Writing Standards Knowledge Pack Files

Banned AI Writing Patterns

29 patterns sourced from humanizer, based on Wikipedia's "Signs of AI writing" (WikiProject AI Cleanup). Each pattern includes a detection signal, before/after example, and fix guidance.

Content Patterns (1-6)

Pattern 1: Significance Inflation

Signal: Uses words like "pivotal", "crucial", "vital", "key role", "indelible mark", "setting the stage for" Before: "This pivotal moment in the company's journey underscores the vital role of innovation." After: "The company shipped its first product that quarter." Fix: Replace inflated significance claims with plain facts. If something is important, the facts will show it.

Pattern 2: Notability Name-Dropping

Signal: Mentions famous people, institutions, or events to add unearned weight Before: "Much like Steve Jobs revolutionized personal computing, this approach transforms content creation." After: "This approach reduces content production time by half." Fix: Cut comparisons to famous figures. Make claims stand on their own evidence.

Pattern 3: Superficial -ing Analyses

Signal: Sentences structured as "[Topic], [verb]-ing [vague implication]" Before: "The platform integrates AI agents, enabling teams to ship faster while reducing overhead." After: "The platform uses AI agents. Teams ship twice as fast at a third of the cost." Fix: Split compound sentences. Replace "-ing" clauses with specific, measurable claims.

Pattern 4: Promotional Language

Signal: Words like "boasts", "showcases", "exemplifies", "commitment to", "groundbreaking", "renowned", "breathtaking", "must-visit", "stunning" Before: "The platform boasts a groundbreaking approach to workflow automation." After: "The platform automates 12 workflow types without custom code." Fix: Replace promotional adjectives with specific facts. What does it actually do?

Pattern 5: Vague Attributions

Signal: "Experts say", "studies show", "research indicates", "many believe" without citing sources Before: "Research indicates that AI-augmented teams are significantly more productive." After: "GitHub's 2024 study found developers using Copilot completed tasks 55% faster." Fix: Name the source, cite the study, provide the number. If you can't cite it, cut it.

Pattern 6: Formulaic Conclusions

Signal: "Challenges and future prospects", "looking ahead", "as we move forward", "the future of X is bright" Before: "As we look ahead, the future of AI-powered development is bright, with challenges remaining but prospects encouraging." After: Cut entirely. Or: "Next quarter, the team plans to add Python and Go support." Fix: Delete formulaic conclusion paragraphs. End with a specific next step or CTA, not a vague forward-looking statement.

Language Patterns (7-13)

Pattern 7: AI Vocabulary Overuse

Signal: Frequent use of: delve, enhance, foster, garner, intricate, landscape (abstract), pivotal, showcase, tapestry (abstract), testament, underscore, vibrant, crucial, additionally, align with, interplay, valuable, enduring, highlight (as verb) Before: "Let's delve into how this intricate interplay of features enhances the vibrant landscape of modern development." After: "Here's how the features work together." Fix: Replace with plain words. "Delve" → examine, look at. "Enhance" → improve. "Landscape" → field, area. "Leverage" → use.

Pattern 8: Copula Avoidance

Signal: Using "serves as", "stands as", "represents", "marks", "features", "offers" instead of "is" Before: "The dashboard serves as the central hub for fleet monitoring." After: "The dashboard is the fleet monitoring center." Fix: Use "is" or "are" when something simply is something. Fancy verbs aren't more professional.

Pattern 9: Tailing Negations / Negative Parallelisms

Signal: "Not just X, but Y", "not merely X — Y", "X, yet not Y" Before: "This isn't just a tool — it's a complete engineering department." After: "This is a complete engineering department." Fix: State the positive claim directly. Drop the negation setup.

Pattern 10: Rule of Three Overuse

Signal: Lists of exactly three items, especially adjectives or phrases Before: "The platform is fast, reliable, and scalable." After: "The platform handles 10,000 concurrent builds without queueing." Fix: Use two items or one. Three is the strongest AI tell in list construction. If you must list, use specific items, not abstract adjectives. Landing page exception: Reduce to LOW. Feature grids commonly use 3-column layouts, making lists of three a design constraint. Only flag if the three items are vague adjectives rather than specific capabilities.

Pattern 11: Synonym Cycling

Signal: Using different words for the same concept within a paragraph to avoid repetition Before: "The agents complete tasks autonomously. These automated workers handle operations independently. The self-directed bots..." After: "The agents complete tasks autonomously. They handle deployments, tests, and code reviews without human intervention." Fix: Repeat the same term. Readers prefer clarity over forced variety.

Pattern 12: False Ranges

Signal: "From X to Y" constructions that cover everything, saying nothing Before: "From startups to enterprises, from development to deployment, the platform serves diverse needs." After: "The platform runs in production at three companies with 20-500 employees each." Fix: Replace ranges with specific examples or data.

Pattern 13: Passive Voice / Subjectless Fragments

Signal: "It was decided", "improvements were made", "the system can be configured" Before: "Significant improvements were made to the deployment pipeline." After: "The team cut deploy time from 45 minutes to 8." Fix: Name the actor. Who did it? Use active voice.

Style Patterns (14-19, 26-29)

Pattern 14: Em Dash Overuse

Signal: Multiple em dashes (—) per paragraph Before: "The platform — which launched last quarter — handles everything from deployment — including rollbacks — to monitoring." After: "The platform launched last quarter. It handles deployment (including rollbacks) and monitoring." Fix: Replace em dashes with periods, commas, or parentheses. One em dash per page maximum, if any.

Pattern 15: Boldface Overuse

Signal: Key phrases bolded for emphasis throughout body copy Before: "Our AI agents deliver real results with zero configuration." After: "AI agents deliver real results with zero configuration." Fix: Remove most bold. Use it only for headings, labels, or genuinely critical terms. Body copy should not need bold to communicate emphasis.

Pattern 16: Inline-Header Vertical Lists

Signal: Paragraphs formatted as bold-header + colon + description, stacked vertically Before: "Speed: Deploy in minutes. Scale: Handle any workload. Reliability: 99.9% uptime." After: "Deploy in minutes. It handles any workload at 99.9% uptime." Fix: Rewrite as flowing prose. Lists are fine for genuinely discrete items, but marketing benefits should flow as narrative.

Pattern 17: Title Case Headings

Signal: Every Word In The Heading Is Capitalized Before: "How Our Platform Transforms Your Engineering Workflow" After: "How our platform transforms your engineering workflow" Fix: Use sentence case. Title case is an AI tell in body content and subheadings.

Pattern 18: Emoji Usage

Signal: Emojis used as section markers, bullet decorations, or emphasis Fix: Remove emojis from professional content. Exception: social media posts where emojis are platform-native.

Pattern 19: Curly Quotation Marks

Signal: "Smart quotes" instead of "straight quotes" in technical or web content Fix: Use straight quotes for web content. Curly quotes can cause encoding issues and are an AI formatting tell.

Pattern 26: Hyphenated Word Pair Overuse

Signal: Frequent compound modifiers: "cutting-edge", "data-driven", "results-oriented", "mission-critical" Before: "Our cutting-edge, AI-powered, data-driven platform delivers mission-critical results." After: "The platform uses AI to analyze data and produce reliable results." Fix: Unpack hyphenated pairs into plain descriptions. One per sentence maximum.

Pattern 27: Authority Tropes

Signal: "The real question is", "at its core", "in reality", "what really matters", "fundamentally", "the heart of the matter" Before: "At its core, the real question is what fundamentally matters for your engineering team." After: "Your engineering team needs faster deploys and fewer bugs." Fix: Delete the authority framing. State the point directly.

Pattern 28: Signposting Announcements

Signal: "Let's dive in", "let's explore", "let's break this down", "here's what you need to know", "without further ado" Before: "Let's dive into how AI agents can transform your workflow." After: "AI agents pick tickets, write code, and open PRs. Here's how." Fix: Delete the signpost. Start with the content.

Pattern 29: Fragmented Headers

Signal: Headers that are sentence fragments used for dramatic effect Before: "The problem. → The solution. → The results." After: "Problem → Solution → Results" or just use descriptive headers Fix: Use descriptive headers or full sentences. Dramatic fragment headers are a strong AI tell. Landing page exception: Suppress. Landing pages use fragment headers as section labels paired with supporting body text (e.g., "Deploy." + "Set up agents on dedicated cloud servers."). This is standard web design, not an AI pattern.

Communication Patterns (20-22)

Pattern 20: Chatbot Artifacts

Signal: "I hope this helps!", "Of course!", "Certainly!", "You're absolutely right!", "Would you like me to...", "Let me know if..." Fix: Delete entirely. Published content should never contain conversational artifacts.

Pattern 21: Knowledge-Cutoff Disclaimers

Signal: "As of my last update", "I don't have access to real-time data", "based on information available to me" Fix: Delete entirely. If information might be outdated, verify it or note the date.

Pattern 22: Sycophantic Tone

Signal: Excessive agreement, complimenting the reader, "great question!", "that's a really interesting point" Fix: Delete. State information directly without complimenting the audience.

Filler & Hedging (23-25)

Pattern 23: Filler Phrases

Signal: "It's worth noting that", "it's important to remember that", "it should be mentioned that", "needless to say" Before: "It's worth noting that the platform handles concurrent deployments efficiently." After: "The platform handles concurrent deployments efficiently." Fix: Delete the filler phrase. The sentence is always stronger without it.

Pattern 24: Excessive Hedging

Signal: "Arguably", "perhaps", "it could be said that", "in some ways", "to some extent", "relatively" Before: "The platform is arguably one of the more effective solutions in the space." After: "The platform ships code faster than manual engineering teams." Fix: Commit to the claim or don't make it. Hedging signals uncertainty without adding nuance.

Pattern 25: Generic Positive Conclusions

Signal: Final paragraphs that summarize what was said and end on a vague positive note Before: "In conclusion, AI-powered development represents an exciting frontier with tremendous potential to transform how teams build software." After: "Start a pilot with three agents on your lowest-risk codebase. Results show in the first sprint." Fix: End with a specific action, not a summary. If the content was good, it doesn't need restating.

Structural Patterns (S1-S6)

These patterns come from stop-slop and address higher-level structural tells.

Pattern S1: Binary Contrasts

Signal: "Not because X. Because Y.", "X isn't the problem. Y is.", "The answer isn't X. It's Y." Before: "The problem isn't the technology. It's the process." After: "The process needs fixing. Here's how." Fix: State the positive claim directly. Drop the false contrast.

Pattern S2: Negative Listing

Signal: "Not a X... Not a Y... A Z." (Rhetorical striptease through negation) Before: "Not a dashboard. Not a report. A complete operating system for your engineering team." After: "A complete operating system for your engineering team." Fix: Skip the buildup. State what it is.

Pattern S3: Dramatic Fragmentation

Signal: "[Noun]. That's it. That's the [thing].", "[X]. And [Y]. And [Z]." Before: "Ship. Review. Deploy. That's it. That's the workflow." After: "The workflow has three steps: ship, review, deploy." Fix: Write complete sentences. Fragmentation for drama is one of the most recognizable AI patterns.

Pattern S4: False Agency

Signal: Inanimate objects performing human actions: "a complaint becomes a fix", "the data tells us", "the market rewards", "the decision emerges" Before: "The platform empowers teams to achieve their goals." After: "Teams using the platform ship 2x faster." Fix: Name human actors. "The market rewards" → "Customers buy". "The data tells us" → "We found".

Pattern S5: Narrator-from-Distance

Signal: "Nobody designed this.", "This happens because...", "People tend to...", "It turns out..." Before: "Nobody planned for this outcome. It emerged naturally from the architecture." After: "The architecture produced this outcome. We didn't plan for it." Fix: Name actors and use first person where appropriate. Avoid the omniscient narrator voice.

Pattern S6: Rhetorical Setups

Signal: "What if [reframe]?", "Here's what I mean:", "Think about it:", "And that's okay." Before: "What if I told you that AI agents could handle 90% of your engineering tasks? Think about it." After: "AI agents handle 90% of engineering tasks in our production fleet." Fix: Delete the setup. State the claim.