Communication Science

AI-bloated communication is not just annoying — it is harmful. This page explains the cognitive science and psychology that make verbose, synthetic language damaging at the individual level, the team level, and across entire organizations.

Each section is grounded in published research. Where content is still being drafted, a summary of the planned material is included.

Why People Over-Communicate with AI

Before understanding the damage, we need to understand the cause. People reach for AI-generated filler for four predictable reasons.

1

Fear of Seeming Terse

People feel that short messages are rude. AI generates length as a proxy for effort and care, reinforcing this misperception. In reality, brevity is a sign of respect.

2

Performative Effort

A 500-word email looks more thorough than a 50-word one — even when the 50-word version is clearer and more actionable. Length is mistaken for quality.

3

Delegating Responsibility

Letting AI write means the sender doesn't have to own their words. Vague AI output provides plausible deniability and removes the social risk of a clear, direct statement.

4

Approval Seeking

Warm AI phrases like "I hope this helps!" are engineered to feel supportive. Senders borrow that warmth — but the recipient pays the cost in time and attention.

1. Cognitive Load Theory: The Hidden Tax of Every Filler Word

Scientific Foundation · Sweller (1988)

John Sweller's Cognitive Load Theory (CLT), published in Cognitive Science in 1988, established that human working memory has a hard, measurable limit — roughly 4–7 chunks of information at one time (Miller, 1956). CLT distinguishes three types of cognitive load:

Intrinsic Load

The inherent complexity of the topic. You can't reduce it — the information is what it is.

Germane Load

Mental effort used to build new understanding. This is productive cognitive work.

Extraneous Load

Wasteful. Load imposed by how information is presented — unnecessary words, redundant phrasing, filler phrases. AI bloat is almost entirely extraneous load.

Every "Certainly! I'd be happy to help you with that!" before an answer consumes working memory capacity that the reader needs to process the actual information. The receiver's brain has already started allocating resources when it detects an incoming message. Filling those initial slots with noise — instead of signal — forces the reader to mentally discard content before they can begin to understand the message.

Applied to email and chat, research informed by CLT (Mayer & Moreno, 2003; Kalyuga et al., 1999) consistently shows that removing extraneous words reduces comprehension time and error rate, with no loss of perceived quality. The reader understands more, faster, with less effort — and attributes the quality to the writer, not the brevity.

2. The Trust Deficit: When Warmth Backfires

Scientific Foundation · Authenticity, Algorithm Aversion

Humans are exquisitely sensitive to authenticity cues. Decades of social psychology research on interpersonal communication establish that we continuously and unconsciously assess whether emotional signals are genuine or performed. This is sometimes called the sincerity detection system.

When that system detects inauthenticity — as it reliably does with AI-generated pleasantries — two things happen:

1

Trust drops

Research on algorithm aversion (Dietvorst et al., 2015) shows people distrust algorithmically-generated content even when it performs better than human-generated alternatives. Once the reader suspects AI authorship, the credibility of the entire message is discounted — including the accurate, useful parts.

2

Psychological distance increases

Synthetic warmth signals that the sender did not consider the recipient enough to write to them personally. This is worse than no warmth at all. Manufactured empathy ("I understand how challenging this must be for you") is perceived as dismissive rather than supportive.

Hancock et al. (2020) found that disclosure of AI involvement in a message reduces trust even when the content is factually accurate and helpful. The trust deficit is not about the information — it is about the relationship signal embedded in how it was written.

3. Grice's Cooperative Principle: Verbose ≠ Polite

Scientific Foundation · Grice (1975)

Philosopher H.P. Grice formalized how cooperative communication works in his 1975 paper "Logic and Conversation." His Cooperative Principle holds that speakers follow four maxims in genuine dialogue:

Maxim of Quantity

Be as informative as required. No more, no less. AI bloat systematically violates this — it adds words without adding information.

Maxim of Quality

Only assert what you believe to be true. Synthetic empathy ("I truly understand your frustration") asserts an emotional state the AI does not have.

Maxim of Relation

Be relevant. Every sentence should advance the communication. Filler openers and closing pleasantries contribute zero relevance.

Maxim of Manner

Be clear, brief, and orderly. Avoid ambiguity and obscurity. AI verbosity violates manner by burying the message under padding.

When a message violates these maxims, the reader's brain immediately starts searching for the implicature — the implied meaning behind the violation. This is extra, involuntary cognitive work. The reader thinks: "Why are they using so many words? What are they trying to hide, soften, or perform?" The subtext creates noise even when there is none.

4. Individual Impact: What It Does to a Single Person

Research Areas · Attention, Decision Fatigue, Stress

Gloria Mark's long-running attention research at UC Irvine (2005–2023) documents a sharp decline in average focus duration: from 2.5 minutes per task in 2004 to 47 seconds in 2020, driven by digital interruption and communication fragmentation. Every bloated message that requires re-reading or parsing is a micro-interruption.

A

Decision Fatigue

Baumeister et al.'s ego depletion research (1998) demonstrated that willpower and cognitive control deplete with use. Parsing bloated messages is low-value but real cognitive work. Over a full inbox, that depletion adds up — leaving less capacity for actual decisions.

B

Stress and Cortisol

Mark et al. (2016) measured cortisol levels in knowledge workers and found a direct correlation between email volume and physiological stress markers. More messages — even when benign — raise baseline stress. Bloated messages increase the effective volume of information to process.

C

Erosion of Agency

When people routinely send and receive AI-generated text, they lose the habit of articulating their own thoughts clearly. Communication becomes outsourced. This gradually degrades the cognitive skill of clear expression — the same way GPS erodes spatial memory.

5. Organizational & Social Impact: When Bloat Scales

Research Areas · Team Trust, Communication Norms, Social Contagion

What happens at the individual level multiplies when it becomes a team or organizational norm.

Psychological Safety Degrades

Amy Edmondson's research on psychological safety (1999, 2018) defines it as the shared belief that it is safe to speak up, disagree, and be direct. Teams with high psychological safety communicate more honestly and perform better. AI-bloated communication is the opposite signal: it says "I won't commit to a clear position." Over time, a team where everyone hides behind AI-generated hedging becomes a team where nobody trusts anyone's actual opinion.

Communication Norms Are Contagious

Communication styles spread through social contagion (Christakis & Fowler, 2009). When one or two influential team members adopt AI-bloated communication, others mirror it — both because it reduces social risk ("if everyone talks like this, it can't be wrong") and because it signals the norm. Within months, an entire department can shift from direct to padded communication without any deliberate decision.

The Productivity Asymmetry

The sender of an AI-bloated message spends 10 seconds pressing "Generate" and "Send." If the message goes to 10 people and each spends 90 extra seconds parsing it, 15 minutes of total human time was destroyed in one click. At organizational scale — thousands of messages daily — this asymmetry represents a measurable, invisible productivity drain that never appears on any dashboard.

What You Can Do

Lead by example

Write short replies to long AI-generated messages. Signal clearly what you expect: signal-to-noise, not word count. Social contagion works in both directions.

Name it directly

It is acceptable to say: "This looks AI-generated. Can you tell me what you actually think in two sentences?" Naming the norm shift is what breaks it.

Set norms with your team

Agree on communication standards. "Slack messages should be readable in under 15 seconds" is a concrete, enforceable rule. Write it down. Revisit it in retrospectives.

Set your AI tools correctly

You can instruct AI tools to be direct before they write for you. See the Prompt Library for copy-paste instructions.

Key References

Foundational and cited works. Additional references will be added as sections are completed.