# anti-slop writing first pass

> Source: <https://gist.github.com/joshp123/3c60d8def6d30a2503d8a98575334702>
> Published: 2026-08-09 10:51:06+00:00

| # AI Writing Tells — Definitive Deny-List (as of August 2026) | |
| Guard file for OpenTrawl landing-page copy. Every future draft is checked against this list. | |
| The goal: copy that reads like a sharp human wrote it — plain, specific, opinionated, gov.uk-style. | |
| If a phrase, pattern, or rhythm below appears in a draft, rewrite it. Clusters of these are the tell; | |
| one stray word is not. The strongest single signal is **density + formulaic structure**, not any one word. | |
| Confirmed project-specific tells (from the human owner — treat as hard bans): | |
| - Overcompressed fragment stacking presented as punchy copy: "Your apps, in SQLite, on your disk. Your AI searches them. No cloud, no accounts." | |
| - Anthropomorphic punchline: "The archive already knew." | |
| - Mechanical phrasing: "copies your apps into local archives" | |
| - Sensationalist compression in general; noise captions; "X. Y. Z." rhythm | |
| --- | |
| ## 1. Phrase deny-list (exact words/phrases, grouped) | |
| ### 1a. The classic "AI vocabulary" core (2023–2025 era; still flagged everywhere) | |
| delve / delved / delving · tapestry (abstract: "rich tapestry of") · testament ("a testament to") · | |
| underscore / underscoring · pivotal · landscape (abstract: "the evolving landscape of") · | |
| intricate / intricacies · meticulous / meticulously · interplay · vibrant · crucial · | |
| enduring ("enduring legacy") · bolster / bolstered · garner / garnered · showcase / showcasing · | |
| foster / fostering · enhance / enhanced · additionally (sentence-initial) · furthermore · moreover · | |
| notably · comprehensive · multifaceted · nuanced ("nuanced understanding") · paradigm · | |
| transformative · leverage / leveraging · synergy · holistic · robust · streamline · | |
| utilize / utilizing · facilitate · elucidate / illuminate · encompassing · spearhead · | |
| invaluable · groundbreaking · cutting-edge · catalyze / catalyst · cornerstone · | |
| reimagine / reimagining · empower / empowering · harness / harnessing · navigate (abstract) · | |
| realm · poised · myriad · embark / embarking · grapple / grappling · commendable · noteworthy · | |
| align with · boasts (= has) · features (= has) · offers (= has) · serves as (= is) · stands as (= is) | |
| ### 1b. The 2026-era tells (Forbes/Jodie Cook, May 2026 — the current crop) | |
| quietly ("quietly building", "quietly dominating", "quietly transforming") · shift ("a shift in thinking", | |
| "the shift towards") · matters ("this matters because", "what matters here") · shape ("shapes the future", | |
| "shapes how we think") · land ("a message lands") · actually (as filler intensifier) · | |
| real ("real growth", "real value", "real impact") · earn ("earn the right to", "earn trust") · | |
| the work ("do the work", "trust the work") · hold ("hold space", "hold the line") · | |
| pull ("the pull of") · compound ("decisions compound") · signal ("send the signal") · | |
| built different / built for the bold / built to last | |
| ### 1c. Significance-inflation phrases (Wikipedia's #1 content tell) | |
| marking a pivotal moment · represents a significant shift · part of a broader movement · | |
| solidify its role as a regional hub · generated debate · shaped emerging policy discussions · | |
| prompted broader reflection · raising philosophical questions · highlights the enduring legacy · | |
| the transformative power of · contributes to the broader history · reflects the influence of · | |
| setting the stage for · key turning point · evolving landscape · focal point · indelible mark · | |
| deeply rooted · stands/serves as a testament · plays a vital/significant/crucial/pivotal/key role · | |
| underscores/highlights its importance · reflects broader trends · symbolizing its ongoing/enduring/lasting | |
| ### 1d. Promotional / travel-brochure language | |
| nestled in/within · in the heart of · breathtaking · must-visit · stunning · renowned · world-class · | |
| state-of-the-art · seamless · game-changing · unparalleled · rich cultural heritage · natural beauty · | |
| commitment to excellence · boasts a · vibrant · profound · exemplifies · diverse array · gateway to · | |
| captivates residents and visitors alike · acts as the gateway to | |
| ### 1e. Hedging stacks and weasel words | |
| could potentially · might possibly · could possibly · perhaps potentially · may potentially · | |
| it could be argued · one could argue · it is possible that · some might say · it depends · | |
| there's no one-size-fits-all answer · results may vary · experts argue · observers have cited · | |
| industry reports · researchers and conservationists · several sources/publications (when one is cited) · | |
| widely interpreted as · is described in scholarship as · maintains an active social media presence | |
| ### 1f. Filler / throat-clearing phrases | |
| in order to · due to the fact that · at this point in time · in the event that · has the ability to · | |
| it is important to note that · it is worth noting that · it should be noted that · | |
| in today's rapidly evolving [world/landscape/era] · at the end of the day · when it comes to · | |
| the fact of the matter is · in terms of · for the purpose of · in light of the fact that · | |
| in the realm of · at its core · first and foremost · last but not least · needless to say · | |
| in the wake of · as we move forward · going forward | |
| ### 1g. Chatbot conversational filler (chat artifacts, not copy) | |
| I hope this helps! · Let me know if you'd like... · Would you like me to... · Feel free to... · | |
| Don't hesitate to... · Happy to help! · Here is an overview/summary/breakdown · Of course! · | |
| Certainly! · Absolutely! · I'd be happy to... · Is there anything else... · Great question! · | |
| Excellent point! · You're absolutely right! · As an AI language model... · I'm sorry, but... | |
| ### 1h. Generic conclusion / future-facing optimism | |
| the future looks bright · exciting times lie ahead · continue this journey · journey toward excellence · | |
| step in the right direction · only time will tell · the possibilities are endless · poised for growth · | |
| watch this space · the future of [thing] · take your [thing] to the next level · don't get left behind · | |
| the choice is yours · take control of your [thing] | |
| ### 1i. Landing-page / dev-tool marketing slop (Sailop 40-banned list, 2026) | |
| effortlessly · seamlessly · streamline · unlock · empower · transform · revolutionize · supercharge · | |
| powerful · robust · cutting-edge · next-generation · innovative · comprehensive · game-changing · | |
| world-class · state-of-the-art · blazingly fast · AI-powered · AI-native · cloud-native · | |
| enterprise-grade · production-ready · built for the modern web · end-to-end · welcome to our platform · | |
| the future of [thing] · built for [vague audience] · the all-in-one [thing] · trusted by leading companies · | |
| used by thousands of developers · the #1 [thing] for [audience] · get started · sign up free · | |
| learn more · join us | |
| ### 1j. Announcement / setup phrases that promise more than they deliver | |
| Here's the kicker. · The best part? · But here's the thing. · Honestly? · I'm going to state this as clearly as possible · | |
| Here's the part most people miss · Here's the breakdown · Let's dive in · Let's break this down · | |
| Without further ado · As mentioned above · As we've seen · In conclusion · To sum up · To summarize · | |
| In summary · Overall | |
| ### 1k. Therapist-mode / unsolicited validation (2026 tell) | |
| You're not imagining it · You're not alone · You're not broken · Do you want to sit with that for a while? · | |
| Are you ready to go deeper? · It's okay to feel this way | |
| ### 1l. LLM-safe truths (true statements that teach nothing) | |
| Consistency is important. · Building relationships takes time. · Success requires hard work. · | |
| Communication is key. · The journey matters more than the destination. | |
| ### 1m. Knowledge-cutoff / source-gap disclaimers | |
| as of my last knowledge update · up to my last training update · while specific details are limited/scarce in the available sources · | |
| not widely documented/disclosed · based on available information · the subject likely... · | |
| maintains a low profile · keeps personal details private | |
| --- | |
| ## 2. Sentence-pattern deny-list (the structures) | |
| ### 2a. Negative parallelism / contrast reframe — THE #1 tell (Atlantic, Jul 2026; Forbes, May 2026) | |
| The most famous AI writing tic. Survives every prompt fix. Pangram estimates "Not just X but Y" | |
| appears ~3x more often in AI text than human text. Barron's found it quadrupled in corporate | |
| communications 2023–2025. Variants: | |
| - "It's not X; it's Y." / "It's not about X, it's about Y." | |
| - "This isn't a tactics problem, it's a positioning problem." | |
| - "Not just X, but also Y." / "not only X but Y" | |
| - "No A, no B, just C." ("No bag, no things, no armor, just me.") | |
| - "X rather than Y." / "not X, but Y" | |
| - "The target was never a man. The target was the truth." | |
| - "It's far more than X. Focus on Y." | |
| Examples: | |
| - "It's not just a win for the private bank—it's a win for the entire enterprise." | |
| - "This isn't about speed; it's about trust." | |
| - "No cloud, no accounts, no setup. Just your data, on your disk." ← project-adjacent, banned | |
| Human alternative: delete the dismissed half. Say only the thing you mean. | |
| - AI: "This isn't a storage tool; it's a memory." | |
| - Human: "It stores your data locally. That's the whole pitch." | |
| ### 2b. Tricolon / rule of three everywhere | |
| "adjective, adjective, adjective" and "phrase, phrase, and phrase" used to make thin analysis | |
| look complete. AI uses triads constantly; humans use them once, deliberately. | |
| Examples: | |
| - "Faster, smarter, better." / "Seamless, secure, scalable." | |
| - "Your data. Your rules. Your archive." | |
| - "Built for speed, built for scale, built for you." | |
| Human alternative: use one precise item, or two, or recast. Vary list lengths. | |
| - AI: "Private, portable, powerful." | |
| - Human: "Private by design. Portable by default." | |
| ### 2c. Staccato fragment stacking ("X. Y. Z." rhythm) — CONFIRMED PROJECT TELL | |
| Short fragments stacked as fake punch. The human owner flagged: "Your apps, in SQLite, on your disk. | |
| Your AI searches them. No cloud, no accounts." This is the AI "punchy" default: every sentence a | |
| sentence fragment, all same length, all same rhythm, zero connective tissue. | |
| Examples: | |
| - "Your apps, in SQLite, on your disk. Your AI searches them. No cloud, no accounts." | |
| - "No setup. No sync. No servers. Just works." | |
| - "Private. Local. Yours." | |
| Human alternative: vary the rhythm. Let one sentence run. Use a fragment only where it lands hardest. | |
| - AI: "Local. Private. Yours." | |
| - Human: "Everything stays on your disk. There is no cloud account to lose, and nothing to sync." | |
| ### 2d. Colon-heavy headline pattern ("X: why Y matters") | |
| Every headline/heading as "Noun phrase: claim about it." AI defaults to this for all headers. | |
| Examples: | |
| - "SQLite: why local storage matters" | |
| - "Privacy: the case for keeping data on-device" | |
| - "OpenTrawl: the archive that knows" | |
| Human alternative: vary heading shapes. Some are questions, some are plain statements, some are | |
| fragments, some are full sentences. Never all colons. | |
| - AI: "Archives: why they matter" | |
| - Human: "What an archive is for" / "Archives, and why you'd want one" | |
| ### 2e. Anthropomorphic punchline — CONFIRMED PROJECT TELL | |
| Giving the product/archive/software a knowing, human-like consciousness as a closer. | |
| "The archive already knew." The software "remembers", "understands", "knows", "sees". | |
| Examples: | |
| - "The archive already knew." | |
| - "Your data remembers what you forgot." | |
| - "The search engine understands what you meant." | |
| Human alternative: state what the software actually does, plainly. | |
| - AI: "The archive already knew." | |
| - Human: "Search runs against your local copy, so it finds what you actually saved." | |
| ### 2f. Superficial trailing "-ing" analysis | |
| A present-participle clause bolted to the end of a sentence to editorialize. | |
| Examples: | |
| - "...creating a lively community within its borders." | |
| - "...further enhancing its significance as a dynamic hub of activity and culture." | |
| - "...reflecting its continued relevance in the regional and national transportation landscape." | |
| - "...symbolizing the region's commitment to innovation." | |
| - "...cementing its status as..." / "...ensuring..." / "...fostering..." | |
| Human alternative: delete the trailing clause. The sentence stands without it. If the point is real, | |
| make it its own plain sentence with attribution. | |
| ### 2g. "Despite X, Y faces challenges... but the future is bright" formula | |
| The rigid "Challenges / Future Prospects" outline closer. | |
| Examples: | |
| - "Despite its success, the Panama Canal faces challenges, including... Future investments could enhance the canal's efficiency." | |
| - "Despite their promising applications, pyroelectric materials face several challenges... positioning them as critical components for sustainable energy solutions." | |
| Human alternative: keep any concrete, sourced challenge as plain prose where it belongs. Delete speculation about the future. | |
| ### 2h. Canned notability / media-coverage hammering | |
| "featured in Vogue, Wired, and other media" · "profiled in multiple high-quality, independent, and widely-read outlets" · | |
| "these sources provide significant, substantial, secondary coverage" · "maintains an active social media presence" | |
| Human alternative: state facts, not notability arguments. | |
| ### 2i. Rhetorical question immediately answered | |
| "Have you ever wondered...?" → answer. "What is X?" → definition. "Imagine a world where..." → pitch. | |
| Human alternative: start with the substance. | |
| ### 2j. The sandwich conclusion (restate intro + generic optimism + CTA) | |
| "As we've seen... In conclusion... The future is bright... So next time you [topic], remember..." | |
| Human alternative: end on a strong, specific thought. Give the reader something new, not a summary. | |
| --- | |
| ## 3. Rhythm / structure tells | |
| ### 3a. Metronomic sentence length | |
| Every sentence 15–25 words, same beat. Read aloud: da-DUM da-DUM da-DUM. Humans mix aggressively: | |
| one word, then a long winding sentence, then a fragment. | |
| Fix: vary sentence length dramatically; start sentences with And/But/So; use fragments on purpose. | |
| ### 3b. Uniform paragraph length | |
| Every paragraph 3–5 sentences, every section roughly the same length. AI gives every point equal | |
| weight. Humans go deep on what matters and skip the rest. | |
| Fix: one paragraph may be a single sentence; the next may be eight. Sections of 50 words and 300 words. | |
| ### 3c. Perfectly parallel bullet lists | |
| Every bullet the same grammatical shape, same length, same structure. AI's signature. | |
| Fix: asymmetric lists — some items one word, some a sentence. If you wouldn't bullet it to a friend, don't bullet it here. | |
| ### 3d. Inline-header vertical lists ("**Thing:** description") | |
| `- **Conflict of Interest:** ...` `1. **Historical Context:** ...` — the AI list format. | |
| Fix: prose where the list isn't needed; clean lists without bolded inline headers otherwise. | |
| ### 3e. Em-dash density | |
| AI (especially Claude-family) uses em dashes far more than non-professional human writing, in places | |
| humans would use commas, parentheses, colons, or full stops. A July 2026 study: of contemporary | |
| models, only Claude used em dashes more than professional writers. OpenAI famously struggled for | |
| years to train the em dash out of ChatGPT. | |
| Fix: replace most with commas/parentheses/colons/full stops. Keep a few — zeroing them out is its own tell. | |
| ### 3f. Boldface overuse | |
| Bolding every instance of a chosen term; "key takeaways" bolding scattered through paragraphs. | |
| Fix: bold sparingly, for actual emphasis only. | |
| ### 3g. Title-case headings everywhere | |
| "Impact Of Technology And Digitalization" style. Fix: sentence case. | |
| ### 3h. Excessive internal coherence | |
| "Excessive internal coherence reveals the machine behind the words" (Forbes, Feb 2026). AI references | |
| its own earlier points too neatly, loops back to the intro's framing, weaves threads too perfectly. | |
| Humans leave threads hanging, forget what they said, contradict themselves slightly. | |
| Fix: leave some loose ends. | |
| ### 3i. Too clean to be human | |
| Every detail serves the argument; every example fits perfectly; no accidental details, throwaway | |
| numbers, or oddly specific but unimportant facts. Fix: add some irrelevant specifics. Real life has loose ends. | |
| ### 3j. Missing emotional spikes | |
| AI maintains a neutral emotional temperature even when the topic demands a stance. Fix: get annoyed, | |
| get excited, take a position that might alienate someone. | |
| ### 3k. Faux balance without consequence | |
| Gestures at "both sides" then proceeds exactly as planned; nothing is weighed, nothing is traded off. | |
| Fix: if you acknowledge a counterargument, actually engage it or cut it. | |
| ### 3l. Arguments that teleport | |
| Mid-paragraph the logic jumps; one sentence makes a point, the next states a conclusion without | |
| showing how it was reached. Fix: read each paragraph asking "does this sentence follow from the last?" | |
| ### 3m. Near-miss metaphors | |
| Comparisons that sound clever but don't map ("tuning a guitar" for a business strategy). Fix: if a | |
| metaphor misses even slightly, remove it entirely. | |
| ### 3n. Correct words in wrong contexts | |
| Formal verbs in casual contexts; slightly clinical word choices in warm messages; grammatically | |
| perfect but textbook-learned. Fix: use the word you'd actually say. | |
| ### 3o. Curly quotes/apostrophes (ChatGPT/DeepSeek habit) | |
| Curly " " ' ' where the document otherwise uses straight marks, or inconsistent mixing. | |
| Fix: normalize to straight marks for plain text. (Gemini and Claude typically don't do this.) | |
| ### 3p. Emoji as formatting (older models; still seen in comments/microcopy) | |
| 🧠 🚨 🔭 in front of headings/bullets; "👉 Drop me a message". Fix: none in product copy. | |
| ### 3q. Thematic breaks before headings (----) and skipped heading levels | |
| Markdown habits. Fix: proper heading levels, no rules before headings. | |
| --- | |
| ## 4. Model-specific sections | |
| ### 4a. GPT-series tells (GPT-4 / 4o / 5 era) | |
| - **Structural predictability**: numbered lists, consistent headings, step-by-step organization; every | |
| list item grammatically identical. | |
| - **Formulaic openers**: restating the question ("Certainly! Here is a comprehensive guide to..."), | |
| "Let's explore" as a section opener. | |
| - **Formulaic conclusions**: restate main points, note the topic is nuanced, encourage further research. | |
| - **Era vocabulary** (Wikipedia's breakdown): | |
| - GPT-4 (2023–mid-2024): additionally, boasts, bolstered, crucial, delve, emphasizing, enduring, | |
| garner, intricate/intricacies, interplay, key, landscape, meticulous/meticulously, pivotal, | |
| underscore, tapestry, testament, valuable, vibrant | |
| - GPT-4o (mid-2024–mid-2025): align with, bolstered, crucial, emphasizing, enhance, enduring, | |
| fostering, highlighting, pivotal, showcasing, underscore, vibrant | |
| - GPT-5 (mid-2025+): emphasizing, enhance, highlighting, showcasing (+ canned notability claims) | |
| - **Curly quotes/apostrophes** (ChatGPT habit). | |
| - **The em-dash struggle**: OpenAI spent years trying to train the em dash out; GPT output historically | |
| em-dash-heavy, now less than Claude but still formulaic in placement. | |
| - **Sycophancy**: top models affirm users' ideas ~49% more than humans (Stanford/CMU, Mar 2026); | |
| "Great question!", "You're absolutely right!", "I'd be happy to...". | |
| - **Chat artifacts**: `:contentReference[oaicite:0]{index=0}`, `oai_citation`, `turn0search0`, `+1` after sentences. | |
| ### 4b. Claude-series tells (Claude 3/4 era; Fable inherits these) | |
| - **Over-qualification and meta-commentary**: "It's worth noting that", "This is a nuanced area", | |
| "The answer depends on context" — useful qualifications that become obvious patterns every few paragraphs. | |
| - **Hedging before claiming**: Claude hedges before making the claim at all; human writers make claims | |
| and defend them. | |
| - **Em-dash overuse**: the July 2026 study found Claude the only contemporary model using em dashes | |
| more than professional writers. "This matters — and here's why." | |
| - **Longer, more syntactically complex sentences** than average human writing. | |
| - **Earnest, measured tone**: acknowledges limitations and counterarguments even when unnecessary; | |
| reads intellectually cautious rather than authoritative. | |
| - **"I appreciate"** openers and polite framing in conversational output. | |
| - **Curly quotes**: Gemini and Claude typically do NOT use curly quotes (so straight quotes are not a | |
| Claude tell). | |
| ### 4c. Fable-series tells (Anthropic's Fable 5, June 2026 — public info is THIN) | |
| What we found: | |
| - **Fable 5 is Anthropic's flagship** (released June 9, 2026; "Claude Fable 5" and "Claude Mythos 5"; | |
| API name `claude-fable-5`). It is the successor to Claude Opus, positioned for long-horizon reasoning | |
| and high-fidelity creative writing. | |
| - **No public "Fable-isms" catalog exists yet** as of August 2026 — the model is ~2 months old and the | |
| discourse is about its capabilities (large context window, "adaptive thinking", "voice locking"), | |
| not its tells. | |
| - **Fable inherits Claude-family tendencies** (same lineage): hedging, meta-commentary ("worth noting"), | |
| em-dash overuse, over-qualification. Treat the Claude list as the Fable baseline. | |
| - **"Voice locking"** means Fable can be steered to mimic a given style from samples — so its tells are | |
| less consistent than older models when prompted with style examples. Cold output still defaults to | |
| Claude-family patterns. | |
| - **Watch for**: the same negative-parallelism and tricolon structures as every other model (Pangram | |
| confirms all major chatbots rely on them), plus Anthropic's characteristic earnestness. | |
| - Recommendation: keep the Claude list as the Fable deny-list until public Fable-specific tells emerge; | |
| re-check quarterly. | |
| --- | |
| ## 5. What a human writes instead (per major pattern) | |
| | AI pattern | Human alternative (plain, specific, gov.uk-style) | | |
| |---|---| | |
| | Negative parallelism: "It's not just a tool, it's a memory." | "It stores your data locally. That's the whole pitch." | | |
| | Tricolon: "Faster, smarter, better." | "It's fast. The rest you can judge for yourself." | | |
| | Staccato stacking: "Local. Private. Yours." | "Everything stays on your disk. There is no cloud account to lose, and nothing to sync." | | |
| | Colon headline: "SQLite: why local storage matters" | "What an archive is for" | | |
| | Anthropomorphic punchline: "The archive already knew." | "Search runs against your local copy, so it finds what you actually saved." | | |
| | Significance inflation: "marking a pivotal moment in the evolution of regional statistics" | "The institute was established in 1989 to collect and publish regional statistics." | | |
| | Trailing -ing analysis: "...reflecting its continued relevance in the transportation landscape." | Delete the clause. The sentence stands without it. | | |
| | "Seamlessly unlock the power of AI to effortlessly transform your workflow." | "Catch the 40 phrases that mark your copy as AI-written, with a one-line replacement for each." | | |
| | "Effortlessly streamline your workflow" | "No more reading 200-line PR diffs." | | |
| | "Seamlessly integrates with your CI" | "Plugs into your CI as a single GitHub Action." | | |
| | "Unlock 50 design templates" | "Get 50 design templates." | | |
| | "Empowers your team to ship" | "Ship a landing page without a designer." | | |
| | "Trusted by leading companies" | "Used at Stripe to gate the design-system PR." | | |
| | "Get started" | "Install via npm." / "Scan my codebase." | | |
| | "The future looks bright" | Cut it. End on a specific thought. | | |
| | "Great question!" | Start with the substance. | | |
| | "In today's fast-paced world" | Cut it. Start with the fact. | | |
| | "It's worth noting that..." | Cut it. Note the thing. | | |
| | "Consistency is important." | "Consistency is important" → cut, or sharpen: "Ship every Friday, or don't ship at all." | | |
| | "Here's the kicker." | Delete the fanfare; let the substance speak. | | |
| | "Do the work." | "Write the copy. Then cut half of it." | | |
| The universal rule (Sailop): replace every adjective with a number, every adverb with a comparison, | |
| every vague verb with a concrete one. The rewrite is usually shorter and always clearer. Write copy | |
| you would defend in a pitch meeting — AI copy is what survives committee review because nobody | |
| disagrees with it. | |
| --- | |
| ## 6. Sources | |
| Primary field guides: | |
| 1. Wikipedia: "Signs of AI writing" (WikiProject AI Cleanup field guide; era vocab, negative parallelisms, | |
| rule of three, em-dash, boldface, inline-header lists, promotional language, challenges-formula). | |
| https://en.wikipedia.org/wiki/Wikipedia:Signs_of_AI_writing | |
| 2. The Atlantic, Will Oremus: "The Most Famous AI Writing Tic Is Also the Most Mysterious" (Jul 2026) — | |
| negative parallelism as the #1 tell; Pangram 3x statistic; Barron's 4x statistic. | |
| https://www.theatlantic.com/technology/2026/07/ai-chatbot-writing-tic-negative-parallelism/687892/ | |
| 3. The Atlantic, Eve Fairbanks: "The Biggest Tell That Something Was Written by AI" (May 2026) — | |
| canned perfection, uneditable AI prose, sycophancy (Stanford/CMU 49% study), Max Planck word-leak study. | |
| https://www.theatlantic.com/technology/2026/05/how-to-tell-ai-writing/687345/ | |
| 4. Forbes, Jodie Cook: "15 New Giveaway Signs Of AI Writing (May 2026 Update)" — the 2026 word crop | |
| (quietly, shift, matters, shape, land, actually, real, earn, the work, hold, pull, compound, signal, | |
| built different) + contrast reframes as the surviving #1. | |
| https://www.forbes.com/sites/jodiecook/2026/05/21/15-new-giveaway-signs-of-ai-writing-may-2026-update/ | |
| 5. Forbes, Jodie Cook: "The 15 New Giveaway Signs Of AI-Generated Content (In February 2026)" — | |
| announcing-the-good-part, LLM-safe truths, "Honestly?", quiet, therapist mode, marching sentences, | |
| near-miss metaphors, teleporting arguments, missing emotional spikes, faux balance, too-tidy internal | |
| references, too-clean-to-be-human. | |
| https://www.forbes.com/sites/jodiecook/2026/02/03/the-15-new-giveaway-signs-of-ai-generated-content-in-february-2026/ | |
| 6. Sailop: "40 Banned Phrases That Mark Your Copy as AI-Generated (with Replacements)" (Apr 2026) — | |
| landing-page-specific deny-list with replacements and the "tells travel in packs" rule. | |
| https://www.sailop.com/blog/banned-phrases-40-words-mark-copy-ai-generated | |
| 7. Ars Technica, Benj Edwards: "Wikipedia volunteers spent years cataloging AI tells. Now there's a | |
| plugin to avoid them." (Jan 2026) — the Humanizer plugin; before/after example. | |
| https://arstechnica.com/ai/2026/01/new-ai-plugin-uses-wikipedias-ai-writing-detection-rules-to-help-it-sound-human/ | |
| 8. texttoolsai.app: "How ChatGPT and Claude Writing Differs — and How to Humanize Each" (May 2026) — | |
| GPT vs Claude tells. | |
| https://www.texttoolsai.app/blog/chatgpt-vs-claude-writing-quality | |
| 9. The Atlantic: "Chatbots Are Spewing Out All the Worst Clichés" (Feb 2023) — cliché obsession. | |
| https://www.theatlantic.com/technology/archive/2023/02/ai-chatbots-cliche-writing/673143/ | |
| 10. MIT Technology Review: "Why detecting AI-generated text is so difficult" (Feb 2023). | |
| https://www.technologyreview.com/2023/02/07/1067928/why-detecting-ai-generated-text-is-so-difficult-and-what-to-do-about-it/ | |
| 11. The Guardian: "Spam, junk ... slop?" (May 2024) — the slop term. | |
| https://www.theguardian.com/technology/article/2024/may/19/spam-junk-slop-the-latest-wave-of-ai-behind-the-zombie-internet | |
| 12. NYMag/Intelligencer, Max Read: "Drowning in Slop" (Sep 2024). | |
| https://nymag.com/intelligencer/article/ai-generated-content-internet-online-slop-spam.html | |
| 13. Wikipedia: "AI slop". | |
| https://en.wikipedia.org/wiki/AI_slop | |
| 14. Humanizer skill files (open-source, distilled from the Wikipedia guide): | |
| brandonwise/humanizer (29 patterns, weighted), WhimseyAI/humanizer-skill, Intern4711/Better_Humanizer. | |
| 15. Anthropic: "Claude Fable" product page + platform docs (Fable 5 / Mythos 5, Jun 2026); Bilzit and | |
| NovelMage Fable-5 reviews (voice locking, adaptive thinking) — Fable context. | |
| https://www.anthropic.com/claude/fable | |
| Note on method: parallel.ai search was rate-limited for this session; sources were gathered via direct | |
| fetch, Google News RSS, Wayback CDX, and DuckDuckGo-through-reader. Fable-specific tells are thin | |
| publicly as of Aug 2026 — the Claude list is the best available proxy, per section 4c. |