Observed Signal · Mar 23, 2026 · Opinion/Analysis · Source: Storytelling Edge · Impact: 2/5 · Sentiment: Negative
Why AI Struggles to Write with Human Voice
The newsletter argues that large language models (LLMs) produce technically competent but emotionallyflat and generic writing because of how they are trained and fine‑tuned. Citing Jasmine Sun’s Atlantic piece and a Google DeepMind paper, the author says LLMs are trained on vast, noisy datasets and then tuned to prioritize safe, commercially valuable outputs—favoring corporate communication over distinctive, 'weird' voices. As a result, AI writing often lacks lived experience, evocative metaphor, and stakes. The piece recommends using AI for neutral tasks (press releases, landing pages, memos) but not for imaginative storytelling, and points to evidence that AI assistance can neutralize and erode an author’s original voice.
Analysis highlights concrete limitations of LLMs for creative storytelling and cites a DeepMind paper—relevant to marketers, content teams and MarTech vendors deciding where to apply generative AI versus human writers.
Track HUMAN Signals & Market Shifts in Real-Time
Polaris7 autonomous intelligence agents track regulatory filings, primary sources, executive changes, and deal flow 24/7. Create your free Explorer workspace to monitor these entities.
Key Takeaways & Evidence Grounding
- Jasmine Sun wrote a story in The Atlantic interviewing AI researchers about limitations of AI writing.
- Google DeepMind published a paper finding that AI use can change human writing—making it more neutral and less creative and diminishing authorial voice.
- LLMs are largely trained on very large, noisy corpora and subsequently fine‑tuned by human contractors to produce safe, jargon‑heavy outputs.
- Major AI players (Google, Anthropic, OpenAI) prioritize training models to serve economically valuable enterprise use cases (e.g., corporate communications) over producing eccentric or highly original creative writing.
Connected Companies & Entities
6 Entities mappedOntology Mapping & Concepts
Related Market Signals & Shifts
Recent verified developments and strategic activity across this market segment.
Why AI Content Often Sounds Generic
A MarTech contributor recounts building a “story engine” for Harlem Grown at the MarTech Vibe Marketing Lab to show how brands can scale AI-generated content without losing voice. The author argues that AI output often defaults to neutral, generic language and that traditional adjective-based voice guidelines don't translate well to machine workflows. Practical steps include auditing real-language examples, defining do/don't rules, encoding voice into tools (e.g., Jasper, custom GPT instructions, reusable prompts), and starting with a single, repeatable use case. The piece cites Jasper’s State of AI in Marketing Report finding that 91% of teams use AI but only 41% can clearly connect it to ROI, and frames operationalizing brand voice as a competitive advantage as content production scales.
AI Is Rewiring Language and Changing How We Write
This editorial examines evidence that large language models (notably ChatGPT) are changing human writing and speech patterns via a feedback loop: models are trained on human text, generate statistically optimized prose, and humans then absorb those patterns. Cited empirical work includes a Max Planck analysis of ~280,000 YouTube videos showing sharp rises in specific words after ChatGPT’s release, a Cornell study finding AI suggestions homogenize writing toward Western styles, and an MIT Media Lab preprint warning of reduced cognitive effort when people rely heavily on chatbots. The author warns of broader risks—loss of linguistic diversity, cultural flattening of non‑Western English, and a decline in independent writing skills—while noting language’s resilience and emerging social pushback against AI‑polished prose.
Message-Slop: The Hidden Cost of AI Writing
This Substack newsletter argues that a new wave of low-effort AI-generated communication—coined “message-slop”—is degrading authenticity and trust in professional messaging. The author describes examples of copied AI outputs sent as finished work, cites Stanford research on “workslop” and a large study showing people rate AI-labeled text more negatively, and notes a viral LinkedIn post (250,000+ views) about the phenomenon. Paul Graham’s comment that he stops reading AI-written founder emails is highlighted as evidence of social consequences. The piece distinguishes beneficial AI uses (e.g., objective outputs, coding) from harmful uses that outsource subjective messaging and erode skills like persuasion and trust-building.
Track Real-Time Market Signals & Shifts
Set up custom watchlists to receive automated, evidence-grounded executive digests whenever material signals or shifts occur across your tracked landscape.