Claim Your Credibility: Publish With Evidence

Responsible AI Use in Journalism and Editorial Ethics

Part 1: When AI Meets the Newsroom. Trust, Accuracy, and What's at Stake

The Sports Illustrated case became a landmark moment, not because AI was new to journalism, but because it showed exactly what happens when editorial ethics don't keep pace with editorial tools. Newsrooms had been experimenting with AI for years: the Associated Press used automation to write earnings reports as far back as 2014, the Washington Post's Heliograf tool produced thousands of short news items during the 2016 Olympics, and Bloomberg had long used AI to flag financial data anomalies for reporters. None of those experiments caused a crisis. The difference was transparency, oversight, and the presence of human editorial judgment at every stage.

The tension at the heart of this lesson is real and ongoing. AI tools can make journalists faster, help fact-checkers cover more ground, and surface patterns in data that would take weeks to find manually. But the same tools can generate confident-sounding misinformation, fabricate sources, flatten nuance, and make plagiarism nearly invisible. Understanding where AI helps and where it creates liability isn't a philosophical exercise, it's a practical skill every journalist, editor, and content professional needs right now. The principles you'll work through in this lesson are drawn directly from cases where newsrooms got it right, and cases where they didn't.

The Core Principle: AI Assists, Humans Are Accountable

The Sports Illustrated disaster extracts a principle that sounds simple but is surprisingly easy to violate under deadline pressure: when AI produces content that reaches an audience, a named human being must be accountable for its accuracy and fairness. This isn't just an ethical position, it's a practical one. Readers, sources, and subjects of stories have no mechanism to contest or correct an AI. They need a person. Editorial accountability is what separates journalism from content generation, and it's what gives a publication the right to call its work trustworthy. Every AI-assisted workflow needs a human checkpoint before publication, and that checkpoint must have real authority to change or kill the piece.

How CNET Learned the Hard Way About AI Disclosure

In late 2022 and early 2023, CNET quietly began publishing financial explainer articles written entirely by an AI system. The articles carried no byline identifying them as AI-generated. When tech publication The Verge and later Futurism exposed the practice in January 2023, CNET faced a second problem on top of the disclosure failure: many of the AI-written articles contained factual errors. A piece explaining how compound interest works included a mathematical mistake. Another article about Social Security benefits misstated the rules. These weren't edge cases, a review found errors in more than half of the AI-generated articles that had been published.

CNET's response was to add an editor's note to the articles and begin a correction process. But the damage to reader trust was significant. The publication had built its reputation on being a reliable source for consumer financial guidance. The AI articles weren't labeled, weren't consistently fact-checked, and in some cases contained errors that could have led readers to make bad financial decisions. CNET's parent company, Red Ventures, had been motivated by efficiency. AI could produce high-volume SEO content at a fraction of the cost of staff writers. The math made sense in a spreadsheet. It didn't survive contact with editorial reality.

What CNET should have done is what the AP actually does. The AP uses AI to generate first drafts of corporate earnings reports, a task that involves pulling structured data and inserting it into templated sentences. Every one of those AI-generated reports is reviewed by a journalist before publication, and the AP is transparent about the process. The output is narrow, the data is structured, and the human review is genuine. The result is thousands of accurate, timely reports that free journalists to work on stories that require actual reporting. That's a model that works. CNET's approach, broad, undisclosed, under-reviewed, is the model that fails.

A Different Industry, the Same Problem: AI in Legal Publishing

Journalism isn't the only field where AI-generated content reaching a professional audience creates accountability problems. In May 2023, a New York attorney named Steven Schwartz submitted a legal brief in federal court that cited six cases as precedent, cases that didn't exist. He had used ChatGPT to research the brief, and ChatGPT had fabricated the citations with confident, detailed descriptions of rulings that never happened. Schwartz told the court he had been unaware that ChatGPT could produce false information. The judge fined him $5,000 and referred the matter to the court's disciplinary committee. His co-counsel faced the same sanctions.

The Schwartz case matters for journalists and editors because it illustrates a specific, dangerous property of large language models: they don't know what they don't know. ChatGPT, Claude, and Gemini are trained to produce fluent, confident text. When they lack accurate information, they don't say 'I'm not sure', they often generate plausible-sounding content that feels authoritative. In a legal brief, that produced phantom court cases. In a news article, it produces phantom quotes, phantom statistics, and phantom sources. The lesson isn't 'don't use AI.' It's 'never publish an AI-generated fact without tracing it to a primary source you've personally verified.'

Comparing AI Use Policies Across Major News Organizations

Click to enlarge

A Fact-Checker's Perspective: How Full Fact Uses AI Without Losing Credibility

Full Fact is the UK's leading independent fact-checking organization, and they've been one of the most thoughtful newsroom operations in terms of integrating AI into their workflow responsibly. Rather than using AI to generate fact-check articles, Full Fact uses it to do the unglamorous work that precedes fact-checking: monitoring political speeches, identifying repeated claims across news sources, and flagging statements that have been previously checked. Their AI tools essentially act as a listening system that alerts human fact-checkers to claims worth investigating. The fact-checkers then do the actual verification, calling sources, reviewing documents, checking official statistics.

Full Fact publishes detailed documentation of how their AI tools work and what they're used for. This transparency is deliberate. Their credibility depends entirely on readers trusting their process. If a fact-check says a politician's claim is false, the politician and their supporters will scrutinize every aspect of how that conclusion was reached. Full Fact's answer has to be airtight. By using AI only for claim detection and monitoring, not for the verification judgments themselves, they've found a way to scale their operation without compromising the human reasoning that makes their verdicts defensible. This is the model journalists should study: AI handles volume, humans handle judgment.

Applying These Principles in Your Daily Workflow

Most journalists and editors using AI tools aren't trying to deceive anyone. They're trying to work faster in newsrooms with fewer resources, tighter deadlines, and growing content demands. The ethical failures covered in this lesson. Sports Illustrated, CNET, the Schwartz legal brief, didn't happen because bad people made bad choices. They happened because institutions and individuals didn't have clear, practical guidelines before they started using tools that were genuinely new. The solution isn't to avoid AI. It's to build a small number of clear habits that make AI use both faster and safer at the same time.

The most important habit is treating AI output as a first draft of your thinking, not a final draft of your work. When you ask ChatGPT or Claude to summarize a topic, identify angles for a story, or draft interview questions, you're using AI the way a good researcher uses a reference librarian, to help you find your footing faster. When you copy AI-generated text into a published article without verifying its claims, you've handed editorial responsibility to a system that has no stake in accuracy. The distinction isn't about how much AI you use. It's about where human judgment enters the process.

A practical test: before you publish anything AI has contributed to, ask yourself two questions. First, can I personally defend every factual claim in this piece if challenged? Second, would my editor, my subject, and my readers feel they were given an accurate picture of how this content was produced? If the answer to either question is no, something needs to change before the piece goes out. These questions aren't bureaucratic hurdles, they're the same questions good journalists have always asked. AI just makes it easier to accidentally skip them if you're not deliberate about the process.

Key Lessons From Part 1

1. The Sports Illustrated and CNET failures were not caused by using AI, they were caused by using AI without disclosure, without verification, and without genuine human editorial accountability at the point of publication.
2. AI tools can fabricate facts, statistics, quotes, and citations with complete fluency and zero uncertainty. This is a structural property of how large language models work, not a bug that will be fixed in the next update.
3. Major news organizations including the AP, Reuters, and the BBC have all adopted binding AI editorial policies that require human review before publication and mandate disclosure when AI contributes to published content.
4. The most defensible AI use in journalism keeps AI in the research, monitoring, and drafting phases, and keeps human judgment in the verification and publication phases. Full Fact's model is a strong template.
5. The 'Primary Source Rule' is your single most important protection: every specific fact in your published work must trace to a source you have personally verified, regardless of how the research process began.
6. Treating AI output as a first draft of your thinking rather than a final draft of your work captures the correct relationship between AI assistance and editorial responsibility.
7. Before publishing AI-assisted content, two questions protect you: Can I personally defend every factual claim? Would my readers feel accurately informed about how this content was produced?

When AI Gets the Story Wrong: Source Fabrication and the Attribution Problem

In June 2023, a New York attorney named Steven Schwartz submitted a legal brief citing six court cases to support his client's position. The cases looked legitimate, proper formatting, plausible case names, realiztic-sounding rulings. Every single one was fabricated. Schwartz had used ChatGPT to research precedents, and the AI had confidently invented cases that never existed. The judge was furious. Schwartz faced sanctions. The story became a landmark warning about AI "hallucinations", the technical term for when AI tools generate false information with total confidence. For journalists, this isn't just a cautionary tale from another profession. It is the central editorial risk of using AI in any reporting workflow.

What makes the Schwartz case instructive for newsrooms isn't that he used AI, it's that he trusted AI output without independent verification. The same failure mode applies directly to journalism. An AI tool asked to summarize a politician's voting record, pull statistics from a report, or identify a source's prior statements will sometimes produce plausible-sounding fabrications. The AI doesn't flag uncertainty the way a junior researcher might say "I'm not sure about this one." It presents invented facts in the same confident tone as accurate ones. Journalists who treat AI-generated research as a starting point and verify everything independently are protected. Those who treat it as a finished product are exposed.

The principle extracted from this case is deceptively simple: AI output is a draft, not a source. A quote attributed to a public figure by an AI tool is not a verified quote until you find the original recording, transcript, or documented publication. A statistic generated by an AI is not a verified statistic until you locate the original dataset or report. This isn't a limitation that will disappear with better AI models, even the most advanced tools today produce hallucinations at measurable rates. The workflow discipline required is identical to what good editors have always demanded from junior reporters: show me where you got this.

The Verification Workflow: How Responsible Newsrooms Are Structuring AI Use

The Associated Press was among the first major news organizations to publish explicit AI usage guidelines, doing so in 2023 after cautious internal pilots. Their framework draws a hard line that has since influenced dozens of other outlets: AI can assist with tasks like transcription, translation, data sorting, and drafting story structures, but all factual claims in published work must be verified through traditional editorial processes. AP also requires that any AI-generated or AI-assisted content be disclosed to editors, and that AI tools not be used on breaking news stories where speed might tempt reporters to skip verification steps. The reasoning is practical, not philosophical, breaking news is precisely when AI hallucination risk is highest because models lack current information.

What the AP framework reveals is that responsible AI use in journalism isn't about restricting AI, it's about sequencing it correctly within an existing editorial workflow. AI tools are genuinely useful for generating interview question frameworks, identifying angles that might have been missed, drafting headlines for A/B testing, summarizing long documents before a reporter reads them in full, or converting a press conference transcript into a rough chronology. None of these tasks require AI output to be treated as publishable fact. They're cognitive support functions, not reporting functions. The moment AI output moves from "helping me think" to "telling readers what happened," verification becomes non-negotiable.

Smaller newsrooms without AP's resources have found practical shortcuts that preserve this discipline. One regional newspaper in the United States. The Buffalo News, has reporters use AI to generate a list of claims made in any AI-assisted research, then manually check each claim against a primary source before writing. It adds roughly 20 minutes to a story's production time. Editors report that this process has caught hallucinated statistics, misattributed quotes, and at least two completely invented organizational names in AI-generated summaries. The overhead is real but manageable, and far less costly than a published correction or a credibility-damaging retraction.

Click to enlarge

A Different Industry, The Same Lesson: How Healthcare Learned to Audit AI Output

Healthcare communications teams faced a version of this problem earlier than most. When AI tools became widely available in 2022 and 2023, hospital marketing departments began using them to draft patient education materials, press releases about clinical trials, and internal communications. Several organizations discovered, some through near-misses, some through published errors, that AI tools would confidently state incorrect drug dosages, misname clinical procedures, or fabricate study citations when generating health content. The FDA and hospital legal teams responded with mandatory human review requirements for any patient-facing AI-generated content, effectively creating the same "AI as draft, not source" principle that journalism ethics demand.

The parallel for journalists is direct. Healthcare communicators learned to treat AI output the way pharmacists treat a prescription, the doctor's order triggers the process, but the pharmacist independently verifies dosage, interactions, and patient identity before anything reaches the patient. In journalism, the AI's output triggers the reporting process. It can suggest angles, surface background, organize information, and draft structure. But before any of it reaches a reader, an editor, functioning like that pharmacist, verifies the critical facts independently. The professional liability in journalism is reputation and public trust rather than physical harm, but the structural discipline is identical.

The Ghostwriting Problem: Bylines, Disclosure, and Reader Trust

In 2023, CNET, a major technology publication owned by Red Ventures, quietly published dozens of articles generated substantially by AI, under a generic byline that obscured the AI's role. When the practice was discovered by journalists at Futurism, the resulting coverage was damaging. Multiple published articles contained factual errors. CNET's editorial credibility took a public hit, and the outlet was forced to issue corrections and acknowledge that its AI content review process had been inadequate. The controversy wasn't simply that AI was used, it was that readers were not told, errors were published, and the normal editorial accountability that a named human reporter provides was absent.

The CNET situation established a clear precedent that the journalism industry has largely converged on: if AI materially contributes to a published piece, that contribution should be disclosed. "Materially" is the operative word. Using AI to fix grammar or check spelling is not typically considered a disclosure-level contribution, editors have always used tools that assist with mechanics. But using AI to research, draft, structure, or generate content that reaches readers without substantial human rewriting crosses a threshold that most major editorial standards now require be disclosed. The Society of Professional Journalists updated its guidance in 2023 to address exactly this question, and outlets from The Guardian to NPR have published their own transparency frameworks.

Practical Application: Building an AI Ethics Checklist for Your Workflow

The gap between knowing AI ethics principles and applying them under deadline pressure is where most problems occur. Journalists working on tight turnarounds don't have time to re-read policy documents before filing. The solution that experienced editors have found effective is a short, specific pre-publication checklist, not a philosophical document, but a five-minute habit. Before any story that involved AI assistance moves to final edit, the reporter answers the same four questions: Did AI generate any statistics or data I'm citing? Did AI produce or suggest any quotes I'm attributing? Did AI identify any sources I haven't independently confirmed exist? Did AI draft any sections that went into the final piece largely unchanged? A yes to any of these triggers the corresponding verification step.

This checklist approach works because it converts an abstract ethical principle, verify AI output, into a concrete, repeatable action. It also creates an audit trail that protects both the reporter and the publication. If a correction is later required, the editor can identify exactly where the verification process broke down rather than conducting a broad investigation into the reporter's entire workflow. Several newsrooms that have adopted structured AI verification checklists report that the process also improves reporter confidence in using AI tools more extensively, because the verification safety net removes the anxiety of not knowing whether an AI-generated fact is reliable.

The checklist also needs a disclosure decision built into it. Before filing, a reporter should answer: does the AI contribution in this story meet my publication's threshold for disclosure? If the answer is yes, the disclosure line should be drafted at the same time as the checklist review, not added as an afterthought by a copy editor who wasn't present for the reporting process. Normalizing disclosure as part of the filing workflow rather than treating it as an editorial exception removes the psychological friction that leads to omissions. The goal is a newsroom culture where AI disclosure feels as routine as attributing a quote to its source, not a special admission, but standard professional practice.

Click to enlarge

Key Lessons from This Section

1. AI hallucination, the confident generation of false information, is a measurable, ongoing risk in all current AI tools, not a bug that better models have eliminated.
2. The foundational rule is that AI output is a draft, not a source. Every specific fact, statistic, quote, or citation produced by AI requires independent verification against a primary source before publication.
3. The AP, SPJ, and most major news organizations have converged on the same framework: AI assists with cognitive tasks (drafting, summarizing, structuring) but does not replace editorial verification.
4. Healthcare communications teams faced nearly identical AI verification challenges earlier than journalism, their mandatory human review protocols offer a practical model for newsrooms building their own workflows.
5. The CNET AI content controversy demonstrated that undisclosed AI use that produces published errors damages editorial credibility more severely than transparent AI use that follows rigorous verification.
6. Disclosure language should be specific about what AI did and didn't do in a given story. Vague disclosures create reader confusion and undermine the trust that transparency is meant to build.
7. A short, repeatable pre-publication checklist is more effective than policy documents alone, it converts ethical principles into specific actions that work under deadline pressure.
8. Disclosure decisions should be made during the filing process, not added as an afterthought, and should be treated as routine professional practice rather than exceptional admission.

When the Story Gets Out Ahead of the Truth

In April 2023, Sports Illustrated published bylined articles later revealed to have been written by AI-generated personas, complete with fake author photos sourced from an AI image generator. The outlet's publisher, The Arena Group, initially denied it, then admitted the content had been produced by a third-party vendor. Readers had trusted those bylines. Editors had signed off without disclosure. The damage was immediate: public trust collapsed, staff were embarrassed, and the story became a textbook example of what happens when speed and cost-cutting override editorial accountability. The problem wasn't that AI was used. The problem was that no one told anyone.

The Sports Illustrated case exposed a fault line running through modern media: AI tools can produce plausible, polished content fast enough to fool editors under deadline pressure. But plausibility isn't accuracy. Polished isn't verified. And fast isn't responsible. The editorial ethics that journalism developed over a century, attribution, sourcing, transparency about methods, don't disappear because a new tool enters the workflow. They become more important, not less, because the tool can fail in ways a human writer typically wouldn't: confidently fabricating quotes, inventing citations, or generating text that sounds authoritative while being factually hollow.

What the Sports Illustrated editors needed wasn't a ban on AI. They needed a policy, a disclosure habit, and a verification step that sat between AI output and publication. Those three things are the foundation of responsible AI use in any editorial environment, and they're available to every journalist, editor, or content professional regardless of budget or technical skill.

The Associated Press offers a more constructive model. AP began experimenting with AI-assisted earnings reports as early as 2014, using Automated Insights software to generate thousands of quarterly financial summaries. But AP built the system with explicit rules: AI handled structured, data-driven content only; human editors reviewed outputs before publication; and AP publicly disclosed its use of automation in a transparency note. Readers knew. Editors were accountable. The workflow had guardrails. A decade later, AP's AI ethics guidelines remain among the most cited in the industry precisely because they treated disclosure and human oversight as non-negotiable from day one.

The principle AP demonstrated is simple: structure produces trust. When you build a clear workflow. AI drafts, human verifies, outlet discloses, you get the efficiency benefit without the credibility risk. The mistake most newsrooms make is treating AI as either completely off-limits or completely autonomous. Neither extreme serves the reader. The middle path requires deliberate process design, and that process can be built by any editor with a checklist and a clear policy document.

Verification is where that process lives or dies. AI tools like ChatGPT and Claude cannot browse the live web by default in their standard modes, and even when they can, they may hallucinate sources, misattribute quotes, or blend facts from different time periods. A journalist using AI to draft a backgrounder on a public figure must treat every factual claim in that draft as unverified, because it is. The workflow discipline is identical to how you'd treat a research intern's first draft: useful starting point, not a finished product.

How Other Industries Handle AI Disclosure, and What Journalism Can Learn

Medical publishing faced a parallel crisis when researchers began submitting AI-generated abstracts to peer-reviewed journals. The journal Nature responded in January 2023 by banning AI authorship outright, no AI tool could be listed as an author, while allowing AI-assisted writing with mandatory disclosure. The JAMA Network followed with similar policies. The reasoning was identical to journalism ethics: accountability requires a human who can stand behind the work, answer for errors, and be held responsible. AI cannot be sued for malpractice. It cannot correct the record. It cannot be interviewed by an ethics board.

The lesson for journalists is that disclosure norms are converging across professional fields. Readers, patients, and audiences are all asking the same question: did a human being take responsibility for this? The answer has to be yes, and the disclosure is how you prove it. Newsrooms that build this habit now will be ahead of whatever regulatory or industry standard eventually formalizes it.

Consider a high school journalism teacher in Ohio running a student newspaper. Her students are using ChatGPT to draft interview questions and summarize school board meeting minutes. She's not fighting the tools, she's teaching alongside them. Her rule: every AI-drafted sentence must be traced back to a primary source before it runs. Students have to highlight AI-generated text in their drafts and annotate each claim with where they verified it. The habit she's building isn't anti-AI. It's pro-verification. Her students graduate knowing both how to use AI efficiently and why the verification step is the actual job.

That teacher's approach models what mature AI integration looks like at every level of journalism, from student papers to national outlets. The tool doesn't change the ethics. The ethics shape how you use the tool. When you internalize that sequence, AI becomes genuinely useful rather than a liability waiting to surface.

Practical AI ethics in journalism isn't abstract. It's a set of habits you build into your daily workflow. Start with the verification reflex: every time AI produces a factual claim, your next move is to find the primary source. Not to check whether AI sounds right, to confirm whether the underlying fact is documented. This takes three to five minutes per claim. It's the difference between a publishable draft and a liability.

Add a disclosure template to your style guide. It doesn't need to be long. One or two sentences explaining what AI did and that facts were independently verified. Attach it to any piece where AI contributed more than minor copy editing. Your readers will respect the transparency, and your editors will appreciate the paper trail if questions arise later.

Finally, treat AI errors as editorial errors. When AI hallucinates a source and you publish it unchecked, that's not the AI's mistake on record, it's yours. Owning that accountability isn't punishing yourself; it's the mindset that keeps your verification habits sharp. The journalists who use AI most effectively are the ones who never quite trust it, who treat every output as a starting point, and who know that the credibility of their byline is the one thing no AI tool can generate for them.

1. AI tools in journalism require the same editorial accountability as any other reporting method, the journalist, not the tool, is responsible for published content.
2. The Sports Illustrated fake bylines case shows that undisclosed AI use destroys reader trust faster than any factual error, because it signals deliberate concealment.
3. The AP model, structured use, human review, public disclosure, is the industry benchmark for responsible AI integration and has held up for over a decade.
4. Verification is not optional: every factual claim in an AI-generated draft must be traced to a primary source before publication, regardless of how authoritative the AI output sounds.
5. Disclosure norms are converging across journalism, medicine, and academia, building the habit now positions you ahead of formal standards that are coming.
6. A one-page AI use policy, drafted before a mistake forces one, is the single most protective step any newsroom or freelance journalist can take right now.
7. Treating AI errors as editorial errors, your responsibility, not the tool's, is the mindset that keeps verification habits sharp and byline credibility intact.

Click to enlarge