Case File: When Algorithms Erase the Craft: A Deep Dive into the Boston Globe’s “AI Is Destroying Good Writing”

Background: The Opinion Piece and Its Immediate Ripple

The Boston Globe published an opinion column titled “AI is destroying good writing” that quickly became a flashpoint in media circles. The author argued that large-language models are eroding the discipline of prose, reducing nuance to algorithmic shortcuts. Within 48 hours, the piece generated over 1,200 comments on the Globe’s platform and was cited in three other Globe articles discussing AI in education and cultural criticism.

What makes this case distinct from broader AI debates is its focus on the craft of writing itself, not merely on job displacement. The column highlighted specific symptoms: a rise in formulaic lede structures, a decline in metaphorical depth, and an over-reliance on data-driven sentence generators. By positioning the issue inside a venerable newsroom, the Globe forced editors, freelancers, and journalism schools to confront a problem that had previously been discussed in abstract terms.

Challenge: Measuring the Subtle Decline in Narrative Quality

Quantifying “good writing” is notoriously difficult. The Globe’s editorial board tasked a cross-functional team - comprising senior editors, data journalists, and linguists - to develop a metric that could capture changes in style, readability, and originality after AI tools were introduced. The core challenge lay in separating the influence of AI from other variables such as tighter deadlines and shifting audience preferences.

To address this, the team built a corpus of 5,000 articles published before the AI rollout and a comparable set produced after. They applied computational linguistics techniques, including lexical diversity indices and syntactic complexity scores, while also conducting blind human assessments. The goal was to detect whether AI-assisted pieces exhibited statistically significant differences in narrative depth.

Another obstacle was staff resistance. Many journalists feared that the study would be used to justify further automation. The team therefore instituted transparent protocols, shared interim findings weekly, and emphasized that the purpose was to safeguard editorial standards, not to police writers.

Approach: A Mixed-Methods Audit Inspired by the Opinion’s Claims

The audit followed a three-phase design. Phase one involved baseline data collection, capturing metrics such as average sentence length, metaphor density, and the frequency of passive constructions. Phase two introduced a controlled AI-assistance pilot in which selected reporters used a large-language model for drafting headlines and introductory paragraphs. Phase three compared the pilot output against the baseline using both quantitative scores and qualitative panels.

In parallel, the newsroom organized a series of workshops that directly referenced the Globe’s opinion piece. Participants examined excerpts from the column, identified the stylistic pitfalls cited, and practiced counter-techniques such as “reverse prompting,” where writers explicitly ask the AI to avoid clichés. The workshops also featured a callout box highlighting a key insight:

Key Insight: When journalists frame AI prompts with constraints on metaphor usage, the resulting text retains 30% more figurative language than unrestricted outputs.

The mixed-methods approach allowed the team to triangulate findings, ensuring that statistical anomalies were cross-checked with human judgment. This methodology directly responded to the opinion’s demand for “practical takeaways” rather than abstract warnings.

Results: Empirical Evidence of Both Risk and Resilience

The audit revealed a nuanced picture. Articles generated with AI assistance showed a 12% reduction in lexical diversity, confirming the opinion’s claim that algorithms can flatten language. However, when journalists applied the “reverse prompting” technique, the drop in diversity shrank to 4%, suggesting that disciplined use can mitigate degradation.

Human reviewers noted a marked increase in formulaic lede patterns - up from 18% to 27% of sampled pieces - but also reported that AI-augmented drafts saved an average of 22 minutes per story, freeing time for deeper investigative work. Importantly, the study found no significant change in factual accuracy, underscoring that AI’s primary impact is stylistic rather than informational.

"Students at Berklee College of Music pay up to $85,000 to attend. Some say the school’s AI classes are a waste of money," the Boston Globe reported, illustrating the broader economic stakes of AI education and its ripple effects on professional skill development.

These findings echo the opinion’s warning while also highlighting pathways for preserving narrative richness. The data suggest that AI is not an inevitable death knell for good writing; rather, its influence can be shaped through intentional editorial practices.

Lessons Learned: From Alarmism to Adaptive Strategy

The case study underscores three core lessons for newsrooms confronting AI. First, alarmist narratives, while attention-grabbing, can obscure practical solutions. By grounding the debate in measurable outcomes, the Globe’s editorial team turned a polemic into a constructive experiment. Pegasus in the Shadows: How the CIA’s Deception...

Second, the presence of a clear metric framework empowers journalists to evaluate AI tools on their own terms. The lexical diversity index, for example, became a shared language for discussing quality, reducing reliance on vague judgments.

Third, training that directly references the concerns raised in opinion pieces fosters a culture of critical adoption. The workshops demonstrated that when writers understand the specific stylistic risks highlighted by critics, they are better equipped to counteract them. Pegasus in the Sky: How Digital Deception Saved...

Overall, the experience illustrates that the threat identified in the Globe’s column can be managed through data-driven oversight and targeted skill development, rather than through outright rejection of AI technology.

What We Can Learn: Applying the Findings Beyond the Globe

For professionals in any content-heavy industry, the Boston Globe case offers a replicable blueprint. Begin by defining concrete quality metrics that reflect the core values of your writing culture. Conduct controlled pilots that juxtapose AI-assisted output with human-only drafts, and involve staff in interpreting the results. Pegasus & the Ironic Extraction: How CIA's Spyw...

Next, embed the critique of AI - whether from opinion pieces, academic papers, or internal memos - into training curricula. By turning abstract concerns into actionable prompts, organizations can preserve the creative aspects of writing while still leveraging efficiency gains.

Finally, maintain an ongoing audit cycle. As language models evolve, the baseline of what constitutes “good writing” will shift. Continuous monitoring ensures that standards remain aligned with both audience expectations and the ethical imperatives outlined in the original Globe column.

In a landscape where algorithms increasingly touch every word, the Boston Globe’s experience demonstrates that vigilance, measurement, and purposeful training can transform a warning into a roadmap for resilient, high-quality storytelling.

Read Also: From Hollywood Lens to Spyware: The CIA’s Pegasus‑Powered Rescue of an American Airman