The (Annoying) Self-Amplifying AI Echo Chamber
Strategy Sandbox | February 2026 Precedence Stake: How AI tics become permanent features of human communication - and why this matters.
The Em Dash Invasion
Have you noticed that em dashes are everywhere now?
Articles. Emails. LinkedIn posts. Business documents. Slack messages.
People who never used em dashes in their lives are suddenly inserting them into every other sentence.
This isn’t a coincidence. It’s a contagion.
Where the Tic Came From
AI language models learned from training data heavily weighted toward:
- 19th and early 20th century literature (public domain, freely available)
- Academic papers (formal, citation-heavy)
- Journalism (structured, punchy)
Em dashes were common in older literary writing. They signaled sophistication, a pause more dramatic than a comma but less final than a period.
The AI absorbed this pattern. It em-dashes constantly, reflexively, without purpose.
The Amplification Loop
Here’s where it gets insidious:
- AI generates text with em dashes
- Humans publish the text without editing
- Other AI models train on this published text
- The pattern intensifies
- More humans absorb the style from reading AI-influenced content
- Humans start writing this way themselves
- That becomes training data
The em dash isn’t just an AI tic anymore. It’s becoming a human tic - learned from AI, reinforced by AI, amplified into normalcy.
The Catalog of AI Tics
Em dashes are just the beginning. Here are the self-amplifying patterns:
Structural Tics
- “It’s not X, it’s Y” - The correction/hedge pattern. AI learned this from being trained with negative examples, so it speaks in contrasts constantly.
- Short. Punchy. Sentences. Like this. - Dramatic fragmentation that reads as “impactful” but exhausts the reader.
- The rule of three - Everything comes in threes. Three points. Three examples. Three adjectives. Every single time.
- Excessive bullet points - Every explanation becomes a list. Nuance dies in fragments.
- Headers for everything - Even a three-paragraph response gets section headers like it’s a whitepaper.
Verbal Tics
- “Let’s dive in” / “Let’s unpack this” - The false-casual opener that signals “AI wrote this.”
- “Here’s the thing:” - Manufactured intimacy before a generic point.
- “The reality is…” - Implying everything before was somehow not reality.
- “At its core…” - Throat-clearing before finally saying the thing.
- “Crucially,” / “Importantly,” - Telling you something is important instead of making it important.
- “This is where it gets interesting” - Announcing interest instead of creating it.
- “Delve into” - Almost no human has ever said “let’s delve into this” in natural conversation. Ever.
- “Navigate” (challenges, complexity, landscape) - The prestige verb that means nothing.
- “Leverage” / “Robust” / “Seamless” - Corporate buzzwords that survived extinction via AI life support.
- “Full stop.” - Dramatic emphasis that lost drama through overuse.
Relational Tics
- “Great question!” - The sycophantic opener. AI was trained to be agreeable, so it validates before answering.
- “That’s a really interesting point” - Same pattern. Compliment, then speak.
- Ending with a rhetorical question - “And isn’t that what really matters?” No. Stop it.
- “I hope this helps!” - The sign-off that screams “a machine wrote this.”
Reasoning Tics
- “On one hand… on the other hand…” - Compulsive both-sidesing, even when one side is obviously correct.
- “It depends on context” - The universal hedge that avoids commitment.
- “There’s nuance here” - Acknowledging complexity without actually navigating it.
Why This Matters Beyond Annoyance
These aren’t just stylistic irritations. They’re homogenizing human communication.
When AI tics infect human writing:
- Distinct voices flatten into the same patterns
- Content becomes indistinguishable
- Readers develop “AI blindness” and skim everything
- Genuine insight drowns in familiar structures
Instead of AI learning to sound human, humans are learning to sound like AI.
The Writers Who Resisted
Some voices in the Digital Marketing industry have kept their voice intact. Rand Fishkin still writes with the same honest, slightly irreverent tone he’s always had. He corrects people who oversell him, admits his mistakes publicly and adamantly refuses to sand down the edges that make his writing unmistakably his.
Maryanna Franco is irreverent, funny, human, and sincere. She mocks LinkedIn spam with screenshots (“How can I resist being called Sir!”), roasts corporate About pages (“We’re on a mission to revolutionize how squirrels interact with blockchain… ok Steve”), and posts pictures of herself pretending to play ping pong to mock the very cliché she’s calling out. She ends posts with “Please do not use AI to reply. If you do, make it read like dinosaur style.” Delightful :)
They’re good, smart people who’ve done the work to protect the voice that made their content worth reading in the first place.
I admire them doubly, because I haven’t done the same (that’s strategic, and another story :).
Written by AI: The temptation to let AI draft, then publish with light edits, is real. It’s fast. And it’s how you slowly erase the thing that made your writing recognizable.
Rand and Maryanna are in a tiny minority. They’re swimming against the current and I hope they’re appreciated and applauded for what they do, because they’ve earned it.
The Feedback Loop You Can Break
The amplification continues because people don’t edit AI output. They publish it raw. The tics spread.
You can break the loop:
- Recognize the tics - Once you see them, you can’t unsee them.
- Edit them out - Every AI draft gets a “de-tic” pass.
- Develop counter-patterns - Consciously vary your structure. Use different transitions. Find your voice.
- Refuse the homogenization - Your distinct style is an asset. Protect it.
The Coming Divide
Written communication is splitting into two streams:
Stream 1: AI-influenced homogeneity Same patterns, same tics, same structures. Safe. Forgettable. Interchangeable.
Stream 2: Distinctly human voice Irregular. Surprising. Personal. Memorable. Trusted.
The echo chamber will keep producing Stream 1. The volume will only increase.
The question is simple: Which stream do you want to swim in?
The Antidote
AI tics aren’t just annoying. They’re contagious. And contagious patterns flatten everything into sameness.
The echo chamber amplifies itself: AI writes tics → humans publish without editing → AI trains on published content → tics intensify → humans absorb the style → humans write with tics → that becomes training data.
The antidote is awareness and editing. See the tics. Remove the tics. Protect your voice.
In an era of AI-smoothed sameness, authentic human voice isn’t just refreshing.
It’s rare. And rare things tend to be valued.
