AI SLOP. Smells Like AI Spirit. Why Does My Writing Taste and Smell Like AI?

We have spent over two years working daily with different large language models (yes, LLMs), and as a user, I am honestly tired of constantly reading and listening to 100% AI-manufactured text and scripts across blog posts and social platforms like LinkedIn, Instagram, YouTube, etc.

Enough is enough, don't you think?

This is something that has fascinated me and pushed me to investigate deeply, especially as media outlets and tech analysts finally begin putting a name to the problem.

This isn't about ditching the tools altogether. Using them to organize thoughts, discover other angles, or speed up drafting is completely valid. The problem starts when we hand everything over to whichever AI model is on hand and let the algorithm take control of our writing, or surrender our editorial judgment altogether.

The result is unmistakable, as demonstrated by these two actual drafts generated by AI when attempting to write the introduction to this very article:

Neither of these drafts reflects my real voice, even if I agree with the underlying idea. Expressions like "rhetorical cowardice" or "verbal plastic" are words I would never use in a real conversation, and I wouldn't take so many detours just to make a direct point.

At first glance, it may look fluent, but it's just the algorithm's default mold talking.

Language models have their language so deeply conditioned by their training that it's remarkably difficult to keep them from stumbling into the exact same pitfalls. You can painstakingly prompt them to avoid every single one of these clichés, but sooner or later they default right back to the same mold.

Let's look under the hood to see why this happens.

Why Does AI Produce Predictable Writing?

The term AI SlopGlosarioAI SlopTérmino que define la morralla, bazofia o contenido basura generado en masa por modelos de inteligencia artificial sin supervisión ni valor editorial. Proviene del inglés slop (literalmente la comida de desperdicios o sobras que se echa a los cerdos). En la cultura digital representa el sucesor directo del spam: textos inflados, imágenes clónicas, código genérico y relleno algorítmico sin criterio ni alma que saturan internet y provocan el colapso de los propios modelos al retroalimentarse de su propia basura.Ver término completo → refers to low-quality, low-effort digital content produced en masse by generative models, characterized by a total lack of human creativity, intent, or editorial care. In modern internet culture, as highlighted by publications like The Guardian and glossaries like UltralyticsGlosarioAI SlopTérmino que define la morralla, bazofia o contenido basura generado en masa por modelos de inteligencia artificial sin supervisión ni valor editorial. Proviene del inglés slop (literalmente la comida de desperdicios o sobras que se echa a los cerdos). En la cultura digital representa el sucesor directo del spam: textos inflados, imágenes clónicas, código genérico y relleno algorítmico sin criterio ni alma que saturan internet y provocan el colapso de los propios modelos al retroalimentarse de su propia basura.Ver término completo →, all these bloated paragraphs, carbon-copy images, and empty filler flooding feeds just to game engagement have emerged as the direct successor to spamGlosarioSpamEnvío masivo, no solicitado y repetitivo de mensajes, correos electrónicos o contenidos publicitarios a través de internet. El término se popularizó a partir de un famoso sketch del grupo humorístico Monty Python (donde repetían incesantemente la palabra de una marca de carne enlatada) y se convirtió en el concepto universal para definir el correo basura en la era digital clásica, predecesor histórico del fenómeno del AI Slop.Ver término completo →—not merely as unwanted email, but in its core definition as digital junk manufactured to pollute public feeds.

To sound human, you have to write like a human. There is no other way. With your own flaws, rhythm, and raw voice. Using AI to structure thoughts or unblock angles is completely valid; the problem comes when we surrender our judgment and let the algorithm take control of our writing.

A language model does not comprehend intent or nuance the way a human does; it calculates mathematically which token is most probable next. By systematically picking the safest, most frequent statistical option in its corpus, the algorithm averages out human language, sanding down any risky, spontaneous, or personal edge.

ANALOGY — Imagine a star-shaped silicone baking mold.

If you pour chocolate batter into it, you get a dark star. If you pour lemon batter, you get a yellow star. If you bake carrot cake, you pull out an orange star.

The ingredients (the topic) change completely, but the silhouette, edges, and final structure are always identical.

AI operates the exact same way. You can ask for an essay on astrophysics, a YouTube script, or a sales pitch. It will swap out the topical ingredients, but force them systematically into the exact same prefabricated rhetorical template.

Below, we break down the four decisive patterns that give this mold away, how to eliminate them from your writing, and other secondary habits to watch out for.

1. False Contrast: "It's Not About X, It's About Y"

This is one of the most overused crutches in online scripts (YouTube, TikTok, and LinkedIn).

We are talking about soundbites like "It's not about making money; it's about buying freedom" or "The issue isn't the technology, but how we choose to use it".

AI leans on this formula to simulate depth by inventing an artificial dilemma or straw man, which it immediately demolishes with a supposedly revelatory conclusion.

How It Manifests in Scripts and Social Media

This pattern runs rampant among creators with heavy daily production schedules (such as Spanish analysts and content creators like Marc Vidal or the Shorts by José Antonio Vizner on Negocios TV). Their underlying analysis is often rigorous, but the pressure to manufacture dramatic punchlines shoves LLMs into the exact same rhetorical mold.

This is a real example (in Spanish) taken from Marc Vidal's analysis in his video "Chaos in Ceuta: one country exports its youth and the other exports blame":

"At the end of all this there's no identifiable villain, or is there? There's a state that figured out it's cheaper to export its young people and collect the remittances they can send than to build them a labor market. And there's another state, with far more resources, that discovered it's cheaper to blame every failure on an outside force than to fix that failure.
 
Every time we hear an official explanation that places the cause of a problem outside the reach of whoever is offering it, stop for a second and ask yourself how long that person has held the authority to fix it.
 
This isn't an exercise in outrage, it's an exercise in accounting."

Analysis of the Closing Line: "This isn't an exercise in outrage, it's an exercise in accounting"

When delivered by a seasoned broadcaster, this closer lands with rhythm, force, and a crisp dramatic finish. However, it doesn't always reflect spontaneous rhetoric, but rather an AI-assisted script tuned for maximum punch.

The problem doesn't lie in the validity of the argument, but in the automated cadence and tone. An attentive listener accustomed to podcasts, video essays, and media broadcasts quickly spots that underlying template—no matter how much the speaker tries to mask it with personal catchphrases to hide the mechanical feel of synthetic clichés.

Guidelines for Your Voice, Videos, or Personal Blog

↑ Back to see this pattern in the intro example

2. Uniform Sentence Length and Lack of Burstiness

AI detectors (like GPTZero or statistical classifiers) don't look for typos or bad grammar. They measure the mathematics of language, analyzing two primary variables:

1. PerplexityGlosarioPerplejidad / PerplexityMétrica estadística que mide el grado de incertidumbre de un modelo de lenguaje al predecir la siguiente palabra en una secuencia. Cuanto más baja es la perplejidad, más predecible y probable resulta el texto. En detección de contenido sintético, los textos generados por IA suelen registrar una perplejidad muy reducida debido a que el algoritmo selecciona sistemáticamente los tokens estadísticamente más frecuentes.Ver término completo →. How predictable each word is in the sequence.
2. BurstinessGlosarioBurstiness / Inyección de VariabilidadMétrica estadística que mide la variabilidad en la longitud de las oraciones. Un texto con alta burstiness alterna frases breves con explicaciones compuestas para crear un ritmo natural. En cambio, los modelos de lenguaje tienden a generar frases de longitud uniforme, lo que produce una cadencia monótona. La técnica de Burstiness Injection consiste en editar borradores de IA introduciendo oraciones cortas de impacto para romper la simetría y recuperar el ritmo humano.Ver término completo → (or Rhythmic Explosiveness). The statistical variance in sentence length, density, and structure.

Now you know where Perplexity AI gets its name.

In traditional writing, we call this cadence. Technical AI detectors measure a related property called burstiness: the statistical variance in sentence length across a passage. They are not the same thing — cadence is broader, encompassing sound, tone, and flow — but they point to the same underlying problem.

Human thought moves in bursts. When we are passionate about an idea, we stretch it into a long, layered sentence loaded with subordinate clauses and momentum; then we brake hard. Three words. A breath.

That gap is what makes prose breathe.

AI has neither lungs nor enthusiasm. Because it optimizes for minimal statistical risk, its burstiness becomes flat as it churns out uniform blocks of 18 to 25 words with the mechanical precision of a metronome.

"This sentence has five words. Here are five more words. Five-word sentences are fine. But several together become monotonous. Listen to what is happening. The writing is getting boring. The sound of it drones. It’s like a stuck record. The ear demands some variety.
Now listen. I vary the sentence length and I create music. Music. The writing sings. It has a pleasant rhythm, a lilt, a harmony."
— Gary Provost, 100 Ways to Improve Your Writing

Guidelines to Break Rhythmic Monotony

↑ Back to see this pattern in the intro example

3. Compulsive AI-Splaining & the Colon Overuse Epidemic

One of the most persistent dead giveaways in AI writing is the compulsive urge to spoon-feed and define every single concept it introduces, as if terrified that the reader won't follow along.

This tic manifests through two specific mechanics:

A. Colon addiction and the set-up/punchline cadence

As content strategist Blake Stockton highlights in his analysis Don't Write Like AI: Colons, Colons Everywhere, AI models lean obsessively on the colon-header formula ("Concept: Explanation of why it matters").

Because the algorithm struggles to build organic narrative bridges, it defaults to a mechanical set-up/punchline cadence: introduce a bolded term, drop a colon, and dump a textbook definition. This habit has grown so pervasive that it has sparked what essayist James Christopher termed "The Great Punctuation Purge"—where human authors intentionally strip colons and dashes from their own drafts for fear of being accused of writing like a bot.

B. The Three Pillars of Synthetic AI-Splaining

1. Faux-Insight: The model's tendency to present a total platitude as a profound philosophical revelation ("The true value of data isn't storing it, but interpreting it").
2. The Strawman Confusion: Derived directly from the classical straw man fallacy. The model plants a flimsy, manufactured dilemma into the reader's mind ("Many mistakenly believe that...") simply to knock it down in the very next sentence and claim a cheap breakthrough.
3. Algorithmic Condescension: The interface psychology and computational linguistics term for why AI over-explains: the model defaults to assuming the reader is confused, adopting the patronizing cadence of a grade-school teacher who halts the prose to define every basic term.

The result is pure illusion of insight: prose that mimics deep analysis without providing any new substance.

Guidelines to Trust Your Reader

↑ Back to see this pattern in the intro example

4. Discourse Scaffolding: Announcing What You Are About to Say

Throat-Clearing in Writing
In professional editing, throat-clearing is the writer narrating the communication process instead of delivering the message. It happens whenever an explanatory preamble is dropped ahead of an idea ("In this section, we will delve into..."). Strip that scaffold away, and the core thesis not only stands on its own, but gains immediate density and momentum.

In applied linguistics and discourse analysis, this phenomenon falls under interpersonal and organizational metadiscourse. Research shows that large language models systematically rely on algorithmic hyper-signposting (AI signposting) to manufacture synthetic authority:

• "In this article / video, we will delve deep into..."
• "The goal of this section is to clarify..."
• "It is crucial to highlight that..."
• "As we will see below in detail..."
• "In conclusion, we can summarize that..."

Why AI Abuses Metadiscourse

Lacking firsthand lived experience or authentic domain intuition, large language models default to a low-friction scaffold and pad the composition by mimicking the external mechanics of formal genres. They generate formulaic preambles and transitional padding simply to prove they are executing the user's prompt.

The Musician and the Microphone Analogy

Imagine a pianist who, before playing each key, leans into the microphone and announces: "Next, I will play an F-sharp".

That is exactly what a language model does with metadiscourse: it breaks the immersion and ruins the rhythm of the piece by explaining what it is about to do before doing it.

Guidelines to Strip the Scaffolding and Maximize Density

↑ Back to see this pattern in the intro example

Extra. Other Secondary Habits to Keep in Mind

There are two other common LLM habits worth watching. Unlike the patterns above, these elements are completely normal parts of human language, but AI relies on them so heavily that they instantly give away a synthetic draft:

A. Empty Adjectives and Overblown Rhetoric

Models default to grandiosity ("groundbreaking", "fascinating", "disruptive", "unprecedented paradigm") to compensate for their lack of firsthand lived experience. In moderation they show enthusiasm, but in excess they destroy technical credibility.

B. Unnecessary Introductions

Formulas like "furthermore", "consequently", "moreover", or "it is worth noting" belong in formal academic journals. In blog posts and social content, dropping them into every paragraph turns your prose into cardboard.

Editorial Checklist: The 6-Step Human Filter

Before publishing an article, newsletter, or script generated or assisted by AI, run your draft through this systematic cleanup protocol:

Editorial Prompting Template for Ongoing Auditing (Lightweight Version)

There is no definitive prompt that guarantees "semi-human" text. Your own critical judgment remains the irreplaceable filter.

That said, running a repeatable audit prompt helps you catch algorithmic clichés and subtle AI tics with every new draft or edit.

You can drop this lightweight XML template into your AI assistant as an editorial review prompt:

<editorial_audit version="1.0">
  <role>Senior technical editor specializing in assertive, human, and direct prose.</role>
  
  <rules>
    <rule id="no_scaffolding">Eliminate introductory throat-clearing and announcements ("In this article...", "Below we will explore...").</rule>
    <rule id="no_false_contrasts">Remove symmetric "It's not about X, but Y" formulas. State positive facts and verifiable realities directly.</rule>
    <rule id="no_colons_as_crutches">Avoid using colons (':') as definitional anchors or dramatic crutches. Integrate terms smoothly into sentences.</rule>
    <rule id="no_didactic_lecturing">Purge condescending advice, patronizing commentary, or moral lessons telling the reader how to feel.</rule>
    <rule id="high_burstiness">Alternate very short sentences (3 to 6 words) with compound analytical sentences to break rhythmic monotony.</rule>
    <rule id="action_verbs">Prioritize active verbs over passive state verbs (is, seems to be, represents, serves as).</rule>
  </rules>
 
  <lexical_blacklist>
    synthetic content, synthetic patterns, fast-paced world, comprehensive analysis, catastrophic collapse, crucial, fundamental, essential, revolutionary, unprecedented, authentic, verbal plastic, In conclusion.
  </lexical_blacklist>
 
  <task>
    Audit the following draft. Identify any algorithmic writing vices and rewrite the text strictly adhering to these editorial guidelines:
    
    [PASTE_YOUR_DRAFT_HERE]
  </task>
</editorial_audit>

This ongoing audit technique acts as a recurring quality gate before final human review, ensuring new revisions never reintroduce automated writing tics.

A machine can spit out an average draft in seconds. The authority, texture, and genuine voice can only come from you.