By Anika Collier Navaroli
In the rapidly evolving landscape of contemporary journalism, the integration of artificial intelligence into the editorial workflow has transitioned from a speculative curiosity to an unavoidable reality. As newsrooms grapple with the potential of large language models (LLMs) to streamline production, the question of how to ethically deploy these tools has become paramount. Transparency is the bedrock of journalistic integrity, yet when it comes to AI, many organizations remain opaque, leaving readers to guess where human insight ends and algorithmic generation begins.
To address this, we must adopt a policy of "radical transparency." Rather than offering vague disclaimers, journalists should embrace a show-don’t-tell philosophy. By documenting the specific ways AI interacts with our research, drafting, and distribution, we can foster a deeper sense of legitimacy and trust with our audience.
The Mirage of Fact: A Case Study in Hallucination
The necessity for skepticism when using AI was made clear during the initial research phase for this column. While outlining my thoughts, I recalled a narrative—common in tech circles—that a prominent tech journalist had successfully trained a local LLM on their own body of work to serve as a personal archive. Seeking to verify this fact, I turned to Google Search’s AI Overview.
The AI confidently corroborated my recollection, confirming that such an experiment had indeed taken place. However, maintaining the "trust but verify" principle essential to the craft, I prompted Google’s Gemini to provide a primary source for the claim. The result was a stark reminder of the "black box" problem: Gemini failed to produce a single citation, eventually contradicting the initial confirmation provided by the AI Overview.
It became evident that the LLM had hallucinated, effectively echoing my own biased recollection back to me. This failed experiment serves as a critical lesson: AI is not a reliable oracle of truth. It is a probabilistic engine that mirrors human patterns, including our propensity for error. Consequently, I excised the anecdote entirely. This process—using AI, identifying its failure, and manually correcting the record—is precisely the kind of disclosure I advocate for.
The Writing Process: Human Alchemy vs. Algorithmic Echoes
There is a pervasive fear that AI is replacing the writer. In my own practice, I have maintained a strict boundary: I have never tasked an AI with drafting my prose. The creative friction of writing—the deliberate choice of syntax, the rhythm of a sentence, and the "breaking of grammar" that literary critic Namwali Serpell identifies as the hallmark of great writing—is inherently human.
In my workflow, I rely on the standard suite of tools built into Google Docs. While I do not use dedicated AI writing assistants, I am subjected to the predictive nature of autocomplete. Often, these tools are useful for mundane tasks, such as correcting the spelling of "chagrin." Yet, they remain secondary to the creative process. As technology writer Will Oremus suggests, generative AI is best understood not as an author, but as a "slightly smarter thesaurus."
When I find myself stuck, I use ChatGPT for precise, low-stakes linguistic queries: "What is another word for a ‘small group’?" or "What is a word for ‘negligence’ starting with the letter M?" By using AI as a tool for vocabulary expansion rather than creative synthesis, I retain full ownership of the editorial voice.
The Social Media Dilemma: Why ‘Cringe’ is a Human Experience
A recurring pain point for modern journalists is the administrative burden of social media promotion. Following my previous column regarding the First Amendment rights of AI, I attempted to offload the task of drafting a LinkedIn post to ChatGPT.
The output was, to put it plainly, disastrous. The platform generated a post so jarringly artificial—filled with buzzwords and corporate platitudes—that I discarded the entire draft. The experience highlighted a fundamental truth: AI lacks the social nuance required to engage a professional community. The "cringe" factor of the AI-generated text was a stark reminder that human connection is built on shared experience and authentic tone, neither of which can be synthesized by an algorithm. I ultimately chose to write the post myself, preferring the risk of human error over the certainty of robotic vacuity.
A Taxonomy of Disclosure: Practical Guidelines for Newsrooms
Transparency does not require every article to be accompanied by an academic essay. However, newsrooms must develop clear, accessible standards for AI disclosure. These could include:
- Detailed Methodology: Clearly naming the tools used (e.g., ChatGPT, Gemini, Claude).
- Prompt Transparency: Disclosing the nature of the prompts used to influence the output.
- Human Verification: Explicitly stating that a human editor reviewed the AI’s contribution for accuracy and bias.
- Visual Indicators: Utilizing standardized icons or badges that, when clicked, link to a comprehensive AI policy page for the publication.
By creating these guardrails, journalists can account for how information has been sourced and synthesized, allowing readers to judge the credibility of the reporting.
Implications: The Ethical Ocean
We must remain clear-eyed about the broader impact of AI. Its development is often rooted in the mass extraction of intellectual property, and its operational requirements carry significant environmental and labor costs. To treat AI as a benign utility is to ignore its foundational ethical problems.
Transparency is not a panacea for these concerns, but it is a necessary mitigation strategy. If we are to use these tools, we must do so with a posture of humility and accountability. We must acknowledge that an algorithm can never replace the "alchemy" of human editorial exchange.
I was reminded of this during a recent conversation with Betsy Morais, the editor in chief of CJR. I noted that I write with a thesaurus; she replied that she edits with a dictionary. That simple exchange illustrates the fundamental difference between the roles of the writer and the editor—a relationship built on mutual challenge and nuance.
This human-to-human connection is the lifeblood of journalism. No matter how sophisticated an LLM becomes, it cannot replicate the lived experience of an editor who understands not just the definition of a word, but the intent behind it. As we move forward, the most valuable tool in any newsroom will not be the latest AI model, but the critical judgment of the journalists who know when to use—and when to ignore—the machine.
This piece was produced with support from the Craig Newmark Center for Journalism Ethics and Security. Anika Collier Navaroli is an award-winning writer, lawyer, and researcher focused on the intersections of technology, journalism, and democracy.
