338,340],{},[33,339,291],{}," Open with a specific instead. A number, a moment, a claim. \"Last week a post with a typo outperformed everything I'd polished for a month.\"",[18,342,344],{"id":343},"the-closing-that-begs","The closing that begs",[11,346,347],{},"\"What do you think?\" \"Agree?\" \"Drop your thoughts below.\" A confident post lands its point and stops. The plea for engagement reads as exactly that.",[11,349,350,352],{},[33,351,291],{}," End on a statement. If a question genuinely belongs, make it a real one, specific enough that answering it takes thought.",[18,354,356],{"id":355},"the-stacked-hedge","The stacked hedge",[11,358,359],{},"\"It's often the case that many teams sometimes struggle to…\" One qualifier can be honest. Three in a row drain the life out of a sentence and hide whether you're actually saying anything.",[11,361,362,364],{},[33,363,291],{}," Keep the qualification that's true; cut the ones added only to sound safe. \"Teams often struggle with this\" says as much as the stacked version, and you can read it without wincing.",[18,366,368],{"id":367},"how-to-actually-apply-this","How to actually apply this",[11,370,371,372,376],{},"You don't need to memorize the list. Do one pass with a single question in mind: ",[373,374,375],"em",{},"would I say this sentence out loud to someone I respect?"," Most tells don't survive it. The puffery sounds fake, the symmetry sounds rehearsed, the begging sounds needy.",[11,378,379,380,383,384,387,388,392],{},"This is the same standard ",[70,381,6],{"href":382},"\u002F"," is built around, so the draft comes back with these patterns already stripped and the editing pass is short instead of a rescue mission. For the fuller argument on why this matters, start with ",[70,385,386],{"href":72},"why AI social posts get ignored",". When you're editing your own chatbot output, ",[70,389,391],{"href":390},"\u002Fblog\u002Fhumanize-chatgpt-draft","this walkthrough"," turns the list above into a repeatable pass.",{"title":246,"searchDepth":247,"depth":247,"links":394},[395,396,397,398,399,400,401,402],{"id":282,"depth":247,"text":283},{"id":295,"depth":247,"text":296},{"id":307,"depth":247,"text":308},{"id":319,"depth":247,"text":320},{"id":331,"depth":247,"text":332},{"id":343,"depth":247,"text":344},{"id":355,"depth":247,"text":356},{"id":367,"depth":247,"text":368},"2026-08-02","A field guide to the patterns readers now recognize as machine-written, with a concrete fix for every one, so your posts read like a person again.",{},{"title":274,"description":404},"blog\u002Fai-writing-tells",[269,268],"c_mykMSYGzeBanFnRrY_2aITCR3Za4SSqygS3tCg8kw",{"id":411,"title":412,"author":6,"body":413,"date":511,"description":512,"draft":258,"extension":259,"image":260,"meta":513,"navigation":262,"path":390,"seo":514,"stem":515,"tags":516,"updated":260,"__hash__":517},"blog\u002Fblog\u002Fhumanize-chatgpt-draft.md","How to edit a ChatGPT draft so it sounds like you",{"type":8,"value":414,"toc":502},[415,418,422,425,429,432,436,439,443,446,450,453,457,460,464,485],[11,416,417],{},"The draft is done in ten seconds. Making it sound like a person takes longer, and it's the part most people skip, which is why so many feeds read like the same voice wearing different logos. You don't have to rewrite from scratch. You have to run a pass. Here's the one I use.",[18,419,421],{"id":420},"_1-cut-the-first-paragraph","1. Cut the first paragraph",[11,423,424],{},"Chatbots warm up. The real opening is usually the second or third sentence, once the model has finished clearing its throat. Read the draft and find the first line that actually says something. Start there. Delete everything above it.",[18,426,428],{"id":427},"_2-replace-the-biggest-adjective-with-a-fact","2. Replace the biggest adjective with a fact",[11,430,431],{},"Scan for the puffed-up words: \"powerful,\" \"seamless,\" \"essential,\" \"game-changing.\" Pick the one doing the most fake work and swap it for something concrete: a number, a name, a moment. The rest you can usually just delete. Nouns and verbs carry a sentence; adjectives are what you reach for when you don't have one.",[18,433,435],{"id":434},"_3-break-the-symmetry","3. Break the symmetry",[11,437,438],{},"If there are three bullet points of equal length, the draft is wearing a uniform. Make it uneven. Merge two, cut one, let one point run long because it deserves to. Human emphasis is uneven; templates are tidy.",[18,440,442],{"id":441},"_4-read-it-out-loud","4. Read it out loud",[11,444,445],{},"This is the step that does the most work and gets skipped the most. Read the whole thing aloud. Every sentence you stumble on, or wouldn't actually say to someone, gets rewritten or cut. \"It's not about X, it's about Y\" dies here. So does the hedging. So does the closing plea.",[18,447,449],{"id":448},"_5-add-one-thing-only-you-know","5. Add one thing only you know",[11,451,452],{},"The fastest way to make writing human is a specific that couldn't have been generated: a detail from your week, a client's exact words, a number from your own work, an opinion you'd defend. One real specific does more than a hundred edits. It's also what search engines and readers both reward, because it's proof a person was actually here.",[18,454,456],{"id":455},"_6-end-on-a-period-not-a-question","6. End on a period, not a question",[11,458,459],{},"Delete \"What do you think?\" Land the point and stop. If a question genuinely belongs, make it specific enough to be worth answering.",[18,461,463],{"id":462},"the-whole-pass-in-order","The whole pass, in order",[465,466,467,470,473,476,479,482],"ol",{},[29,468,469],{},"Cut the warm-up opening.",[29,471,472],{},"Swap the biggest adjective for a fact.",[29,474,475],{},"Make the structure uneven.",[29,477,478],{},"Read aloud; fix anything you wouldn't say.",[29,480,481],{},"Add one thing only you know.",[29,483,484],{},"End on a statement.",[11,486,487,488,491,492,495,496,498,499,238],{},"It takes about five minutes once it's a habit. If you want to know exactly what you're editing ",[373,489,490],{},"out",", the ",[70,493,494],{"href":124},"full list of AI tells"," is the companion to this pass. And if doing this by hand for every post across several brands is the actual bottleneck, that's the part ",[70,497,6],{"href":382}," automates, so the draft arrives already past step four. The step before this one — getting the model to write in your voice at all, rather than editing it in afterwards — is ",[70,500,501],{"href":263},"here",{"title":246,"searchDepth":247,"depth":247,"links":503},[504,505,506,507,508,509,510],{"id":420,"depth":247,"text":421},{"id":427,"depth":247,"text":428},{"id":434,"depth":247,"text":435},{"id":441,"depth":247,"text":442},{"id":448,"depth":247,"text":449},{"id":455,"depth":247,"text":456},{"id":462,"depth":247,"text":463},"2026-07-31","A five-minute editing pass that turns generic chatbot output into something in your voice, no rewrite from scratch.",{},{"title":412,"description":512},"blog\u002Fhumanize-chatgpt-draft",[269,268],"SdXUzZvHI8yik8UOH3bCwmfsw55i-m_8TU8cDEYFrnw",{"id":519,"title":520,"author":6,"body":521,"date":629,"description":630,"draft":258,"extension":259,"image":260,"meta":631,"navigation":262,"path":236,"seo":632,"stem":633,"tags":634,"updated":260,"__hash__":635},"blog\u002Fblog\u002Fwhat-ai-detectors-catch.md","What AI content detectors actually catch (and what they miss)",{"type":8,"value":522,"toc":623},[523,526,530,533,547,554,558,561,575,578,582,585,588,592,595,609],[11,524,525],{},"People keep asking: will this pass an AI detector? It's the wrong question, but it's worth understanding why, because the answer changes how you edit.",[18,527,529],{"id":528},"how-detectors-actually-work","How detectors actually work",[11,531,532],{},"AI-text detectors don't \"know\" a machine wrote something. They measure statistical properties of the text, mainly two:",[26,534,535,541],{},[29,536,537,540],{},[33,538,539],{},"Perplexity:"," how predictable each next word is. Model output runs smooth and predictable; human writing is bumpier and more surprising.",[29,542,543,546],{},[33,544,545],{},"Burstiness:"," how much sentence length and rhythm vary. People write a long sentence, then a short one. Models drift toward an even, uniform cadence.",[11,548,549,550,553],{},"A detector sees low perplexity and low burstiness and raises its hand. That's the whole trick. It's a probability guess about ",[373,551,552],{},"how the text is shaped",", not a fact about who made it.",[18,555,557],{"id":556},"why-they-get-it-wrong-in-both-directions","Why they get it wrong in both directions",[11,559,560],{},"Because it's a shape-based guess, it fails predictably.",[26,562,563,569],{},[29,564,565,568],{},[33,566,567],{},"It flags real human writing."," Clear, plain, well-structured prose, the kind good writers aim for, can look \"too smooth\" and get marked as AI. Non-native English writers get flagged more often, unfairly.",[29,570,571,574],{},[33,572,573],{},"It clears real AI."," Add a few odd word choices, vary the sentence lengths, and generated text sails through. The patterns detectors measure are the ones easiest to perturb.",[11,576,577],{},"So a passing score proves very little, and a failing score might just mean you wrote cleanly.",[18,579,581],{"id":580},"the-right-goal-isnt-a-score-its-a-reader","The right goal isn't a score, it's a reader",[11,583,584],{},"Here's what the detector debate misses: your audience is the detector that matters. A person doesn't run a perplexity check. They feel the tells, the puffery, the tidy symmetry, the \"not X but Y\" reflex, and they lose interest. You can pass every automated detector and still get scrolled past, and you can fail one while writing something people actually read.",[11,586,587],{},"Optimizing for a detector score pulls you toward gaming statistics. Optimizing for a human pulls you toward specifics, real opinions, and uneven, natural rhythm. That also lowers your detector score, as a side effect rather than the goal.",[18,589,591],{"id":590},"what-to-do-instead","What to do instead",[11,593,594],{},"Forget the score. Edit for the reader:",[26,596,597,600,603,606],{},[29,598,599],{},"Lead with something specific and true.",[29,601,602],{},"Vary your sentence lengths on purpose.",[29,604,605],{},"Cut the adjectives that stand in for facts.",[29,607,608],{},"Add at least one detail only you could know.",[11,610,611,612,615,616,619,620,622],{},"That's the same standard behind the ",[70,613,614],{"href":124},"list of AI tells"," and the ",[70,617,618],{"href":390},"editing pass",", and it's what ",[70,621,6],{"href":382}," optimizes for by default: sound like a person, because that's who's actually reading.",{"title":246,"searchDepth":247,"depth":247,"links":624},[625,626,627,628],{"id":528,"depth":247,"text":529},{"id":556,"depth":247,"text":557},{"id":580,"depth":247,"text":581},{"id":590,"depth":247,"text":591},"2026-07-26","How AI-text detectors really work, why they flag human writing and clear real AI, and why chasing a detector score is the wrong goal.",{},{"title":520,"description":630},"blog\u002Fwhat-ai-detectors-catch",[268,269],"ivIVWYDJGhPwmmo9TB8N8WDqcYvZ4RYGdzQpOuu3fJE",1787652583417]