synthesis-content-quality（综合内容质量）

Content Quality

A systematic methodology for developing high-quality AI-assisted content and identifying content that falls short. This framework defines 36 criteria organized into confidence tiers for evaluating whether AI-assisted content meets professional publishing standards.

When to Use This Skill

• Reviewing AI-assisted drafts before publication
• Editing content that may contain unrevised AI output
• Building or calibrating AI content detection tools
• Training writers or editors on AI content quality standards
• Performing editorial review of submitted content

Core Philosophy

AI-assisted content creation is legitimate and valuable. The distinction that matters is between:

• Unedited AI output: Raw generation copied and published without human refinement
• AI-augmented work: Human expertise enhanced by AI capabilities, with proper oversight
• Systematic human-AI collaboration: Methodical integration where humans maintain judgment, add genuine expertise, and ensure quality

The goal is quality assessment, not origin detection. No single indicator proves AI generation definitively. Detection requires pattern recognition across multiple indicators.

The 36 Quality Criteria

Each criterion is tagged with its confidence tier: [HIGH], [MED], or [LOW].

Language and Tone Patterns

1. Undue Emphasis on Importance and Symbolism [MED] -- Inflating significance with phrases like "stands as a testament to" or "plays a vital role in." Fix: describe subjects accurately rather than inflating importance.
2. Promotional and Travel Brochure Language [MED] -- Marketing copy tone with "rich cultural heritage," "breathtaking," "nestled in the heart of." Fix: replace promotional adjectives with specific, factual descriptions.
3. Editorial Commentary and Meta-Analysis [MED] -- Injecting interpretation with "it's important to note," "notably," "one cannot overlook." Fix: state facts and trust readers to determine importance.
4. Superficial Analysis with Participial Phrases [MED] -- Sentences ending with "-ing" phrases adding shallow commentary: "highlighting its commitment to sustainability." Fix: provide real analysis with evidence, or let the statement stand alone.
5. Negative Parallelism [MED] -- Overusing "not X but Y" constructions: "not just a place to eat, but a cornerstone of community." Fix: use this structure sparingly and only for genuine contrast.
6. Overuse of Transition Words [LOW] -- Excessive "Moreover," "Furthermore," "Additionally" placed mechanically. Fix: let ideas connect through logical flow; vary transitions.
7. Section-Ending Summaries [MED] -- Explicit summary phrases: "In summary," "Overall," "In essence." Fix: remove these; if a section needs a summary to be understood, restructure it.
8. The Rule of Three [MED] -- Formulaic grouping in threes: "innovative, impactful, and transformative." Fix: vary list lengths; let content dictate structure.
9. Passive Voice and "Has Been Described As" [LOW] -- Overreliance on "is widely regarded as," "has been praised for." Fix: use direct statements with specific attribution.
10. Uniform Sentence and Paragraph Length [MED] -- Mechanically consistent structure without natural variation. Fix: vary deliberately -- short sentences for emphasis, longer ones for complexity.

Style and Structural Indicators

1. Excessive Em Dashes [LOW] -- Overusing em dashes where commas, parentheses, or colons fit better. Fix: match punctuation to function.
2. Bulleted Lists with Bolded Lead-ins [MED] -- Formulaic bullets with "Term: explanation" structure throughout. Fix: vary list formats; sometimes prose serves better than a list.
3. Excessive Bolding and Formatting [LOW] -- Mechanical bolding of every "important" term. Fix: bold sparingly. If everything is emphasized, nothing is.
4. Emoji Usage in Inappropriate Contexts [LOW] -- Emojis in formal content where they do not belong. Normal in casual content; a tell in formal contexts.
5. Markdown Formatting Mixed with Standard Text [HIGH] -- Raw Markdown syntax (asterisks, backticks, hash headers) appearing in published content.
6. Curly vs. Straight Quotes [LOW] -- Inconsistent quote styles or wrong type for context.
7. Title Case in Headers [LOW] -- Every major word capitalized instead of sentence case, especially in journalism contexts.

Technical and Formatting Tells

1. Placeholder Text and Incomplete Elements [HIGH] -- Bracketed placeholders like [Insert source here] or [Citation needed] left in published content.
2. Chatbot Communication Artifacts [HIGH] -- Salutations, valedictions, knowledge cutoff disclaimers, or offers to assist further appearing in published content.
3. Broken or Fabricated Links and Technical Codes [HIGH] -- URLs leading to 404 errors, invalid DOIs/ISBNs, or ChatGPT-specific artifacts like "turn0search0." Verify all links and identifiers.
4. Citation Abnormalities [MED] -- Citations repeated without proper reference tagging, real sources cited for unrelated content, or generic "According to experts..." without naming them.
5. Suspiciously Long Edit Summaries [MED] -- Unusually formal, comprehensive edit summaries in first-person paragraphs on platforms with edit tracking.

Citation and Sourcing Issues

1. Hallucinated Citations [HIGH] -- Fabricated sources, misattributed quotes, non-existent journal articles with plausible-sounding titles. This is among the most dangerous AI content problems.
2. Vague Attribution to Unnamed Authorities [MED] -- "Experts say," "Studies have shown," "Research indicates" without specific attribution. Professional standard requires verifiable sources.

Context-Specific Indicators

1. Industry-Specific Slop Patterns [MED] -- Domain-characteristic AI patterns: "innovative, cutting-edge" in tech; "hidden gem" in travel; "synergy, leverage" in business; uniformly positive product reviews.
2. Lack of Personal Detail or Specificity [MED] -- Generic descriptions without specific examples, personal anecdotes, or experiential details. Humans who have experienced something provide sensory details and concrete examples.
3. Superficial Depth Without Expertise [MED] -- Covering topics broadly without demonstrating actual understanding. Restating common knowledge, using technical terms superficially, avoiding nuance, lacking case studies. A specific sub-pattern: the "balanced hedging" conclusion that refuses to take a definitive stance, ending with "Ultimately, finding a balance between X and Y is crucial" or "Both approaches have merits." If every section ends with equivocation rather than a position, the writer is summarizing a field rather than working in it.

Hyperbolic and Dramatic Patterns

1. Hyperbolic Subheadings and Section Titles [MED] -- Subheadings that inflate significance: "The word that changed everything," "A game-changing approach," "The revolutionary insight." Fix: subheadings should describe what the section contains, not advertise it. Use specific, factual titles.
2. Dramatic Fragment Construction [MED] -- Short dramatic sentences or fragments used for artificial emphasis: "And it was a disaster." "Everything changed." "The results were stunning." Fix: let the content create impact through specificity and evidence, not through theatrical sentence structure. One or two per article for genuine rhetorical effect is fine; a pattern of them is AI-style pacing.
3. Borrowed Canonical Examples [MED] -- Using the same illustrative examples that appear in every article on a topic: "A jet engine is complicated; a market is complex" (Cynefin), "the bus route that nobody rides" (design thinking), "the restaurant with great food but no customers" (systems thinking). These signal that the writer is summarizing a field rather than working in it. Fix: use examples from your own experience or construct novel ones.

Confidentiality and Exposure Risks

1. Scenario Fingerprinting in "Anonymized" Examples [HIGH] -- Removing company names while keeping the scenario, specific numbers, stakeholder dynamics, vocabulary, and industry context. The scenario IS the identifier — names are the least important part. A story about "a content platform used by journalists" where you "changed fourteen components" from "generate" to "draft" is identifiable to anyone who knows the author's work. Fix: apply the four-test protocol: (1) Outsider test — could a stranger narrow this to a small set of companies? (2) Insider test — does this confirm something an insider suspected? (3) Adversary test — could a reporter use this as evidence? (4) Irony test — does publishing this undermine the thing the example describes protecting?
2. Operational Decisions Presented as Teaching Material [HIGH] -- Internal product strategy decisions, risk mitigation choices, and confidential operational changes described as case studies — even without names. If the decision was made to manage risk, describing it publicly re-creates the risk. Fix: use genuinely universal patterns, publicly known examples from other companies (with attribution), fictional scenarios clearly marked as illustrative, or the author's personal methodology (which is already public).

Behavioral and Tonal Patterns

1. Saturated AI Vocabulary [MED] -- Clustering of words that appear at disproportionately high frequency in unedited AI output: delve, tapestry, nuanced, robust, foster, beacon, catalyst, synergy, pivotal, overarching, multifaceted, landscape (used abstractly), leverage (as verb), streamline, spearhead, underscore, harness. Any single occurrence is unremarkable — these are legitimate words. The signal is clustering: three or more from this list in a single piece, or repeated use of the same word, suggests unrevised AI output. Fix: replace with specific, concrete alternatives. "Delve into" becomes "examine" or "investigate." "Robust framework" becomes a description of what makes the framework strong.
2. Exhausted Metaphors as Structural Filler [MED] -- Dead metaphors used as connective tissue to simulate analytical sophistication: "navigating the complex landscape of," "viewed through the lens of," "a symphony of moving parts," "at the intersection of X and Y," "the fabric of," "a tapestry of," "unpacking the layers of." These function as transitions that add zero meaning — they connect ideas without saying anything about the connection. Fix: state the actual relationship between ideas directly. "Navigating the complex landscape of AI regulation" becomes "AI regulation is fragmented across jurisdictions" — a claim with content instead of a metaphor without any.
3. Unprompted Moral Cadence [MED] -- Injecting ethical reminders, aspirational wrap-ups, or "looking toward a brighter future" codas at the end of factual or technical content where the topic does not warrant moral framing. An article about database indexing that ends with "As we build these systems, we must remain mindful of their impact on society" is an AI-typical domain mismatch — the moral register does not match the content's register. Fix: end technical content with technical conclusions. If the topic genuinely raises ethical questions, address them with specificity, not platitudes. The test: does this moral conclusion follow from the preceding analysis, or was it appended because AI defaults to inspirational endings?
4. The Concierge Tone [HIGH] -- Sycophantic agreement, sterile professional empathy, and service-register language appearing in content that is not customer service. Manifestations include: excessive validation ("That's a great question!"), hedged positivity that avoids any negative assessment, formulaic empathy ("I understand your concern"), and a pervasive agreeableness that treats every statement as a customer interaction. Distinct from chatbot artifacts (criterion 19), which are structural tells like valedictions and help offers. The concierge tone is a tonal quality that pervades the entire piece — the writer never disagrees, never says something is wrong, never takes a position that might displease. Fix: take positions. Disagree where warranted. State limitations directly. Professional writing has a perspective; service writing has a customer.

For detailed explanations, examples, and fix guidance for each criterion, see references/detailed-criteria.md.

Confidence-Based Evaluation Process

Step 1: Scan for High-Confidence Indicators

Check for: hallucinated citations, chatbot artifacts, placeholder text, raw Markdown formatting, broken/fabricated links, concierge tone, multiple indicators clustering together.

If any are present: very likely unedited AI output.

Step 2: Count Medium-Confidence Indicators

• 3-4 present: likely AI-generated
• 5+ present: very likely AI-generated

Step 3: Assess Overall Pattern

• Uniform structure + promotional tone + shallow analysis = strong AI signal
• Specific details + personal voice + varied structure = human-written

Step 4: Consider Context

• Is this from an established author with a portfolio?
• Does other work by this author show similar patterns?
• Is the publication known for quality control?

Ineffective Detection Methods

These do NOT reliably signal AI generation:

• Perfect grammar -- Skilled humans and professional editors produce polished prose
• "Bland" prose -- Corporate communications from humans can sound formulaic
• Common phrases -- "Rich cultural heritage" exists in human writing too
• Em dashes -- Professional human writers use them frequently
• Technical terminology -- Experts naturally use jargon

Systematic Revision Process for Creators

When using AI to assist content creation, revise through these five passes:

1. Eliminate formulaic patterns -- Vary sentence/paragraph length, reduce mechanical rule of three, remove promotional language and editorial commentary, replace generic descriptions with specifics.
2. Add genuine expertise and experience -- Include personal anecdotes and specific observations, provide depth beyond surface-level analysis, take clear positions with genuine reasoning.
3. Verify and enhance sourcing -- Check all citations are real and relevant, add specific attribution, include original research or first-hand sources. Use the synthesis-fact-checking skill for thorough verification.
4. Inject personality and voice -- Use natural transitions, vary rhetorical structures, include humor or perspective where appropriate, let imperfections remain if they sound natural.
5. Apply the Human Touch test -- Would a reader recognize this as distinctly yours? Does it include knowledge only you would have? Does it sound like how you actually write?

Quick-Reference Checklist

High-Risk Phrases

• "stands as a testament to," "plays a vital/significant role"
• "rich cultural heritage," "breathtaking," "nestled in the heart of"
• "it's important to note," "it is worth mentioning," "one cannot overlook"
• "not only... but also," "it's not just X, it's Y"
• "Moreover," "Furthermore," "Additionally," "Nevertheless"
• "In summary," "In conclusion," "Overall," "In essence"

High-Risk Vocabulary (flag when 3+ cluster in a single piece)

• delve, tapestry, nuanced, robust, foster, beacon, catalyst
• synergy, pivotal, overarching, multifaceted, landscape (abstract)
• leverage (verb), streamline, spearhead, underscore, harness

Exhausted Metaphor Phrases

• "navigating the complex landscape of..."
• "viewed through the lens of..."
• "a symphony of moving parts"
• "at the intersection of X and Y"
• "the fabric of..." / "a tapestry of..."
• "unpacking the layers of..."
• "Ultimately, finding a balance between X and Y is crucial"

High-Risk Subheading Patterns

• "The X that changed everything"
• "A game-changing approach to..."
• "The revolutionary/transformative..."
• "Why X will never be the same"
• "The surprising truth about..."
• "What nobody tells you about..."

Anonymization Checks (Before Publishing)

• Outsider test: could a stranger narrow this to a small set of companies?
• Insider test: does this confirm something an insider suspected?
• Adversary test: could a reporter use this as evidence?
• Irony test: does publishing this undermine what the example describes protecting?
• Are specific numbers (14 components, 6 engineers) identifying?
• Are stakeholder dynamics (journalists worried about AI) identifying?
• Are vocabulary choices (exact terminology changes) identifying?

Structural Checks

• Sentences vary in length naturally
• Paragraphs vary in size
• Does not group everything in threes
• Transitions feel natural, not mechanical
• No section-ending summaries
• Formatting is strategic, not excessive
• Citations verify and are relevant
• Voice and personality are present
• Includes specific examples and details
• Demonstrates genuine expertise
• No placeholder text or artifacts
• No chatbot communication remnants

The Human Touch Test

Before publishing AI-assisted content:

1. Would a reader recognize this as distinctly mine?
2. Does it include knowledge only I would have?
3. Does it sound like how I actually write?
4. Would anyone else write it exactly this way?
5. Have I added genuine value beyond what AI provided?

If you cannot answer yes to most of these, revise further.

Related

Part of the synthesis writing craft — the writer writes, the AI assists.