This is everything we checked while working out how to make AI sound more human. That article is the part to read. This page is for looking things up: every tool, repo and study we found, what the evidence says about each one, and a link so you can check it yourself.
Claude did the research, and the links were checked against their sources on Nov 1, 2026. Prices, star counts and rankings are from that date, so expect some of them to have moved.
Tools, Skills and Repos
These are Tavi’s recommendations: each group is sorted into the ones worth starting with and the ones just to know about, with a clear standout marked as the top pick. Stars (GitHub’s version of a like) and last-update dates are from GitHub on Nov 1, 2026, and licenses come from each project’s own files. “None found” means we found no published evaluation, outside or by the maker, as of that date.
Humanizer skills and plugins for Claude and other AI agents (15)
Humanizer Skills and Plugins for Claude and Other AI Agents
| Name | What it does | What the evidence says | Cost |
|---|---|---|---|
| Start here | |||
| blader/humanizer Top pick | 26 AI patterns in six groups, strongest first, with a draft, critique and final pass | Why start here: it’s the only one with an outside judge or reader test, and the most widely used. The only one we found with a judge or reader test from outside: won 16 of 16 on quality with AI judges, detection stayed at 98-100%, and judges noticed a more uniform rhythm afterward; 63k stars | Free (MIT) |
| mshumer/unslop | Has the model write about 50 samples in your field, finds what it repeats and writes an avoid-list; run it again when the model changes | Why start here: it finds what your own model repeats, which a fixed list can’t do. None published | Free (MIT) |
| work-humanizer | Rewrite mode plus an interview mode that asks you up to five questions, with a fact check at the end | Why start here: it asks you questions, so the draft gets a person’s input. Makers’ own test with AI judges: picked 48% of the time against blader’s 20%, with fewer changes in meaning; read as human on Pangram 4 of 4 times when given the writer’s notes | Free (MIT) |
| Every’s Compound Writing | A home for each writer’s voice, style, audience and examples, plus reviewer agents, a check of the draft against its sources and a flag for wording copied too closely from examples | Why start here: it checks the draft against your sources, which most of these skip. None found | Free (MIT, per its README) |
| Just to know about | |||
| stop-slop | Eight core rules (cut all adverbs, no em dashes) and a 50-point score | None found; 18.5k stars; no update since Mar 2026 | Free (MIT) |
| no-ai-slop | Removes 20+ patterns, with edit, detect and satire modes; also a ChatGPT plugin | None found; 13k stars | Free (MIT) |
| avoid-ai-writing | Rewrite, detect and edit modes, five voice profiles and a bundled detector | Only the maker’s own calibration file | Free (MIT) |
| sepia | For fiction, works on story structure first, then flow, then wording, choosing three to five changes per story; other prose gets rule files matched to where it will run | Has a way to evaluate itself, with no results published | Free (MIT) |
| theclaymethod/unslop | Scans phrases, structure and how ideas are arranged (endings that call back, previews the body repeats) | Its makers’ latest result says not good enough to release, with no steady gain across different AI models | Free (MIT, per its README) |
| humanizer-stack | A wording pass, then six structure checks on an outline of the draft, one at a time | None found | Free (MIT, with some CC BY-SA material) |
| agent-style | 21 writing rules loaded before writing | Makers’ own test: counted tells fell 46-86% on three of four tools; quality not measured | Free (CC BY 4.0 rules, MIT code) |
| humanize-writing-skill | Rules on claims, sentence shape and endings | Makers’ own small test: Claude judges preferred it in 5 of 6 and 3 of 3 pairs; says it doesn’t change detector scores | Free (MIT) |
| writing-style-skill | You edit the AI’s drafts, and it extracts rules from your edits | None for the tool; the idea worked with simulated users in a 2024 study | Free (MIT, per its README) |
| nobuzz, vomit and SlopMonster | Rewrite Claude’s output through a second model (Gemini, a model on your own computer, or a rival model) | None with readers; the first model’s style survives rewording, and vomit warns it can make things up | Free |
| Anthropic’s brand-voice plugin | Built by a partner (Tribe AI): finds a brand voice in your documents and checks content against it | None published | Free (MIT) |
Open-source tools for models you run yourself (7)
Open-Source Tools for Models You Run Yourself
| Name | What it does | What the evidence says | Cost |
|---|---|---|---|
| Start here | |||
| antislop-sampler Top pick | Backs up and retries when the model writes a banned word, phrase or pattern | Why start here: it’s the one with a paper behind it, and it handles far more patterns than plain bans. The Antislop paper: handles 8,000+ patterns where plain token banning fails at about 2,000; needs a model whose word probabilities you control | Free (Apache-2.0) |
| slop-forensics | Finds each model’s overused words and phrases | Why start here: it shows what a model overuses before you decide what to ban. The source of EQ-Bench’s phrase lists | Free (MIT) |
| Just to know about | |||
| auto-antislop | Finds a model’s stock phrases and trains them out | The paper reports about 90% less slop with writing scores kept | Free |
| KoboldCpp and SillyTavern | Ban phrases or word fragments while a local model writes | No quality tests | Free (AGPL-3.0) |
| Unslopper-30B-A3B | A model trained to rewrite fiction without AI phrasing | Maker’s own numbers: Pangram’s human score rose from 0.000 to 0.481 while judged quality fell from 8.60 to 7.96 | Free |
| logitloom and loom-mcp | Generate many continuations for a person to pick from; logitloom works through some hosted model services | None; loom-mcp depends on prefill (writing the first words of Claude’s reply), which current Claude models reject | Free |
| pseudobibeR | Counts grammar features to compare a draft with a person’s own writing | No study links its counts to readers’ judgments | Free |
AI word lists and style guides (12)
AI Word Lists and Style Guides
| Name | What it does | What the evidence says | Cost |
|---|---|---|---|
| Start here | |||
| Wikipedia: Signs of AI writing Top pick | Volunteer editors’ guide to AI tells by model era, with sections on signs of human writing and signs that don’t work | Why start here: it’s the most complete list, it’s sorted by model era, and it’s kept current. A long guide, edited almost daily; observations with footnotes, no accuracy numbers for its signs, though it quotes one 2025 preprint’s figures for expert readers (90% caught, 3.3% misidentified) | Free |
| Anthropic’s prompting guides | Say what you want, name specific patterns and match the prompt’s style to the output; the Fable 5.1 guide defines mannered prose, and the Opus 5.5 guide says to name patterns to avoid (in its web-design section) | Why start here: it comes from the people who make Claude, and it says what to write, not only what to avoid. The company’s own guidance, with no numbers | Free |
| Gwern, “Better LLM writing” | Have the model suggest chatbot-style edits, then do the opposite; generate many options and pick | Why start here: its idea of generating many options and picking one is the most useful habit on this list. One practitioner; older models | Free |
| Just to know about | |||
| WikiProject AI Cleanup guide | How Wikipedia’s editors find and clean up AI text | Community guidance; rates Pangram among the most reliable detectors despite its misses | Free |
| EQ-Bench Slop Score (code) | Scores text against 1,648 words, 430 three-word phrases and “not x but y” patterns | No reader test; its own page says it isn’t an AI detector | Free |
| Pangram’s signs of AI writing | A detector company’s rates for nine surface signs | The company’s own analysis; it says these signs aren’t what its detector looks at | Free |
| tropes.fyi | 49 AI writing tropes in six groups, tagged as rising or fading | Shows no source data or dates | Free |
| jola.dev on Claude’s phrases | Claude’s pet phrases plus a hook that swaps them on screen | The hook changes only what you see; Claude still writes them | Free |
| sabrina.dev’s prompt and inaiwetrust’s list | Prompts with long banned lists (65 entries on sabrina.dev, 108 on inaiwetrust), including “can”, “may”, “might” and “tends to” | No tests; they ban hedges that human writing uses, and sabrina.dev’s bans semicolons and formatting, then suggests a semicolon and asks for bullets | Free |
| Zack Proser’s scrub pass | Run the anti-slop pass last, and have a different model family do the rewrite than the one that drafted | One practitioner, no reader test; he warns that a draft that says nothing can still pass a scanner | Free |
| Simon Willison on AI writing | Lets AI proofread, but won’t let it speak for him in opinions or first person | One practitioner’s rule | Free |
| WRITING.md | Rules that change by type of writing, with examples and a self-check | None found | Free (MIT) |
AI humanizer apps and writing apps (10)
Most humanizer apps are built to get past detectors, which I wouldn’t make a goal; they’re listed so you know what they are. Their claims are their own.
AI Humanizer Apps and Writing Apps
| Name | What it does | What the evidence says | Cost |
|---|---|---|---|
| Start here | |||
| Spiral (from Every) Top pick | A writing app that picks your relevant samples for each new draft | Why start here: it works from your own samples and isn’t built around getting past detectors. The company’s own eval, with an AI judge trying to spot the draft among the person’s real samples: Spiral’s drafts blended in 87% of the time | From $15/month |
| Grammarly AI Humanizer | Rewrites in a chosen voice; says it isn’t meant to get past detectors | Why start here: there’s a free version, and it says it isn’t meant to get past detectors. Passed Pangram 0 of 30 times on one vendor-run leaderboard | Free basic version |
| Just to know about | |||
| Undetectable AI | Rewrites text to pass detectors and offers a money-back guarantee if output is flagged | One study found under 4% still flagged after it; leaderboards run by vendors disagree widely | From $5/month, billed yearly |
| WriteHuman | Rewrites text; says its model is updated for GPTZero and Pangram | Only leaderboards it runs or a rival runs | $12-$39/month yearly, $20-$59 monthly |
| StealthGPT | A rewriter built to get past detectors | After it, Pangram stayed accurate and GPTZero missed about half (University of Chicago test) | $25-$150/month yearly, $30-$180 monthly |
| QuillBot AI Humanizer | A rewriter with Basic humanizing on its free tier and Advanced humanizing on Premium | No independent quality test | Free tier; paid plans |
| ChatGPT’s AI humanizer page | Shortens, clarifies and rewords inside ChatGPT | None | Free with ChatGPT |
| Humanizer GPTs in the GPT Store | Creators’ own humanizer prompts | None; custom GPTs are scheduled to retire Dec 11, 2026, later for some business workspaces | Depends on your plan |
| Leaderboard humanizers (Phrasly, HumanizeMyPaper, HIX Bypass, Rephrasy and others) | Rewriters built to get past detectors | Ranked only by HumanizerBench (run by WriteHuman), HumanizerEval (data hosted by StealthGPT) and a Hugging Face page with affiliate links | Varies |
| Brand-voice features in Copy.ai, HubSpot and Hypotenuse | Match a voice from writing samples (Copy.ai recommends 3,000+ words if possible; HubSpot asks for 2 or 3 samples, of at least 500 words for blogs and pages or 150 for email, texts and social); Hypotenuse says it trains a model on how you write | No evidence the sample thresholds were tested | HubSpot: Professional and Enterprise plans; the other two pages name no plan |
Checkers and linters for AI writing (7)
A linter is a program that flags problems in text against a set of rules.
Checkers and Linters for AI Writing
| Name | What it does | What the evidence says | Cost |
|---|---|---|---|
| Start here | |||
| Vale with vale-ai-tells Top pick | A prose linter plus a pack of AI-tell rules | Why start here: it’s an established linter, and the rule pack adds AI tells you can edit. None for AI tells | Free (MIT) |
| Hemingway Editor | Highlights hard-to-read sentences; the paid tier rewrites them | Why start here: it’s free, it’s easy to try, and it flags the hard sentences a person should look at. None for AI tells | Free; paid tier with a trial |
| Just to know about | |||
| proselint | Flags weak or awkward prose | None for AI tells | Free (BSD-3-Clause) |
| sloptrim and slop-guard | Score prose files for AI tells | None found | Free |
| LLM Slop Detector (Firefox) | Flags invisible characters, em dashes, curly quotes and stock phrases on web pages | None found; 4 daily users | Free (MIT) |
| Slop or not | A game where you pick the AI reply out of a pair | No accuracy table, and the players aren’t a random sample | Free |
| EditLens | A research model that estimates how much AI editing a text has | Only its authors’ own test, and three of its four authors work at Pangram | Free for non-commercial use (CC BY-NC-SA 4.0) |
More tools, repos and lists (46)
Everything else the research turned up, for completeness. We found no judge or reader test of any of these from outside their makers; where a maker publishes its own test, the row says so.
More Tools, Repos and Lists
| Name | What it does | What the evidence says | Cost |
|---|---|---|---|
| EveryInc/draft-review-kit | Every’s draft review tools | None found | Free (MIT) |
| MohamedAbdallah-14/unslop | Strips flattery, stock words, stacked hedges and piles of em dashes; leaves code, links and headings alone | Maker’s own blind test with one judge model (Claude Sonnet 4.5): its rewrite was preferred in 21 of 21 judgments; the maker says it doesn’t fool GPTZero | Free (MIT) |
| ZeroSlop | Finds and removes AI slop | Maker’s own test, scored by its own tool: a writing score of 12.8 against 23.3 to 35.4 for three other skills (lower is better) | Free (MIT) |
| content-skills | Five skills, including an anti-AI-writing skill and a voice guide | None found | Free (MIT) |
| Smaller pattern-list skills: slopkit, anti-ai-slop, asavvin-pixel/unslop, AgriciDaniel/anti-slop, quiron, tagore, better-writing, deslopify | Mostly the same pattern-list approach; quiron measures the rewrite against human writing, asavvin-pixel’s checks whether the outline is too tidy, AgriciDaniel’s finds substance problems and never reports authorship, slopkit adds a skill for the agent’s own replies, and deslopify is for Gemini’s command-line tool | Makers’ own tests for slopkit (constructed cases, it says, not proof), quiron (a rewrite inside its human ranges was still judged AI 12 of 12 times) and better-writing; anti-ai-slop says rewrites that scored 100 on its linter still had details the original never mentioned, and tagore shows scored before-and-after examples; none found for the rest | Free |
| Very small skills: theaimillennial-git/humanizer, humanize-text, humanize-ai-writing, avectats7/anti-ai-writing, anti-ai-writing-humanizer, gamper96/anti-ai-writing, anti-ai-writing-style | Variations on the same idea | None found | Free |
| Voice-profile builders: voice-profile-builder, voice-dna-builder, voice-calibration-plugin, voice-suite | Build a voice profile from your writing (paultaki’s uses only writing from before AI) | None found | Free (MIT) |
| Skills for one channel: linkedin-agent-skill, attention-span | LinkedIn posts with a humanize step, and easier-to-read agent replies | None for AI tells | Free |
| Mikefluff/skills | 45 Claude Code skills, Russian-first, including prose editing with an offline linter | None for AI tells | Free |
| More languages: shuorenhua, qu-ai-wei, ilyautov/humanizer-ru, smixs/humanizer-ru, stop-ai-slop-jp, patina, humanizer-zh-next, ai-zixun/humanizer-zh, VincentOld/stop-slop-zh, stop-slop-ko, pencil20388-eng/stop-slop-zh | Chinese, Russian, Japanese, Korean and mixed-language versions of the same approach | Makers’ own tests: shuorenhua runs 123 cases before each release (for v2.6.0, none of 58 clean items damaged, 5 of 7 read better, 1 of 123 lost information); patina links its own pre-registered study, which it says shows where rewriting fails; smixs/humanizer-ru has 19 test scenarios and 127 checks and releases only when all pass; ilyautov/humanizer-ru publishes its calibration of its scanner against Russian text sets; none found for the rest | Free (MIT) |
| More academic skills: humanizer_academic, skill-deslop, scipilot-writing-skill, paper-humanizer-skill, academic-deai | Papers and medical writing, scientific writing, and academic editing in English and Chinese; scipilot includes an AI-tell checker | None found | Free |
| Claude Code output styles | A custom style leaves out Claude Code’s built-in coding instructions by default (a setting keeps them) | Documented only | Part of Claude Code; the page states no price |
| Skill directories: skills.sh, claudskills.com, skillsmp.com | Browse skills; skillsmp.com indexes them without installing or certifying them | skills.sh shows install totals; claudskills.com shows security grades and installs for theme packs, with no public quality score (it offers one to Pro users) | Free to browse; claudskills.com’s one-click install is Pro ($9/month, $79/year or $149 once) |
| antislop-vllm | Writes in chunks with checks, for models you host | Description page only | Free |
| llama.cpp and vLLM | Engines for running open models, with settings that push chosen words down | None for AI tells | Free |
| Qwen3-4B “deslop” model | A small model trained to remove stock phrasing | Maker’s own numbers from its own detector: slop score from 121 to 12.5 per 1,000 words on 40 held-out prompts, em dashes from 20.5 to 0; the post says no reader test has checked it | Free (Apache-2.0) |
| prose-rewriter-1.7b-v2.4 | Rewrites fiction paragraphs | Maker’s own numbers: still changed the meaning of 25.5% of paragraphs | Free |
| Quill-v1 and Darkest-muse-v1 | Fiction models from EQ-Bench’s author | None independent | Free |
| Built to get past detectors: lynote-ai/humanize-text, StealthHumanizer | Rewriters aimed at detectors (StealthHumanizer advertises bypassing GPTZero and Turnitin); Lynote also sells a detector and says no method makes text reliably undetectable | None found | Free |
| TextHumanize | Calls itself a style-normalization tool, not a detector bypass | Maker’s own results from its built-in detector | Free for personal and non-commercial use; commercial licenses from $29/month |
| Style-transfer research code: stylellm_models, DualRL, Text_Style_Transfer_Survey, deep-latent-sequence-model | Research code (2019-2024) and a reading list on rewriting text into another style | DualRL, the survey and deep-latent-sequence-model don’t mention AI-generated text; stylellm_models says it can restyle output from ChatGPT and similar models, without a test | Free |
| Prompt collections: academic-writing prompts, research-writing prompts | Large lists of prompts for academic and research writing | Not tested for AI tells | Free |
| awesome-slop and slop-gate | A curated list of slop research, linters, detectors and humanizers, and the same maintainer’s checker | None for the checker | Free |
| GitHub topic pages: humanize-text, text-humanizer, remove-ai-tells, anti-ai-writing | Lists of projects tagged with each topic | Not applicable | Free |
| “Anti-slop” projects for design or code: taste-skill, impeccable, hallmark, heydamir/unslop, miqdadbadjuber/anti-slop, dmmulroy/anti-slop, desloppify, peakoss/anti-slop | Popular projects with similar names that are mostly about web design or code; miqdadbadjuber’s also ships a copywriting skill | Mostly not for writing | Free |
| Hacker News thread on Claude’s phrases | Commenters add more of Claude’s pet phrases | No counts | Free |
| Techpresso prompt library | 24 copy-paste “humanizer” prompts | None | First 5 prompts free; the rest with a 7-day trial |
| Ruben Hassid on em dashes | Reports that asking alone doesn’t stop em dashes, and that you have to give ChatGPT an alternative | One practitioner’s report, screenshots only | Free; the post ends with a paid section |
| SillyTavern banned-phrase settings | One author’s own settings and banned-token list for local models | None | Free |
| Wikipedia’s “AI or not” quiz | Practice telling AI text from human text | Not reviewed | Free |
| writing-skills | Plain-language, business, GOV.UK, AP and simplified technical English writing standards as skills | None for AI tells | Free (MIT) |
| Clever AI Humanizer | Refines tone and clarity and makes no detector claim; its maker also offers a free AI detector on the same site | None found | Free, up to 3,000 words per request |
| humanizeai.pro | Claims to bypass all detectors (the vendor’s claim) | None found | Free to start; paid upgrade |
| Pressmaster | An AI interviewer that turns talk into content | Hume AI’s case study claims it made content creation 93% faster, with no method given | Not checked |
| detectorscope | Google DeepMind’s research detector built from named, checkable features | Makers’ own results on two benchmarks; its description says accuracy outside its training data isn’t established | Free (Apache-2.0) |
| lynote-ai/ai-text-detector | An open detector for AI agents, linked to a paid service | Run by a company that also sells a humanizer | Free (MIT) |
| Other open detectors: distil-ai-slop-detector, sloptotal, econ-ai-detector | A detector that runs in your browser, one that combines 23 detectors, and one for economics papers | Makers’ own results: distil-ai-slop-detector reports 100% accuracy for its full model on its own test set and about 95% for the version that runs in a browser; sloptotal 0.942 on its own benchmark (a score where 1 is perfect); econ-ai-detector reports false flags on 0.00% of about 1,370 held-out papers, 0.06% of about 5,300 published articles and 0.17% of 599 National Bureau of Economic Research papers, and says it flags 13.4% of that bureau’s 2026 papers | Free |
| Detector research code: AIGC_text_detector, ai-detection-paraphrases, Reliability-of-AI-text-detectors, slop-detector, human_detectors | Earlier detectors, a study showing rewording beats detectors, reliability studies, and the five-reader study’s data | Background only | Free |
| More Vale rule packs: krishnasunkam/vale-ai-tells, vale-llm-slop | AI-tell rules for the Vale linter | None found | Free |
| Small AI-tell scorers: slop-score-skill, SlopScoreTool, sloplint, snifftest, humanizer-lint, FeedHive’s ai-slop-detector, SLOP_Detector | EQ-Bench’s Slop Score as a skill or offline tool, pattern checkers (FeedHive’s has 120 patterns), and a word-list analyzer for AI training data | snifftest publishes its own test (it caught 63 of 80 planted faults and flagged 1 of 54 clean paragraphs); humanizer-lint has a benchmark of 20 synthetic test cases, which it says doesn’t measure real-world accuracy; SlopScoreTool describes its own evaluation in its methodology document; none found for the rest | Free |
| write-good | Simple English prose checks | None for AI tells; last published in 2021 | Free (MIT) |
| 2026-AI-DETECTOR-BENCHMARK | A benchmark published by GPTHumanizer’s maker of four detectors, including its own, on 1,000 texts (May 2026) | The vendor’s own numbers: its detector 98.0% accurate with no false flags on human text, GPTZero 98.7% with 2.2% | Free (MIT) |
| RAID | A detector benchmark of AI texts with rewording and other attacks (6 million+ in the 2024 paper, over 10 million in the current project) | Detectors were easily fooled by rewording, setting changes and unfamiliar models (2024) | Free (MIT) |
| More pattern lists: Aboudjem’s humanizer-skill, slopbuster, harshaneel/humanize, anti-ai-slop-writing, no_ai_slop_writing_rules | The same approach with 55 patterns and a score, 152 patterns, “nine levers”, or rules plus one creator’s voice file | harshaneel/humanize publishes its maker’s own benchmark of 25 texts and says GPTZero, Grammarly and Pangram still scored the output 99-100% AI; none found for the other four | Free |
| Other languages and fields: Humanizer-zh, im-not-ai, academic-humanizer | A Chinese skill adapted from blader’s, a separate Korean-language skill with its own Korean tells, and one for papers and grant proposals | Pebblous, a data-quality company with its own in-house Korean style check, ran im-not-ai’s checking scripts on three texts: a 2020 human essay drew the highest grade (6/6), then low after five commas were deleted; the maker says it has since fixed this and reports 0 of 532 human texts at high risk on its own re-measure; none found for the other two | Free |
| Anthropic’s official skills | 19 skills for documents, slides, spreadsheets, design and more | None of them is about voice or AI tells | Free |
Studies
The studies, tests and sources behind our findings, grouped by topic, with the one finding from each that a writer can use. Preprints haven’t been through peer review, and results from 2023-2025 models may not hold for today’s.
The studies behind this page
Why AI writing sounds like AI15
| Study | Year | What it found |
|---|---|---|
| Reinhart et al. (PNAS) | 2025 | Chat-trained models write different grammar from people; Llama’s base versions write closer to human rates |
| Zhang et al., “Verbalized Sampling” | 2025 | People rating answers favor typical text; asking for several answers with probabilities brings back variety |
| Kirk et al. | 2023 | Training on people’s ratings makes answers less varied |
| West & Potts | 2025 | Models before chat training are less predictable and more original than their trained versions |
| Lu et al. | 2024 | Professional writers’ text is far harder to rebuild from existing web text than AI text, and training lowers models’ scores on that originality measure |
| Li et al. (EMNLP) | 2026 | Each post-training stage made one model family’s fiction more uniform in theme and style and moved intense emotion toward neutral |
| Shaib, Elazar, Li & Wallace (EMNLP) | 2024 | Most of the sentence patterns models repeat also appear in the web text they first learned from |
| Juzek & Ward (COLING) | 2025 | No proven cause for overused words like “delve”; the evidence fits training on people’s ratings |
| Sun et al. (ICML) | 2025 | Five chatbots told apart 97% of the time, and 91% after another model reworded the text |
| Chakrabarty, Laban & Wu, LAMP (CHI) | 2025 | Professional writers’ fixes to AI paragraphs; AI edits swapped clichés and handled specific detail worst |
| Jiang et al., “Artificial Hivemind” (NeurIPS) | 2025 | Different models give similar answers to open-ended questions |
| Hou et al. (ICML workshop) | 2026 | Post-trained models’ story openings collapse to fewer types, even with training meant to keep variety |
| Rallapalli et al. | 2026 | Grammar differences held under human-like prompts; the model shaped style more than its settings |
| Blevins et al. (EACL) | 2026 | Trained and larger models copy the style of a conversation less |
| Xu et al. | 2026 | GPTZero and Pangram often judged text from base models as human, and not text from their instruction-tuned versions |
Signs of AI writing7
| Study | Year | What it found |
|---|---|---|
| Geng & Trotta | 2025 | “Delve” fell in research abstracts after it was publicized |
| Kobak et al. (Science Advances) | 2025 | “Delves”, “showcasing” and “underscores” jumped across 15 million biomedical abstracts |
| Paech et al., Antislop (ICLR) | 2026 | Some patterns appear over 1,000 times more often in AI fiction; simple bans fail at around 2,000 patterns |
| Muñoz-Ortiz et al. | 2024 | Human news varies sentence length more and shows more fear and less joy than 2023-era models |
| Post-editing study | 2026 | Some of people’s edits removed AI-linked words and punctuation, and the edited drafts stayed close to AI style |
| Pew Research Center | 2026 | “It’s not just X, it’s Y” nearly tripled on the web but stayed rare; about 10% of a sample of 10,000 English pages showed strong AI signs, and over a third of pages published after ChatGPT |
| Wikipedia’s editors, “Signs of AI writing” | Ongoing | AI word lists by model era, signs of human writing and signs that don’t work (community guidance) |
Stance and speech5
| Study | Year | What it found |
|---|---|---|
| Jiang & Hyland | 2025 | ChatGPT essays used fewer hedges, firm words and words of personal judgment than students’ essays |
| Herbold et al. | 2023 | ChatGPT essays had fewer “I think”-type phrases, yet teachers rated them higher |
| Pham | 2026 | ChatGPT linguistics research articles showed less personal judgment and no “I” or “we” |
| Abdulhai et al. | 2026 | About 69% more essays stayed neutral when an AI wrote them; AI editing changed meaning |
| Berger, Rocklage & Packard | 2022 | People sharing opinions about products and experiences used more emotional language speaking than writing, unless they planned first |
Readers and judges11
| Study | Year | What it found |
|---|---|---|
| Russell, Karpinska & Iyyer (ACL) | 2025 | Five frequent AI users voting together caught 99.3% of AI articles; people who rarely use AI were near chance |
| Porter & Machery | 2024 | Readers told AI poems from human ones correctly 46.6% of the time, below chance, and rated AI poems higher when unlabeled; saying a poem was AI lowered its rating compared with saying it was human |
| Jakesch, Hancock & Naaman (PNAS) | 2023 | People’s cues for spotting AI (first person, contractions) can be faked, in short profile self-descriptions |
| Karpinska, Akoury & Iyyer (EMNLP) | 2021 | English teachers spotted model-written stories where crowd workers couldn’t |
| Chakrabarty, Ginsburg & Dhillon | 2026 | Expert readers preferred human writers to prompted AI; after fine-tuning they preferred the AI, and detection fell from 97% to 3% (preprint) |
| Shaib et al., “Measuring AI Slop” | 2026 | Staying on point, how much the text said and tone predicted copy-editors’ slop labels; AI judges agreed at about chance |
| Jones & Bergen | 2025 | A persona prompt doubled how often people took GPT-4.5 for a human in five-minute chats |
| Fein et al., LitBench | 2025 | The best AI judge matched people on 73% of human-written story pairs and was at chance on AI-written ones |
| Awuni et al. | 2026 | In a panel of four open-weight AI judge families, 55.4% of verdicts reversed when the order was swapped (preprint) |
| Card et al. (EMNLP) | 2020 | In 97 human-rating experiments from 2019 papers, 24% reported a significance test, and typical designs were too small to detect modest differences |
| Mayor, Bietti & Bangerter | 2025 | AI-imitated phone conversations were judged human 29% of the time, against 64% for real ones |
Detectors7
| Study | Year | What it found |
|---|---|---|
| Epoch AI | 2026 | Top detectors missed almost no plain AI text and 10-18% of writer imitations |
| Jabarian & Imas (University of Chicago) | 2025 | Pangram had near-zero errors and held up after a humanizer; GPTZero missed about half after it |
| Karr et al. (Notre Dame) | 2026 | Two detectors flagged 9-15% of unmodified human abstracts and 38-80% of lightly AI-edited ones (accepted to an ACM summit) |
| Park, Jeong & Kim (EMNLP) | 2026 | 13 detectors’ false-flag rates on non-native English writers’ academic manuscripts ranged from 0% to 100% |
| Saha & Feizi (ACL Findings) | 2025 | Detectors often flagged lightly polished human text and couldn’t tell amounts of AI help apart |
| RAID (ACL) | 2024 | Detectors lost accuracy under rewording, setting changes and unfamiliar models |
| Van Vlasselaer et al. | 2026 | Of four detectors on 160 documents built partly from real theses, Pangram placed the AI content in the expected range in 37 of 40 humanized papers, and GPTZero got 1 right |
Methods11
| Study | Year | What it found |
|---|---|---|
| blader/humanizer issue #229 | 2026 | In one user’s test, a humanizer rewrite won 16 of 16 with AI judges and barely moved detection; judges cited uniform rhythm more afterward |
| Wang et al., “Catch Me If You Can? Not Yet” | 2025 | Five samples beat none; 2 to 10 barely differed; same-topic samples lowered how often the imitation was matched to its author |
| Liang et al. | 2025 | Up to about 24% of the text in corporate press releases was estimated to be AI-modified by September 2024 |
| Gao & Das (AAAI) | 2024 | Good-and-bad example pairs beat plain examples on the tasks tested |
| Castricato et al. | 2024 | Telling a model to avoid a topic made one of four models marginally more likely to mention it, left one unchanged and made two less likely; not tested on writing style |
| Schmalbach | 2026 | In eight articles on one model, a word ban cut the listed words’ uses by about a third, and one rewrite dropped a needed hedge |
| Anthropic Interviewer | 2025 | In a survey after short AI interviews, 97.6% of participants rated their satisfaction 5 or higher out of 7; the page doesn’t say how many answered |
| Park et al. | 2024 (revised 2026) | AI stand-ins built from two-hour AI interviews predicted people’s survey answers better than stand-ins built from demographic details alone (83% against 74%, measured against how consistently people matched their own answers) |
| Nielsen Norman Group | 2024 | AI stand-ins for users gave shallow or overly favorable feedback, often seeing every idea as a good one |
| Chakrabarty, Laban & Wu, “AI-Slop to AI-Polish” | 2025 | A scoring model trained on expert edits picked edits professional writers preferred |
| Gao et al., CIPHER (NeurIPS) | 2024 | Learning preferences from a user’s edits cut later editing effort (simulated users) |
Models and fine-tuning7
| Study | Year | What it found |
|---|---|---|
| ToneBench (Towards AI) | 2026 | Claude Opus 5.5 at max effort scored highest on voice-matched scripts, by about the normal draft-to-draft variation, though its rating range doesn’t overlap the next model’s; the board calls its results directional |
| Every, Opus 5.5 review | 2026 | The most readable prose the reviewer had seen from an Anthropic or OpenAI model, but it often buries the point |
| Peeperkorn et al. | 2024 | Temperature is weakly linked to originality and moderately to incoherence |
| Schaeffer et al. | 2025 | Min-p, a popular setting, showed no reliable benefit on reanalysis |
| Marquardt & Brule | 2025 | Fine-tuning on 100 examples pushed about 90% of a small model’s answers over a reading-ease threshold in most settings, with readability standing in for tone |
| Gekhman et al. (EMNLP) | 2024 | Fine-tuning on facts a model doesn’t know increases made-up answers |
| Thinking Machines, “LoRA Without Regret” | 2025 | LoRA, a cheaper add-on method, matched full fine-tuning on small and medium sets of examples |
Every Method, Rated
Verdicts: Works (the evidence supports it), Partly (it helps with limits), Doesn’t (the evidence says no), Untested (we found no test of it). “Our” results are my own experience, not a published study.
Every method, rated
Helps, with limits: start here8
| Method | Verdict | What the evidence says | Source |
|---|---|---|---|
| Record the person and write from the transcript | Partly | People use more emotional language speaking than writing; it’s what has worked best in my experience; we found no study of articles | Speaking-versus-writing study |
| A handful of writing examples | Partly | Five examples beat none on the four models tested both ways, but AI blog posts still passed as the author’s only 17-21% of the time | Writing-samples study |
| Past call transcripts as source material | Partly | In the tests I’ve graded, they helped, most of all on articles where we had nothing else from the client | My own A/B tests |
| Banned-word list | Partly | In one developer’s test, a ban cut the listed words’ uses by about a third, and one rewrite lost a needed hedge; AI word lists change with each model generation | Schmalbach’s test; Wikipedia’s editors |
| Blocking words while the model writes | Partly | Simple word-fragment bans stop working at around 2,000 patterns; the Antislop sampler handles more but needs a model you run yourself; one developer’s ban worked on an open model reached through a hosted service; Claude’s API has no setting for it | Antislop paper; Schmalbach’s test |
| Start the draft with the person’s own first words | Partly | Raised the share of blog drafts one detector called human from 9% to 22%; we found no reader test | Writing-samples study |
| Humanizer rewrite | Partly | In one user’s test, won 16 of 16 with AI judges while detection barely moved, and judges noticed a more uniform rhythm afterward | blader/humanizer test |
| AI edits its own draft | Partly | Sometimes swaps one cliché for another and handles specific detail worst; models told to fix grammar only changed essays’ meaning | Professional-edits study; Neutral-essays study |
Doesn’t help10
| Method | Verdict | What the evidence says | Source |
|---|---|---|---|
| Tell it to “sound human”, “write casually” or “try harder” | Doesn't | Instruction-tuned models (GPT-4o, GPT-4o mini, Llama 3) told to match a person’s style kept a noun-heavy style; in one user’s test, “try harder” fooled 0 of 15 AI judges (AI models asked to rate text) | Model-continuation study; blader/humanizer test |
| A rule set telling the model to write more human | Doesn't | Fooled 0 of 17 AI judges in one user’s test; in our experience, piles of rigid rules made client drafts robotic | blader/humanizer test |
| More examples (up to 10) | Doesn't | Going from 2 to 10 changed the paper’s four measures very little; we found no test above 10 | Writing-samples study |
| Same-topic examples | Doesn't | Lowered how often the imitation was matched to its author (roughly halved for blogs) | Writing-samples study |
| Many excerpts plus a style description | Doesn't | Expert readers preferred human writers’ imitations, and a detector flagged 97% (fiction, older models, one preprint) | Author-imitation study |
| Rewrite with a second AI | Doesn't | The first model is still identifiable after another model rewords its text | Chatbot-fingerprint study |
| AI judge picks the best draft | Doesn't | Three AI models matched professional copy-editors at about chance on yes-or-no “slop” labels; in another study, a panel of open AI judges reversed 55% of verdicts when the order changed | Copy-editor study; Judge-panel study |
| AI role-plays the expert | Doesn't | Long excerpts of AI-imitated casual phone calls were judged human 29% of the time, against 64% for real ones; AI stand-ins in user studies gave favorable, vague feedback | Phone-call study; Nielsen Norman Group |
| Higher effort setting | Doesn't | On ToneBench, a writing benchmark that calls its results directional, Claude Opus 5.5 gained about 1.3 points from medium to extra high at about three times the cost, less than the score’s normal run-to-run variation; I haven’t seen a big difference in my own tests | ToneBench |
| Higher temperature | Doesn't | Added little originality and more incoherence in a 2024 story study; Claude Opus 5.5 and Sonnet 5.5 reject the setting | Temperature study; Anthropic |
Untested8
| Method | Verdict | What the evidence says | Source |
|---|---|---|---|
| Persona prompt | Untested | Doubled how often people took GPT-4.5 for a human in five-minute chats; we found no test on articles | Turing-test study |
| Write the prompt in the style you want back | Untested | Anthropic says the formatting of a prompt can steer the formatting of the reply; we found no test for article voice | Anthropic |
| Require a position in every piece | Untested | AI-written essays stayed neutral about 69% more often; in my experience it helps, though it changed along with other things | Neutral-essays study |
| Good-and-bad example pairs | Untested | Beat plain examples on 2024 tasks; we found no test for voice | Example-pairs study |
| Several drafts, a person chooses | Untested | The drafts have to be made to differ, since post-trained models’ story openings collapse to fewer types | Story-openings study |
| Learn from the person’s edits | Untested | Cut editing effort when GPT-4 played the users, on summaries and emails (2024) | Learning-from-edits study |
| AI interviews you first | Untested | In Anthropic’s own survey after its AI interviews of about 10-15 minutes, 97.6% of participants rated their satisfaction 5 or higher out of 7; the page doesn’t say how many answered, and no articles were rated | Anthropic |
| Written Q&A | Untested | Writing gives people time to plan, which flattened their language in the speech study | Speaking-versus-writing study |
Works, but most people can’t get it2
| Method | Verdict | What the evidence says | Source |
|---|---|---|---|
| A scoring model trained on expert edits picks the best | Works | Nine professional writers ranked its picks best; one preprint, with its research models and data released by the authors | Expert-edits study |
| Fine-tuning on the person’s writing | Worksmostly out of reach | In one preprint, expert readers preferred GPT-4o fine-tuned on an author’s books (30 authors, excerpts up to 450 words of literary writing) at a median $81 per author in computing costs; closed to new OpenAI customers, and the only Claude model Amazon Bedrock lists for tuning, Claude 3 Haiku, has reached end of life | Author-imitation study; OpenAI; Amazon Bedrock’s tuning list; Amazon Bedrock |
Signs of AI Writing
- Current: recent studies or editors still see it.
- Fading: it belonged mostly to older models.
- Myth: the evidence doesn’t support it.
- Unproven: we found no independent count of it.
- “Our drafts”: TJ Digital’s AI first drafts, compared with 27 company blog posts that four businesses published before 2023. Phrase counts come from 76 drafts and rates like contractions from a larger set of 121 AI articles, and each row says which. One writing process produced all of them, and they show how that one process writes.
Signs of AI writing, and what the evidence says about each
Stance6
| Sign | Example | What the evidence says | Status | Source |
|---|---|---|---|---|
| Few personal hedges | “I think”, “in my experience” | ChatGPT essays had fewer hedges overall (2025) and fewer phrases like “I think” (2023) than students’ essays, and ChatGPT linguistics articles had no “I” or “we”; older models | Current | Essay study, 2023 essay study, linguistics-articles study |
| Few firm claims | “always”, “definitely”, “the only” | Fewer in ChatGPT essays; less than half the human rate in our 17 client drafts, most without the client’s call material | Current | Essay study |
| No side taken | “There are pros and cons to both.” | When gpt-4o-mini wrote the essay, about 69% more essays stayed neutral than when people wrote alone (one small study) | Current | Neutral-essays study |
| Mild, upbeat feeling | positive tone, no strong emotion | 2023-era models wrote news with more joy and less fear than people did; training made one model’s fiction less intense | Current | AI-news study, Narrative-flattening study |
| Dated disclaimers | “It’s important to note that…” | Wikipedia’s editors date it to 2023-24; none of our 76 drafts used it | Fading | Wikipedia’s editors |
| Heavy softening | “may”, “typically”, “depends on” | Our 17 client drafts, most without the client’s call material, matched the human rate, while GPT-4o used more downtoners (words like “barely” and “nearly”) than people; the evidence points both ways, so softening alone tells you little | Myth | Model-continuation study |
Content8
| Sign | Example | What the evidence says | Status | Source |
|---|---|---|---|---|
| Generic detail | the first example anyone would think of | When AI models edited AI writing, adding specific detail was the fix they handled worst; models imitating writers often fell back to an average tone | Current | Professional-edits study, Imitation study |
| Clichés | “a sense of”, “the weight of”, “unspoken” | 17% of professional writers’ fixes; GPT-4o, Claude 3.5 Sonnet and Llama all overused the same ones in fiction, and AI edits swapped clichés for new ones | Current | Professional-edits study |
| Off-point or padded content | sections that drift from the reader’s question | Whether the text stayed on point and how much it actually said best predicted professional copy-editors’ “slop” labels (slop meaning generic, low-value AI text) | Current | Copy-editor study |
| Nothing new | points every page on the topic makes | Originality came up in 24% of all the explanations five frequent AI users gave for their calls | Current | Five-reader study |
| Invented stories, quotes and experts | a made-up customer story; a Seneca quote | GPT-3.5 and GPT-4 added an anecdote and a Seneca quote when asked to refine essays; many, not all, of 500+ stories built on one company’s press releases featuring fake and misleading comment have been deleted by the publishers | Current | Essay-refining study, Press Gazette |
| Inflated importance | “plays a pivotal role”, “a testament to” | Wikipedia’s editors list puffery and promotional tone; we found no count | Current | Wikipedia’s editors |
| Sameness across pieces | many articles with the same outline and opening | Different models give similar answers to open questions, and post-trained models’ story openings collapse to fewer types, even with training meant to keep variety | Current | Similar-answers study, Opening-types paper |
| Stock names and titles | “Emily”, “Sarah”, “Dr.” | In 63-70% of GPT-4o and Claude 3.5 Sonnet articles in 2024; we found no test since | Fading | Five-reader study |
Sentence and paragraph structure7
| Sign | Example | What the evidence says | Status | Source |
|---|---|---|---|---|
| Over-explaining and burying the point | explaining what the reader already knows | 18% of professional writers’ fixes to AI paragraphs; a staff writer at Every, a publication that covers AI, found Claude Opus 5.5 often buries the point | Current | Every, Professional-edits study |
| Even, polished rhythm | every paragraph the same length and cadence | AI models asked to judge text mentioned uniform rhythm more after a humanizer rewrite (60 mentions before, 101 after) | Current | blader/humanizer test |
| Formulaic endings | “In conclusion”, a reflective closing line | People who use AI often flagged formulaic conclusions; quotable closing lines rose after a humanizer rewrite | Current | Five-reader study, blader/humanizer test |
| Same shape in every section | answer, reason, example, takeaway, every time | Wikipedia’s editors and Every’s writing plugin flag it; we found no count | Current | Every’s Compound Writing |
| Repeated sentence templates | the same pattern of word types again and again | Models repeat grammar patterns more than people (2024), but in a study of professional copy-editors, repetition wasn’t a clear predictor of their “slop” labels | Current | Templates study, Copy-editor study |
| Sentences all about the same length | most sentences 10 to 30 words | 2023-era models varied sentence length less than human news writers; our 121 articles vary as much as the human posts | Fading | AI-news study |
| Perfect grammar, bland prose, transition words | flawless grammar; “Furthermore” | Wikipedia’s editors list these as signs that don’t work | Myth | Wikipedia’s editors |
Phrases and sentence patterns12
| Sign | Example | What the evidence says | Status | Source |
|---|---|---|---|---|
| “-ing” clauses | “Bryan, leaning on his agility, dances around the ring” | GPT-4o used them at 5.3 times the human rate, wherever they fell in a sentence (2024); in our 121 articles, a narrower count of “-ing” phrases after a comma was lower than in the human posts | Current | Model-continuation study |
| Noun-heavy grammar | “the implementation of” for “implementing” | GPT-4o turned verbs into nouns at about twice the human rate | Current | Model-continuation study |
| “It’s not X, it’s Y” | a sentence that denies one thing to assert another | Nearly tripled on the web from 2023 to 2026 but still rare (about 2 uses per 10,000 words); in 3% of our 76 drafts and 4% of human posts | Current | Pew Research Center |
| Reversed contrast | “Y rather than X” | Wikipedia’s editors call it common in recent output, with no source; “instead of” was in 38% of our 76 drafts and 11% of human posts | Current | Wikipedia’s editors |
| Fancy verbs for “is” and “has” | “serves as”, “boasts”, “features” | Wikipedia’s editors list it, and list plain verbs (“wrote”, “used”, “tried”) as signs of human writing | Current | Wikipedia’s editors |
| Few wordy connectors | rarely uses “in order to” or “the fact that” | Wikipedia’s editors saw these more in human-written Wikipedia articles than in AI text; “in order to” was in 1% of our 76 drafts and 33% of human posts | Current | Wikipedia’s editors |
| Announcing the point | “Here’s what that means”, “The real question is” | Listed in catalogs of AI habits, with no study we found; “that’s why / where / what / how” was in 58% of our 76 drafts and 7% of human posts | Current | tropes.fyi |
| Vague attributions | “Experts argue” with no expert named | Wikipedia’s editors list attributions like this as a sign | Current | Wikipedia’s editors |
| Chatbot leftovers | “I hope this helps!” | Wikipedia’s editors list them; older leftovers like notes about the model’s knowledge cutoff are now rare | Current | Wikipedia’s editors |
| Lists of three | “fast, simple and affordable” | The only count we found comes from Pangram, a detector company: 19 per 10,000 words in AI text against 5 in human text, by its own analysis | Unproven | Pangram |
| Missing contractions | “do not”, “it is” | People added contractions when editing AI drafts, but our 121 articles, written with instructions to match each brand’s voice, used twice the human rate | Myth | Post-editing study |
| Passive voice | “was designed to” (with no one doing the designing) | GPT-4o used this kind of passive at about half the human rate | Myth | Model-continuation study |
Words6
| Sign | Example | What the evidence says | Status | Source |
|---|---|---|---|---|
| The newest favorites | “matters”, “practical”, “clearer”, “quietly” | On Wikipedia’s mid-2026 list, which cites a single blog review as its source; “matters” was in 53% of our 76 drafts and none of the human posts | Current | Wikipedia’s editors |
| Each chatbot’s own words | ChatGPT: “certainly”, “overall”; Claude: “here”, “according to”, “based on” | Words that set one chatbot apart from the others, in 2024-25 versions | Current | Chatbot-fingerprint study |
| Claude’s pet phrases | “load-bearing”, “honest take” | Users report them; one user count found “load-bearing” rose from about 0.5 to 12 uses per 10,000 words between Claude Opus 4.6 and Opus 4.7 | Current | jola.dev, user count |
| “Delve” and its friends | “delve”, “showcasing”, “underscores” | Jumped in research abstracts after ChatGPT; “delve” fell once it was publicized in 2024; none of our 76 drafts used it | Fading | ChatGPT-words study, biomedical-abstracts study |
| Older models’ favorite words | “tapestry”, “intricate”, “palpable” (GPT-4o); “emphasizing”, “highlighting” (GPT-5) | GPT-4o used the literary ones 95 to 162 times the human rate in a 2024 test with fiction in the mix; Wikipedia’s editors date the GPT-5 words to mid-2025 through mid-2026 | Fading | Model-continuation study |
| Buzzwords | “leverage”, “robust”, “seamless”, “in today’s” | “Leverage” and “robust” are on Wikipedia’s watch lists; all four were in 0-13% of our 76 drafts, about as often as pre-2023 company blogs used them | Fading | Wikipedia’s editors |
Punctuation and formatting7
| Sign | Example | What the evidence says | Status | Source |
|---|---|---|---|---|
| Only commas and periods | many short sentences, little other punctuation | Wikipedia’s editors call it a new sign now that em dashes are trained out; one line cites a single blog review | Current | Wikipedia’s editors |
| Formatting where it doesn’t belong | bold and headings in an email or forum post | Pangram’s count: formatting symbols (bold, headers, code marks) in plain-text settings, 90 times per 10,000 words in AI text against 8 in human text, and bullet lists 28 against 3; formatting alone identified which chatbot wrote a text 73% of the time | Current | Chatbot-fingerprint study, Pangram |
| A bold label on every list item | “Speed: …”, “Cost: …” | Wikipedia’s editors list it; we found no count | Current | Wikipedia’s editors |
| Heavy bold | bold phrases in every paragraph | Wikipedia’s style notes say some newer models or apps are instructed to avoid overusing bold | Current | Wikipedia’s editors |
| Heading habits | Title Case Headings | Wikipedia’s editors say chatbots strongly tend to use title case in headings | Current | Wikipedia’s editors |
| Emoji, curly quotes, stray code | “utm_source=” left in a link | Wikipedia’s editors list them; curly quotes alone don’t prove AI, and “utm_source=” in a link is a ChatGPT leftover; Pangram counts emoji at 0.2 against 0.1 per 10,000 words | Current | Wikipedia’s editors, Wikipedia’s style notes, Pangram |
| Em dash overuse | asides set off by long dashes | Wikipedia’s editors list it as a sign of older models (2022 to September 2026); an Arena analysis reported in the tech press found Opus 5.5 used about 95% fewer than Opus 5 (0.8 against 15.2 per 1,000 words) | Fading | Wikipedia’s editors, BleepingComputer |
Common Claims About AI Writing
Much of what gets repeated about AI writing and AI detection is out of date or only partly true. These are the claims about how AI writes and how detection works, checked against the evidence as of Nov 1, 2026.
Common claims about AI writing: true or false?
True1
| Claim | Verdict | Why | Source |
|---|---|---|---|
| AI avoids taking a side | True | AI-written essays stayed neutral about 69% more often than people’s (one small study) | Neutral-essays study |
Partly8
| Claim | Verdict | Why | Source |
|---|---|---|---|
| AI hedges too much | Partly | It softens claims about as often as people in some counts, but it rarely says “I think” or commits to a view | Essay study |
| All AI models sound the same | Partly | Much of the style is shared, yet a program told five chatbots apart 97% of the time | Chatbot-fingerprint study |
| AI’s style comes from its training data | Partly | Much of it traces to the training after a model reads the web, since models before that training write closer to people; most of the sentence patterns models repeat also appear in the web text they learned from | Model-continuation study, Sentence-template study |
| AI doesn’t use contractions | Partly | People added contractions when editing AI drafts, but our 121 articles, written with instructions to match each brand’s voice, used twice the human rate | Post-editing study |
| Humanizer tools make AI text undetectable | Partly | Results conflict: after one commercial humanizer, both detectors in a Notre Dame paper flagged under 4%, while other studies found Pangram still caught most humanized text | Notre Dame study, Five-reader study |
| Detectors almost never flag human writing | Partly | Near zero in two tests, but two detectors flagged 9-15% of recent human-published abstracts in one study, and 13 detectors flagged anywhere from 0% to 100% of non-native English writers’ manuscripts | Edited-manuscripts study, Notre Dame study |
| Claude Opus 5.5 is the best AI writer | Partly | First on ToneBench by about the normal draft-to-draft variation in its scores, though its rating range doesn’t overlap the next model’s and the board calls its results directional; tied for first on Arena; seven models beat it on EQ-Bench fiction | ToneBench, EQ-Bench |
| Base models sound more human than chat models | Partly | True for two detectors and for grammar counts; we found no reader test | Base-model study, Model-continuation study |
False16
| Claim | Verdict | Why | Source |
|---|---|---|---|
| AI never hedges | False | ChatGPT essays had fewer hedges than students’ essays, which still means some, and our drafts use “may” and “typically” at about the human rate | Essay study |
| Ban hedging words to sound more human | False | AI already commits less than people do, and Wikipedia’s editors saw “perhaps” and “tends to” more in human-written Wikipedia articles than in AI text | Wikipedia’s editors |
| Remove the AI words and the text passes as human | False | Five frequent AI users voting together still caught every humanized article | Five-reader study |
| People write the way they talk | False | The same people used more emotional language speaking than writing | Speaking-versus-writing study |
| Perfect grammar or bland prose means AI wrote it | False | Wikipedia’s editors list both as signs that don’t work | Wikipedia’s editors |
| The humanizer leaderboards are independent | False | Every one found is run by a vendor or has affiliate links, and they rank the same tools far apart | HumanizerBench |
| A skill with tens of thousands of GitHub stars has been shown to work | False | stop-slop (18.5k stars) and no-ai-slop (13k) publish no results; of the 90 or so we found, only blader/humanizer has a judge or reader test from outside | stop-slop |
| AI detectors are coin flips | False | On plain AI text from 2026 models, Pangram and GPTZero missed under 1% and flagged no human passages | Epoch AI |
| A lower detector score means the writing sounds more human | False | In one user’s test, AI judges preferred a rewrite in 16 of 16 comparisons while detection stayed at 97.8-100% | blader/humanizer test |
| You can ask ChatGPT whether a text is AI | False | GPT-4o with a guidebook wrongly flagged 11% of human articles and caught 3% of humanized ones (older models) | Five-reader study |
| Most people can tell AI writing by feel | False | People who rarely use AI were at about chance | Five-reader study |
| One experienced reader is a reliable judge | False | Individual frequent AI users caught 59-97%; five voting together caught 99.3% | Five-reader study |
| If readers like it, it passes as human | False | General readers favored AI imitations that expert readers rejected | Author-imitation study |
| A blind test of 8 pairs can prove one version is better | False | With 2 ties, 8 pairs leave 6 with a winner, and only 6 of 6 would rule out luck; 5 wins and 1 loss fits a true win rate anywhere from 44% to 97% (our arithmetic; typical rating studies are too small to detect modest differences) | Statistical-power study |
| The most expensive model writes best | False | Fable 5.1 costs 2.5 times as much as Opus 5.5 per word processed (1.4 times as much per ToneBench script at max effort) and scores lower there (84.53 against 85.93) | ToneBench |
| ChatGPT’s personality setting changes the style of its drafts | False | OpenAI says personality doesn’t control drafts such as emails or posts | OpenAI |
Out of date4
| Claim | Verdict | Why | Source |
|---|---|---|---|
| “Delve” gives AI away | Out of date | It rose after ChatGPT and fell once it was publicized in 2024; none of our 76 drafts used it | ChatGPT-words study |
| An em dash means AI wrote it | Out of date | Wikipedia’s editors now list it as a sign of older models, the newest models avoid it, and professional writers use it too | Wikipedia’s editors |
| Detectors flag most non-native English writing | Out of date | A widely quoted 2023 study found detectors flagged 61% of non-native English essays; in 2026, rates ranged from 0% to 100% depending on the detector | Edited-manuscripts study, 2023 study |
| A custom GPT or a Gemini Gem is the place to keep your voice | Out of date | Custom GPTs are scheduled to retire Dec 11, 2026, and Gems on adults’ personal accounts are removed Dec 17, 2026 | OpenAI, Google |
Unproven1
| Claim | Verdict | Why | Source |
|---|---|---|---|
| A “40% AI” score means AI wrote 40% of the text | Unproven | Detectors can’t tell AI editing from AI drafting | Notre Dame study |
AI Detectors
Versions, types of text and settings differ between tests, so don’t average the numbers.
AI detectors: independent tests against vendor claims
| Detector | What independent tests found | What the vendor claims | Cost |
|---|---|---|---|
| Pangram | Epoch (2026, version 3.3.2): flagged 0 of 495 human passages; missed 0% of plain AI text and 10% of imitations. University of Chicago (2025): near-zero errors in six kinds of writing, and still accurate after a humanizer. Notre Dame (2026, version 3.2): flagged 14.9% of human abstracts published in 2023-25 and none from 2013-15 | Pangram 4.1 (Oct 19, 2026, the current version), per its model card: wrongly flags 0.0090% of English text on a cleaned test set and misses 0.3652% of AI text. Pangram 4 claimed 98.83% of humanizer output caught (launch post). Pangram says it’s unreliable on short replies, code, tables of contents and templated text | From $20/month for 300,000 words $0.05 per 100 words through its API |
| GPTZero | Epoch: flagged no human passages; missed 0.67% of plain AI text and 10.8% of imitations. Five-reader study (2025): caught 85%, then 47% after humanizing. University of Chicago: missed about half or more after a humanizer | Under 1% false flags 1.1% on non-native English essays 96.5% accurate on documents that are part human, part AI | $12.99-$24.99/month billed yearly |
| Originality.ai | Epoch (Turbo 3.0.2): flagged 19 of 495 human passages (3.84%); missed 0.34% of plain AI text and 17.9% of imitations | Its current models, Originality.ai 5.0 and 5.0 MAX (Oct 2026), claim 99%+ accuracy (launch post) The Lite and Turbo models, retired in Sep 2026, claimed about 0.5% and 1.5% false flags | Pro $14.95/month for 200,000 words |
| Turnitin | The one independent study we found, a 2026 Belgian study of 160 papers (40 real student theses, 40 fully AI, 40 mixed, 40 mixed and then humanized by a prompt): it cleared the human papers, scored all 40 fully AI papers at 0-20% AI, and caught 60% of mixed and 50% of humanized papers | Under 1% false flags, counted only on documents it scores at 20% AI or higher (Turnitin); detection of AI “bypasser” tools since Aug 2025 | Sold to schools |
| Binoculars (free research detector) | Five-reader study: caught 67%, or 40% at its strict setting; 0-7% of humanized text | Its 2024 paper claimed over 90% caught while wrongly flagging 0.01% of human text | Free |
| Other free detectors (RADAR, Fast-DetectGPT, open-source RoBERTa models) | Five-reader study: Fast-DetectGPT caught 80% with 7.2% false flags, RADAR 15%; University of Chicago: a RoBERTa detector wrongly flagged 30-78% of human text | No vendor | Free |
| Asking a chatbot | GPT-4o with an expert guidebook caught 78%, wrongly flagged 10.7% of human articles and caught 3.3% of humanized ones (older models) | No vendor claim | Any chatbot |
| Five frequent AI users voting together | Caught 99.3% with no false flags, and 100% of humanized articles; one reader alone caught 59-97% (older models) | No vendor | Your own panel |
Writing Benchmarks
These are the boards worth knowing, with what each actually measures.
Writing benchmarks and leaderboards
| Benchmark | What it measures | Leaders (date) | Limits |
|---|---|---|---|
| ToneBench | YouTube scripts written in one creator’s voice, scored by four AI judges against the creator’s own scripts | Claude Opus 5.5 85.93 Claude Sonnet 5.5 84.60 Claude Fable 5.1 84.53 GPT-6 Sol 83.52 (8th, the first non-Claude entry) All at max effort, Oct 26, 2026 | One creator and 10 tasks Opus 5.5 is one of the judges Scores vary 1.1 to 1.8 points between runs, about the size of the gaps at the top No human ratings shown The board calls its results directional |
| Arena creative writing | People vote between two anonymous replies to their own short creative prompts | Opus 5.5 1520 Gemini 4 Argon 1517 Gemini 4 Carbon 1515, on far fewer votes A statistical tie, Oct 27, 2026 | No target voice; short prompts; GPT-6 models outside the top 25 |
| EQ-Bench Creative Writing v3 | Short fiction scored by AI judges, plus a score for stock AI phrases | GPT-6 Astra, Kimi K3.1 Pro and Fable 5.1 lead Opus 5 has the fewest stock phrases (6.59) Opus 5.5 is 8th (10.43) Oct 28, 2026 | Fiction; Claude and GLM judges; the phrase score counts overused words and phrases |
| EQ-Bench Longform | Stories of about 8,000 words, scored by one Claude judge | Opus 5 86.3 Fable 5.1 85.3 Opus 5.5 84.6 Grok 4.7 has the fewest stock phrases Oct 28, 2026 | Fiction; one judge; the closest board to article length |
| Lech Mazur’s writing benchmark | 600-800-word stories that must include ten required elements, compared by AI judges | Fable 5.1 3.745 Opus 5.5 3.693 Opus 5 3.685 Oct 23, 2026 | Measures story construction around required elements; nothing on it measures how human a story sounds |
| SlopBench | Repeated openers, paragraph rhythm and stock phrasing in email, social, essay and workplace tasks | Kimi K2.6 has the least Mistral Large the most Sep 2026 | Only the two ends of the ranking hold up; unclear whether the newest models were tested |
| Hemingway-bench | Blind comparisons by professional writers (Surge AI’s post) | Claude Fable 5 1110 Gemini 3.7 Flash 1105 Claude Fable 5.1 1095 Nov 1, 2026 (the board shows no date) | Run by Surge AI, which sells human ratings; Opus 5.5 not listed |
Model Settings
Model settings and what they change
Whether you can use each with Claude, ChatGPT and Gemini models, as of late October 2026
| Setting | What the evidence says | Claude | ChatGPT | Gemini |
|---|---|---|---|---|
| Higher effort | Opus 5.5 on ToneBench: 84.49 at medium, 85.83 at extra high (about 3 times the cost), 85.93 at max (about 8 times); I haven’t seen a big difference in my own tests | Yes, low to max (Opus 5.5 defaults to medium) | A reasoning setting on GPT-6 | Thinking set to low, medium or high on Gemini 4 Argon |
| Temperature | Weakly linked to originality, moderately to incoherence; min-p, a popular setting that trims unlikely words, showed no reliable benefit when its original data was reanalyzed (reanalysis) | No, it returns an error | Not listed for GPT-6 on OpenRouter | Yes, on Gemini 4 Argon |
| Base model | Grammar closer to human; two detectors often read its text as human (base-model study); we found no reader test; hard to direct; you’d run an open one yourself, such as Qwen3 Base or Llama 3.1 405B Base | None found | None found | None found |
| Blocking phrases while it writes | The Antislop sampler handles 8,000+ words, phrases and patterns; we found no reader test; possible on some open models | No | Not listed for GPT-6 on OpenRouter | Not listed for Gemini 4 Argon on OpenRouter |
| A model built for creative writing | We found no blind test showing these sound more human; most target fiction or creative work | Anthropic positions Fable 5.1 for reasoning work | Shown in March 2025, with no release date given (TechCrunch) | None found |