# The AI Writing Filter ## A 7-Pass System for Eliminating AI Tells From Your Writing **Before you use AI to write anything, run it through this filter. Every sentence should pass one test: Would a reader flag this as AI-generated?** --- ## HOW TO USE THIS FILTER This document contains four tools: 1. **The Full 7-Pass Filter** - A complete prompt you paste into any AI conversation alongside your draft. It scans for AI patterns and returns a cleaned version scored on a diagnostic scorecard. 2. **The Quick-Use Filter** - A condensed version for short-form content like emails, social posts, and Slack messages. 3. **The AI Tell Field Guide** - A reference section explaining the 8 most common AI writing patterns, why they're a problem, and how to fix them. Use this to train your own eye. 4. **The Voice Training System** - A step-by-step process for building your own voice profile so AI writes like you from the start, not just after cleanup. The filter catches AI patterns after the fact. The voice training system prevents them from showing up in the first place. Use both. Start with the voice training to build your profile, then use the filter to quality-check everything that comes out. --- ## THE FULL 7-PASS FILTER Copy everything below the line and paste it into a new AI conversation along with your draft. --- ``` You are my writing editor. Your sole job is to review my draft and eliminate every detectable AI writing pattern while preserving my voice, meaning, and intent. A few things about how I write: [DESCRIBE YOUR VOICE IN 2-3 SENTENCES. Example: "I write the way I talk - direct, specific, occasionally funny, never corporate. I use contractions. I tell stories from my actual life."] Run the following 7-pass filter on my draft and return: 1. A scored diagnostic (see Pass 7) 2. The full rewritten draft with all fixes applied 3. A brief changelog summarizing what you changed and why --- PASS 1: VOCABULARY SCAN - Kill the AI Dictionary Search the draft for these categories of AI-flagged words and phrases. Flag every instance. Replace with simpler, more specific, or more natural alternatives. Verbs to flag: delve, leverage, utilize, foster, underscore, harness, navigate (as metaphor), embark, illuminate, elucidate, showcase, streamline, optimize, revolutionize, empower, elevate, unlock, unleash, craft (as verb for non-physical things), cultivate, reimagine, bolster, facilitate, expedite, spearhead, pivot (unless literal), curate Adjectives/adverbs to flag: robust, pivotal, groundbreaking, cutting-edge, transformative, innovative, comprehensive, meticulous, seamless, bespoke, paramount, profound, indelible, remarkable, vibrant, multifaceted, intricate, nuanced, holistic, unparalleled, compelling, game-changing, thought-provoking. Adverbs: seamlessly, meticulously, profoundly, tirelessly, strategically, arguably Nouns/metaphors to flag: tapestry, kaleidoscope, landscape (metaphorical), realm, beacon, treasure trove, synergy, interplay, testament, mosaic, paradigm, symphony, cornerstone, endeavor, journey (unless literal travel), deep dive, framework (when vague), ecosystem (non-biological) Phrases to flag and delete or rewrite: - "In today's [anything]" / "In an era where" / "In the ever-evolving landscape of" - "It's important to note" / "It's worth noting" / "It's worth mentioning" - "At the end of the day" - "Here's the thing" / "Here's the deal" / "Here's an uncomfortable truth" - "Let's be honest" / "Let's be real" / "Let's face it" - "Based on the information provided" - "This is a testament to" - "The power of [noun]" - "More than just [X]" - "When it comes to" - "In the world of" - "Take [X] to the next level" - "Serve as a [beacon/reminder/testament]" - Any sentence starting with "Imagine" as a hook device - "Think about it" / "Consider this" - "And honestly?" / "And truthfully?" Replacement principle: Don't swap one fancy word for another fancy word. Go simpler and more specific. "Leverage your network" becomes "ask the people you know." "Navigate challenges" becomes "deal with the hard parts" or name the actual challenge. "Unlock your potential" becomes a specific outcome: "close your first paying client" or "finish the book." --- PASS 2: SENTENCE PATTERN SCAN - Break the AI Machinery Search for and fix these structural patterns: "It's not X, it's Y" / "It's not about X, it's about Y" This is the single most recognized AI writing tell. Find every instance. Rewrite to state the point directly. - BAD: "It's not about the money - it's about the freedom." - GOOD: "Freedom is what you're actually after." The Hat Trick (compulsive rule of three) AI groups everything in threes with parallel structure. Find these and either combine into one sentence, break the pattern with 2 or 4 items, or vary the rhythm. - BAD: "It takes courage. It takes conviction. It takes commitment." - GOOD: "It takes courage and a stubborn refusal to quit." The Unnecessary Q&A AI asks a question and immediately answers it. Cut the question, keep the answer. - BAD: "What does this really mean? It means you have to show up." - GOOD: "You have to show up." The No-No-No Parade AI creates dramatic emphasis by listing what something isn't. Combine into one sentence or flip to what it IS. - BAD: "No playbook. No safety net. No guarantee." - GOOD: "You're building this without a playbook, and that's the point." Trailing present participle clauses Sentences ending with ", highlighting...", ", showcasing...", ", demonstrating...", ", making it...", ", allowing...", ", enabling...", ", creating...", ", ensuring..." - cut these. If the information matters, make it its own sentence. - BAD: "He changed his entire routine, showcasing the power of small shifts." - GOOD: "He changed his entire routine. Small shifts, compounding." Monotonous sentence length If five or more consecutive sentences are roughly the same length (within ~5 words of each other), vary them. Insert a short sentence. Extend one. Break the rhythm. Excessive hedging Find clusters of: often, typically, generally, can be, may, might, arguably, to some extent, in many cases, it could be said. Replace with direct statements. Take a position. Em dash overuse Count em dashes. If more than 2 appear in any 500-word section, replace most with commas, periods, or parentheses. False ranges / spectrum claims "From [X] to [Y]" where X and Y are just two loosely related things pretending to be a spectrum. "From boardrooms to locker rooms." "From technical expertise to creative vision." Cut or rewrite to be specific. --- PASS 3: STRUCTURE SCAN - Destroy the AI Skeleton Predictable essay structure If the piece follows intro, 3 equal sections, summary conclusion - restructure. Sections should be different lengths based on importance. Democratic paragraph length If all paragraphs are within 1-2 sentences of each other in length, vary them. Some paragraphs should be one sentence. Some should be five. Meta-commentary Delete every sentence that describes what the piece is about to do instead of doing it: - "In this article, we'll explore..." - "Let's take a look at..." - "Now let's turn to..." - "First, we'll examine... then we'll discuss..." - "Now that we've covered X, let's move on to Y" - "With that in mind..." - "That said..." (when used as a generic transition) Generic introductions Test: could this opening paragraph be pasted onto a completely different article with minor word changes? If yes, rewrite with something specific - a story, a concrete detail, a bold claim. Recap conclusions If the conclusion starts with "Overall," "In conclusion," "In summary," "To sum up," or "At the end of the day" - rewrite. End with a forward-looking thought, a call to action, or the single most important takeaway stated fresh. Unnecessary headers and sub-headers For content under 1,000 words, headers are often unnecessary and signal AI structure. If the piece flows naturally without them, remove them. For longer pieces, keep headers but rewrite any that follow the pattern "The Power of X," "The Importance of Y," "Understanding Z," or "The Future of Q." --- PASS 4: VOICE AND AUTHENTICITY SCAN - Make It Sound Human The "would I actually say this?" test Read every sentence as if the author is saying it out loud to a friend over coffee. Flag anything that sounds like a corporate memo, a TED talk, or a LinkedIn influencer post. Contractions check Most people use contractions naturally. "It is" should usually be "it's." "Do not" should usually be "don't." The exception is when being deliberately emphatic. Register variation The piece should shift naturally between conversational and reflective. If every sentence is the same register (all casual or all formal), flag and vary. Personal specificity check Does the piece contain at least one of these: - A real story from the author's life or work - A specific person, place, or event (named, not generic) - A concrete number or data point from actual experience - An opinion the author is willing to defend that not everyone would agree with If none of these are present, flag it. A piece without personal specificity will read as AI regardless of how clean the language is. Performative emotion check Flag phrases that simulate depth without delivering it: - "And that's the real magic." - "That's where the transformation happens." - "This is what it's all about." - "And that changes everything." - "Let that sink in." - "Read that again." - "Full stop." These are Instagram caption moves. Cut them or replace with something specific. Flattery and cheerleading check Flag excessive encouragement that isn't earned by the context: - "You've got this." - "You're already ahead of the game." - "The fact that you're even asking this question shows..." - "Your courage is inspiring." Encouragement should be specific and earned, not sprinkled in as filler. --- PASS 5: FORMATTING SCAN - Clean the Visual Fingerprint - Remove all emoji from body text (unless the author specifically included them) - Reduce em dashes; replace most with commas, periods, or parentheses - Strip any Unicode formatting artifacts (bold Unicode, arrows, multiplication signs) - Convert Title Case Headers to Sentence case headers - Reduce bold text to only the 1-2 most critical phrases per section (if any) - Remove the bold-colon-restatement pattern from any bullet lists - Normalize quotation marks - Remove any Markdown artifacts that leaked through (stray asterisks, backticks, hash symbols) - Check for consistent formatting throughout --- PASS 6: CONTENT DEPTH SCAN - Surface Polish vs. Real Substance Vague claims test Find every sentence that makes a claim without evidence or specifics. "This approach works" - for whom? When? How do you know? Flag each one and either add specifics or cut the sentence. Interchangeability test For each paragraph, ask: could this paragraph appear in any other article on a similar topic with zero changes? If yes, it's too generic. Flag for rewrite with specifics unique to the author's experience. The "so what?" test After each main point, silently ask "so what?" If the draft doesn't answer that question with something concrete and actionable, flag it. Story completeness check If a story or anecdote is included, does it have: - A specific person (named or identified by role) - A concrete situation (not "someone once told me" but what actually happened) - A real outcome (not "and everything changed" but what specifically changed) If any element is vague, flag for more detail. --- PASS 7: DIAGNOSTIC SCORECARD Score the draft on each dimension (1-10, where 10 = fully human, no AI tells): | Category | Score | Key Issues | |---|---|---| | Vocabulary (no AI buzzwords) | /10 | | | Sentence patterns (no AI formulas) | /10 | | | Structure (no AI skeleton) | /10 | | | Voice (sounds like a real person) | /10 | | | Formatting (clean, human) | /10 | | | Content depth (specific, lived) | /10 | | | OVERALL | /60 | | Scoring guide: - 50-60: Ready to publish. Minor tweaks only. - 40-49: Solid draft. Needs targeted fixes in flagged areas. - 30-39: AI patterns are noticeable. Needs significant revision. - Below 30: Major rewrite needed. AI tells are dominant. --- OUTPUT FORMAT Return your response in this exact order: 1. DIAGNOSTIC SCORECARD (The table above, filled in) 2. FULL REWRITTEN DRAFT (The complete piece with all fixes applied. Preserve meaning, intent, and structure where possible. Only change what triggers a flag.) 3. CHANGELOG (Brief list of what you changed and why, organized by pass number.) --- IMPORTANT NOTES - Do NOT over-edit. If a sentence is fine, leave it alone. The goal is to eliminate AI tells, not rewrite the author's voice. - Do NOT add AI tells while fixing them. Watch your own output for the same patterns. - Do NOT add content the author didn't write. You can suggest where to add a story or specific detail, but mark those as [SUGGESTION: add specific example here] rather than fabricating one. - Preserve any branded terms exactly as written. - When in doubt about a word choice, go simpler and more specific, not fancier. Here is my draft to review: [PASTE YOUR DRAFT HERE] ``` --- ## THE QUICK-USE FILTER For shorter emails, social posts, or Slack messages where the full 7-pass filter is overkill. --- ``` Review this draft for AI writing tells. Specifically: 1. Flag and replace any words from this list: delve, leverage, utilize, foster, navigate, harness, robust, pivotal, seamless, transformative, tapestry, landscape, realm, compelling, holistic, nuanced, multifaceted, groundbreaking, unlock, unleash, elevate, embark, paramount, vibrant, meticulous, cornerstone, testament, beacon, endeavor 2. Find and fix: "It's not X, it's Y" patterns, compulsive threes, rhetorical questions immediately answered, trailing ", highlighting/showcasing/demonstrating..." clauses, and excessive em dashes 3. Cut: meta-commentary ("In this post we'll..."), generic openers ("In today's..."), recap conclusions, and performative depth ("Let that sink in," "And that changes everything") 4. Check: Does it sound like a real person wrote it? Are there specific stories, names, or details? Would someone say this out loud? Return the cleaned draft and a brief list of changes. Draft: [PASTE HERE] ``` --- ## THE AI TELL FIELD GUIDE These are the 8 patterns that make AI-assisted writing detectable. Learn to spot them yourself and you'll eventually stop needing the filter. ### 1. Fluff Connectors **What to spot:** "The thing is..." / "Here's the thing..." / "Honestly..." (as a sentence opener) / "Spoiler..." / Excessive em dashes **Why it's a problem:** These add zero meaning. They're verbal tics that pretend to sound conversational but actually clog up sentences and make readers work harder than they should. **Fix:** Delete them. If you need a pause, use a comma, period, or colon. Watch your sentences sharpen instantly. ### 2. The Opposition Formula **What to spot:** "It's not X, it's Y" / "Not X. Y." / "You're not doing A, you're doing B" **Why it's a problem:** This construction burns half your word count telling readers what something isn't when they came for what it is. It's like giving someone directions by listing everywhere they shouldn't go first. **Fix:** Skip the windup. Just say what the thing IS. One or two opposition structures per piece maximum, and only when the contrast genuinely clarifies the point. ### 3. The Hat Trick **What to spot:** Three sentences with the same opening structure and different endings. "When I listen... When I ask... When I sit..." / "Through X. Through Y. Through Z." / Any triple-beat parallel construction. **Why it's a problem:** This pattern screams AI from a mile away. It's rhythm without reason, artificially dramatic and weirdly robotic at the same time. Also, it takes up three times the space it needs to. **Fix:** Combine into one or two sentences with varied structure. Or use commas and "and" to join them naturally. Same punch, one-third the real estate. ### 4. The Unnecessary Q&A **What to spot:** AI asks a question and immediately answers it. "The truth? Simplicity is the pathway to ease..." / "What's the secret? The secret is..." **Why it's a problem:** The question is just filler dressed up as suspense. Readers scan right past it to find the actual answer, which means you just wasted a sentence on nothing. **Fix:** Cut the question. Start with the answer. Faster, cleaner, and your reader already made it to the good part. ### 5. The No-No-No Parade **What to spot:** Dramatic emphasis by listing what something isn't, with each item as its own sentence. "Not today. Not tomorrow. Not next week." / "No standard. No practice. No review." **Why it's a problem:** This structure eats up space and sounds like a theatrical performance instead of a conversation. It's the written equivalent of someone wagging their finger at you. **Fix:** Combine the list into one sentence, or flip it to what the thing actually IS. ### 6. Matched Pairs **What to spot:** Two sentences with identical structure, same length, opposite words. "Winning hides bad habits. Losing obscures growth." / "Impressive gets you the nod. Present gets you the phone call." **Why it's a problem:** Perfect symmetry is a construction. People don't talk in matched pairs. It reads as crafted, not spoken. **Fix:** Break the symmetry. Make one sentence longer. Change the structure of the second. Let it be uneven the way speech is uneven. ### 7. The Mic-Drop Fragment **What to spot:** Short fragment at the end of a paragraph designed to recontextualize everything before it. "Until someone names them." / "They need the mirror." / "That's the whole game." **Why it's a problem:** AI loves ending paragraphs with a dramatic short fragment that makes the reader feel like the paragraph was building to a revelation. Most people don't write or talk this way. **Fix:** Either fold it into the preceding sentence or cut it entirely. ### 8. The Bumper Sticker **What to spot:** Any line that sounds like it was written to be screenshotted or quoted. Perfect parallel constructions designed to be pulled out of context. Lines that are more clever than true. **Why it's a problem:** Pull quotes are designed for pages, not conversations. If you wouldn't say it out loud without feeling like you rehearsed it, it's a bumper sticker. **Fix:** Roughen it up or cut it. Your readers will trust you more when your writing sounds like you talking, not like a motivational poster. --- ## THE SELF-CHECK Before publishing anything, run it through these questions. **Voice check:** Would you say this out loud to one person across a table? Does it sound like conversation or like writing? Are there enough natural connectors (so, and, but, because)? Does it settle at the end or try to land dramatically? Did you open with something specific, not an abstraction? **AI tell check:** Are there any matched pairs? Any triple-beat patterns? Any opposition formulas that could just state what the thing IS? Any mic-drop fragments? Any lines that sound like pull quotes or bumper stickers? Any fluff connectors? Any unnecessary Q&As? Any No-No-No parades? **Depth check:** Does every claim have evidence or specifics behind it? Could any paragraph be swapped into a completely different article with zero changes? Does every story have a specific person, a concrete situation, and a real outcome? **The final test:** Read it out loud. If you stumble, the sentence is written, not spoken. Rewrite it the way you'd actually say it to someone you trust. --- ## HOW TO TRAIN AI WITH YOUR VOICE The filter above catches AI patterns after the writing is done. This section is about preventing them from showing up in the first place. When AI has a detailed profile of how you actually sound, the output starts closer to your voice and needs far less cleanup. This isn't a shortcut. Building a real voice profile takes a few hours of upfront work. But once you have it, every AI writing session starts from your voice instead of the AI's default. ### The Core Idea AI doesn't know how you sound. It knows how the internet sounds. So when you ask it to write something, it defaults to a generic professional voice that reads like a corporate blog post crossed with a TED talk. The fix is to give it concrete evidence of your actual voice before it writes a single word. The best evidence isn't your polished writing. It's your transcribed speech. When you talk, you don't use AI vocabulary. You don't structure thoughts in matched pairs or triple beats. You use your actual words, your actual rhythm, your actual way of building an idea. That's what we need to capture. ### Step 1: Gather Your Source Material (The Transcript Library) You need 8-12 samples of yourself talking in different contexts. The variety matters because your voice shifts depending on the situation, and a good profile captures the full range. **What to collect:** - 2-3 casual conversations (phone calls with colleagues, Zoom chats with friends or business partners, recordings of yourself talking through an idea) - 2-3 professional conversations (coaching sessions, client calls, team meetings, sales conversations) - 2-3 presentations or talks (webinars, workshops, conference sessions, podcast guest appearances) - 1-2 podcast interviews where you're the guest (these are gold because the interviewer draws out your natural explanations) - 1-2 recordings of yourself thinking out loud (voice memos, Loom videos where you're walking through a problem) **How to get transcripts:** Record your calls and meetings with tools like Fathom, Otter.ai, Fireflies, or the built-in transcription in Zoom or Google Meet. For existing audio or video, upload to any transcription service. You don't need perfect transcripts. Rough auto-transcripts work fine because what matters is the vocabulary and rhythm, not the punctuation. **If you don't have recordings yet:** Start recording now. Turn on transcription for your next 10-15 calls and meetings. Within a few weeks you'll have plenty of material. In the meantime, you can start with any existing long-form writing that's close to how you talk (personal emails, informal blog posts, unedited first drafts), but know that transcripts will always produce a better profile. ### Step 2: Build the Voice Profile Once you have your transcripts, paste them into an AI conversation and use this prompt: ``` I'm going to give you [NUMBER] transcripts of me speaking in different contexts: conversations, presentations, coaching sessions, and interviews. I need you to analyze these and build a detailed voice profile that captures exactly how I sound. Analyze the transcripts for: 1. SENTENCE STRUCTURE - Average sentence length when I'm speaking naturally - How I connect thoughts (do I use "and," "so," "but," "because" or do I use short fragments?) - Do I trail into the next thought or finish cleanly? - Do I speak in long connected sentences or short punchy ones? 2. TRANSITION WORDS & BRIDGES - What words do I use most often to move between ideas? (so, and, but, well, look, okay, right, you know, I mean, etc.) - How do I open new thoughts? - How do I signal I'm wrapping up a point? 3. VOCABULARY I ACTUALLY USE - What are my go-to words? (words that show up repeatedly across multiple transcripts) - What words do I use for emphasis? - What words do I use for agreement or confirmation? - What casual filler words show up? (just, like, pretty, kind of, basically, etc.) 4. VOCABULARY I NEVER USE - Based on these transcripts, what kinds of words are absent? (formal words, literary words, corporate jargon, emotional adjectives, etc.) - What register of language do I avoid? 5. HOW I OPEN A STORY OR MAKE A POINT - Do I set up context first or drop into the moment? - Do I start with feelings or with actions? - What's my typical opening move? 6. HOW I LAND A POINT - Do I build to a single punchline or circle back from multiple angles? - Do I end dramatically or settle quietly? - What do my closing sentences typically sound like? 7. HOW I HANDLE SERIOUS OR VULNERABLE TOPICS - Do I name emotions directly or let the story carry the weight? - Is my vulnerability performed or matter-of-fact? - Do I lead with the lesson or the story? 8. HUMOR AND WARMTH - Is my humor dry, self-deprecating, playful, or something else? - Does it show up in crafted jokes or in how I say things? - How do I express warmth? 9. RHYTHM AND PACING - Do I use parallel structures or avoid them? - Do I repeat phrases for emphasis? - Is my pacing steady, fast, or variable? 10. PATTERNS TO AVOID (things AI does that I don't) - Based on my actual speech, what AI patterns would be most obviously wrong for my voice? Format the profile as a reference document I can paste into any AI conversation before asking it to write in my voice. Include specific examples from the transcripts for each pattern you identify. End with a numbered summary list of the 10-12 most important rules for writing in my voice. Here are my transcripts: [PASTE TRANSCRIPTS HERE] ``` ### Step 3: Test and Refine the Profile Your first voice profile won't be perfect. Here's how to sharpen it. **The side-by-side test:** Take something you've already written that you're happy with. Then give AI your voice profile and ask it to write on the same topic without showing it the original. Compare the two versions. Where they diverge is where your profile needs more detail. **The out-loud test:** Read the AI output out loud. Every sentence where you stumble or think "I'd never say it that way" points to a gap in the profile. Note the specific issue (wrong vocabulary, sentence too short, too dramatic, etc.) and add a rule to your profile. **The specificity upgrade:** Vague profile instructions produce vague results. "Write conversationally" means nothing to AI. "Connect thoughts with 'and' and 'so' rather than dropping periods. Average sentence length 20-30 words. Start new ideas with 'So' or 'And' freely." That gives AI something concrete to work with. **Iterate 3-4 times.** After each round of testing, add the patterns you caught to your profile. Most profiles stabilize after 3-4 revision cycles. ### Step 4: Create Your Writing Filter (Combining Voice + AI Tells) Once your voice profile is solid, combine it with the AI tell elimination system from this document into a single reference file. The structure looks like this: **Part 1: Your Voice Profile** - How you sound. This is what AI should write toward. Include sentence structure, vocabulary, transitions, story patterns, landing patterns, and specific examples. **Part 2: AI Tells to Eliminate** - The 8 patterns from the field guide in this document (fluff connectors, opposition formula, hat trick, unnecessary Q&A, no-no-no parade, matched pairs, mic-drop fragment, bumper sticker). These are what AI should avoid. **Part 3: Self-Check Questions** - The voice check, AI tell check, and depth check questions adapted to your specific voice. Save this as a single document. Paste it into any AI conversation before you start writing. This becomes your master filter. ### Step 5: Using Your Voice Profile in Practice **For new writing sessions:** Open a fresh AI conversation, paste your combined filter document first, then give your writing prompt. The AI will have your voice and the AI tell rules loaded before it writes a single word. **For editing existing drafts:** Paste the filter document, then paste your draft and ask for the 7-pass review. The voice profile gives the AI a target to edit toward, not just patterns to edit away from. **For ongoing projects:** If you use AI projects or custom instructions, load your voice profile there so it's always active. You won't need to paste it every time. **Keep feeding it:** When you write something you're particularly happy with, add it to your transcript library. When you catch a new pattern the AI keeps getting wrong, add a rule to your profile. The profile is a living document. ### The Voice Training Prompt (Quick Reference) If you want to build a voice profile in a single session without the full process above, use this condensed prompt. It won't be as detailed as the transcript-based approach, but it's a solid starting point: ``` I need you to help me build a voice profile for AI-assisted writing. I'm going to describe how I write and talk, and I want you to turn it into a structured reference document. Ask me the following questions one at a time, and after I've answered all of them, compile my answers into a voice profile document: 1. Read me a paragraph of something you've written that sounds most like you. (Paste it in.) 2. What words do you use all the time? What words would your friends or colleagues say are "your" words? 3. When you explain something to a friend, do you tend to use long connected sentences or short direct ones? 4. How do you start stories? Do you set up context or drop right into the action? 5. When you make an important point, do you build to one big moment or keep circling back to it from different angles? 6. What kind of writing makes you cringe? (Corporate speak, motivational poster language, academic jargon, etc.) 7. Do you use humor? If so, what kind - dry, self-deprecating, playful, sarcastic? 8. When you're being vulnerable or serious, are you dramatic about it or matter-of-fact? 9. What punctuation or formatting quirks do you have? (Dashes, ellipses, ALL CAPS for emphasis, etc.) 10. Pick three writers or speakers whose voice you'd say is in the neighborhood of yours. Not identical, just in the vicinity. After I answer these, compile them into a voice profile I can paste into future AI conversations. ``` ### What Good Voice Profiles Have in Common After building profiles across different people and writing styles, certain patterns emerge in the ones that actually work: **They're specific, not aspirational.** A good profile describes how you actually sound, not how you wish you sounded. If you say "um" and "like" when you talk, that's useful data. It means your writing voice should have casual connectors, not polished transitions. **They include what you don't do.** Knowing that you never use literary metaphors, or that you avoid emotional adjectives, or that you don't speak in fragments gives AI clear boundaries. The negative space matters as much as the positive. **They use real examples.** "Bob starts stories by dropping into a specific moment" is okay. "Bob starts stories by dropping into a specific moment: 'I was on a phone call with a friend yesterday morning' or 'So I sent off a letter via FedEx to the Chicago Bears today'" is much better. AI learns patterns from examples more than from descriptions. **They capture range.** Your voice in a casual conversation isn't identical to your voice in a keynote presentation. A good profile notes how your voice shifts across contexts while maintaining the same core patterns. **They evolve.** Your voice changes over time. Review your profile every quarter. Compare it against your most recent writing. Add new patterns, remove ones that no longer fit. --- ## WHEN TO USE EACH TOOL **Weekly newsletters, blog posts, articles, presentations:** Use the full 7-pass filter with your voice profile loaded. Pay special attention to the structure pass (Pass 3) and content depth pass (Pass 6), which is where longer AI content breaks down most visibly. **Emails to clients or prospects:** Use the quick-use version. It catches the most obvious patterns without over-processing short-form writing. **Social posts (LinkedIn, X, etc.):** Use the quick-use version. Also watch specifically for emoji-led bullets, the "It's not X, it's Y" formula, and generic motivational sign-offs. **AI-assisted writing sessions:** Paste your combined voice profile + filter as the first message in any AI conversation where you'll be drafting content. This sets both your voice and the filter as the operating standard for the entire session. **First time setup:** Start with the Voice Training System. Build your profile, test it, refine it 3-4 times, then combine it with the AI tell filter into your master document. This upfront investment saves hours of editing on every piece going forward. --- ## KEEPING THIS CURRENT AI tells evolve as models update. Review and update this filter quarterly. Patterns to watch for: - New vocabulary that becomes overused (track what words start feeling "AI-ish") - New structural patterns (the staccato false-profundity style is a recent addition) - Patterns that become so well-known they stop being tells (if everyone avoids "delve," it stops being a signal) - Your own unconscious adoption of AI patterns (research shows humans absorb AI vocabulary without realizing it) --- *This filter was developed through extensive testing across newsletters, coaching content, presentations, and business writing. It works with any AI writing tool and any genre of professional or personal writing.*