Master library
AI copywriting tell library
An exhaustive taxonomy of the words, phrases, structures and habits that make copy feel generated — organised so you can score them systematically.
~250 tells · 18 diagnostic categoriesAI-associated does not mean forbidden.
AI uses ordinary English. Trying to purge every word it favours would make the writing worse. The point of this library is to spot overuse, predictability, genericity and lack of intention — then give the writer better choices.
Humans write these tells too. The diagnosis is about autopilot writing, not machine authorship.
It might not always be AI
When a “tell” isn’t a tell — and what to look for instead.
Stop hunting for “AI words.” Most of them are just standard English that algorithms learned from us. One word doesn’t make copy synthetic — lazy patterns do.
Robust, seamless and leverage existed before ChatGPT. A single appearance isn’t proof of a prompt — it’s vocabulary. Purging useful words only weakens the writing.seamless is fine. Seamless, effortless, streamlined and innovative crammed into one sentence is where copy dies. Repetition is the culprit, not the word itself.Utilise works in a technical spec. On a homepage, use wins. Words aren’t inherently “AI” — they’re either useful or out of place.18 diagnostic categories
Score the draft against these. Drill into individual tells underneath.
| # | Category | What it catches |
|---|---|---|
| 1 | Word choice | AI-associated vocabulary |
| 2 | Corporate language | Inflated / formal wording |
| 3 | Generic marketing | Clichés and empty claims |
| 4 | Rhetorical patterns | Not X/Y, rule of three, templates |
| 5 | Metaphors | Landscape, ecosystem, journey… |
| 6 | Vagueness | Abstract / general language |
| 7 | Specificity | Missing concrete information |
| 8 | Repetition | Words, ideas, structures |
| 9 | Sentence-level | Rhythm and syntax |
| 10 | Page architecture | Formulaic organisation |
| 11 | Conversational tells | “Here’s the thing…” |
| 12 | Manufactured personality | Fake human voice |
| 13 | Hype / evidence | Unsupported claims |
| 14 | Formatting | Visual AI patterns |
| 15 | Voice | Distinctive human voice |
| 16 | Originality | Original thought / expression |
| 17 | Information density | Substance per sentence |
| 18 | Authenticity | Truthfulness / real experience |
What a cluster looks like
One paragraph, five tells. This is what “overuse” means in practice.
Word choice
AI-associated vocabulary. Score clusters and overuse — not single hits.
1.1 AI-associated verbs
1.2 AI-associated adjectives
1.3 AI-associated nouns
1.4 Inflated alternativesPrefer the plain form unless the formal one earns its place
Corporate language
Inflated and formal wording that adds weight without adding meaning. (See also 1.4 inflated alternatives.)
Latinate stack
Formal verbs where plain ones are clearer.
Noun chains
Stacked abstractions with no actor.
Prepositional bloat
Long prepositions for short ideas.
Ability circumlocution
Three words where one works.
Generic marketing language
Clichés and empty claims that could sit on any site in the category.
2.1 Generic claims
2.2 Generic transformation language
2.3 Generic superiority
2.4 Generic reassurance
Rhetorical patterns
Templates AI reaches for when it needs structure without substance.
Contrast
Not X, but Y
Not only X, but Y
X without Y
Less X, more Y
Transformation & dual frames
Whether / Imagine / What if
Metaphors
Environment, journey, construction, movement, change — and the literary clichés.
Environment
Journey
Construction
Movement & change
Literary clichés
Vagueness
A major scoring category. Core test: Can I replace the generic noun with something concrete?
Generic audiences
Generic problems
Generic outcomes
Generic actions
Generic objects
Specificity failures
What’s missing when the sentence sounds precise but isn’t.
Missing people
Role erased into a blob audience.
Missing situations
“Manual processes” with no scene.
Missing mechanisms
Outcome with no how.
Missing evidence
Assertion without proof.
Missing consequences
Saved time that goes nowhere.
Missing context
Could be any industry.
Fake specificity
Precision theatre.
Repetition
Words, ideas, benefits, structures, sections, CTAs.
Lexical
Same word repeated until it dulls.
Semantic
Same idea restated in different clothes.
Benefit chain
One benefit spun into three near-synonyms.
Structural
Same sentence machine, new nouns.
Section
Every block: heading → claim → three bullets → soft close.
CTA echo
Mini-CTA at the end of every section.
Sentence-level tells
Rhythm, symmetry, fragments, openings, overqualification.
Sentence symmetry
Rule-of-three saturation
Three benefits everywhere.
Uniform sentence length
Same approximate length, page after page.
Excessive short sentences / fragments
More growth.
Better results.
Repeated openings
Overly neat cadence
Everything equally polished and balanced.
Overqualification
Paragraph & page architecture
Formulaic organisation at section and page level.
The AI blog structure
A fixed pipeline regardless of the topic.
Identical section structures
You can predict the next block before you read it.
Over-sectioning
Headings doing the work of sentences.
Over-signposting
Announcing structure instead of using it.
Manufactured conclusions
Every block ends with the same bow.
Information symmetry
Minor points get the same airtime as the real differentiator.
Conversational AI tells
Artificial hooks, engagement, reassurance, urgency.
Artificial hooks
Artificial engagement
Artificial reassurance
Artificial urgency
Manufactured personality
Fake vulnerability, discovery, experience, authority.
Fake vulnerability
Fake discovery
Fake personal experience
Fake authority
…without identifying the evidence.
Hype vs evidence
Score the claim-to-proof ratio. High-risk claims need evidence.
Where’s the evidence?
If the claim is high-risk and the proof is missing, the score should reflect that — whether or not the vocabulary is “AI-ish”.
Formatting tells
Visual patterns that often travel with generated drafts.
Voice
Beyond detection: does this sound like a believable human author?
- A recognisable point of view
- Appropriate informality
- Natural contractions
- Natural vocabulary
- Personality
- Appropriate humour
- Genuine opinion
- Industry fluency
- Audience awareness
- A believable human author
Originality
The positive side of humanisation — what to reward, not only what to flag.
- Original observations
- Uncommon examples
- Genuine opinions
- Specific customer insights
- Unexpected comparisons
- Distinctive phrasing
- Real-world details
- Company-specific knowledge
- Useful disagreement
- Interesting anecdotes
- Information competitors wouldn’t naturally have
Information density
How much useful information each sentence delivers.
Our innovative platform empowers modern businesses to unlock greater efficiency.
Connect Salesforce to Snowflake and let sales reps pull customer data without asking an analyst.
Look for: who + what + how + why + evidence + consequence.
Authenticity
Separate from “AI-ishness”. Is the writing true to experience and evidence?
- Are anecdotes real?
- Are customer claims verifiable?
- Are numbers sourced?
- Are opinions genuinely held?
- Is personal experience genuine?
- Are testimonials real?
- Are claims appropriately qualified?
- Is the brand actually capable of doing what it says?
This list will keep growing. Spotting a tell doesn’t mean the copy is bad — it means AI may have had a hand in it. The job is to decide whether that pattern is earning its place.
From library to scorecard
These 18 categories feed the Humanisation Methodology: diagnose the tells, then rewrite with judgement.
See the methodology →