What NOT to Do with AI

I'm one of the biggest AI advocates you'll meet, but even I would never hit 'publish' on unreviewed AI content. The technology is powerful and useful, but it's not reliable enough for unsupervised publishing. Think of it as a draft machine, not a publishing machine.

Every major AI publishing failure shares the same root cause: skipping human review. The pressure to publish faster is real, but the cost of publishing wrong is always higher. No efficiency gain justifies putting your newsroom's credibility at risk.

Our readers don't care how fast we published — they care that we got it right. One viral correction does more damage to audience trust than a dozen delayed stories. The review step isn't a bottleneck; it's our quality guarantee.

This is the one area where I'm as cautious as Edmund. Innovation cannot come at the cost of source safety. Use local LLMs for sensitive work, anonymize everything, and when in doubt, don't use AI at all. Our sources' trust is more valuable than any efficiency gain.

Source protection isn't a guideline — it's a covenant. Journalists have gone to jail to protect sources. The idea that we'd casually enter their information into a corporate AI tool is unconscionable. This must be an absolute red line for every newsroom.

Our sources are real people who trusted us with sensitive information. If word gets out that a newsroom leaked source details through AI tools, no source will ever trust that organization again. The reputational damage would be permanent.

AI is brilliant at helping you figure out WHAT to fact-check and WHERE to look. But the actual checking? That's still a human job. Use AI to build your verification roadmap, then walk the road yourself.

The moment we outsource fact-checking to AI, we've abandoned our most fundamental responsibility. AI can help us organize our verification process, but every fact must be checked against primary sources by a human journalist. There are no shortcuts to truth.

I've seen how quickly a single factual error can go viral on social media. Our readers are our fact-checkers too — they will find our mistakes. Using AI to plan better fact-checking processes actually makes us more accurate, but only if we do the actual checking ourselves.

AI image generation is incredible technology with legitimate creative uses, but it has no place in news photography. For illustrations, opinion pieces, or feature content, use it with clear labeling. For news coverage, stick to real photographs. Always.

Photojournalism's power lies in its truthfulness — the photograph as witness. AI-generated images destroy that covenant. I don't care how good the technology gets; a fabricated image is a fabricated image. Label everything, and never use AI images where readers expect documentation of reality.

Our audience's ability to distinguish real from fake is already under assault. As a news organization, we have a responsibility to be part of the solution, not the problem. Clear labeling of AI images isn't just ethical — it's how we maintain our credibility in a post-truth landscape.

AI bias isn't a bug to be fixed — it's a feature to be managed. Every AI tool carries the biases of its training data, and pretending otherwise is dangerous. I actively look for bias in every AI output and use it as a signal to dig deeper.

Bias has always been journalism's greatest challenge, and AI introduces new dimensions of it. We must be even more vigilant about representation and fairness when using AI tools, because algorithmic bias can be harder to detect than human bias.

Our audience is diverse, and they notice when our coverage isn't. If AI tools are pushing us toward biased framing, we're failing the very communities we're supposed to serve. Bias checking should be as routine as spell checking.

AI should amplify your instincts, not replace them. I've learned to use AI as a second opinion — when it disagrees with my gut feeling, that's often a signal to investigate further, not to override my judgment. Your nose for news is irreplaceable.

Journalistic instinct is built over years of practice, thousands of interviews, and hard-won experience. No algorithm can replicate the feeling that something doesn't add up, or the instinct to ask one more question. Protect these skills fiercely.

The journalists our audience trusts most are the ones with strong voices and sharp instincts. AI can help us work faster, but our readers come to us for human insight, context, and judgment. Those are the things that make us indispensable.

I love AI tools, but I never trust them blindly. The most dangerous AI errors are the ones that sound completely plausible. I've trained myself to be especially skeptical of AI output that sounds too perfect — that's often when it's making things up.

Accuracy is not optional in journalism. AI tools fail the accuracy test frequently enough that no responsible journalist should treat their output as reliable without verification. Every number, every name, every claim must be checked. This is basic journalism, AI or not.

Our readers hold us to a high standard of accuracy, and they should. When we publish an error that came from AI, our audience doesn't blame the AI — they blame us. The responsibility for accuracy always rests with the journalist, not the tool.

I know training feels slow when you're excited to dive in, but I've seen the difference between trained and untrained AI users. Trained journalists get better results, catch more errors, and find more creative applications. The investment pays for itself within weeks.

We don't let reporters cover beats they haven't been trained for. Why would we let them use AI tools without training? Proper education in AI capabilities, limitations, and ethical guidelines is a prerequisite for responsible use, not an optional extra.

The newsrooms getting the most value from AI are the ones that invested in training first. It's not about being slow — it's about being smart. Well-trained teams produce better content faster and with fewer errors. That's a win for our audience.

This should go without saying, but I'll say it anyway: fabricating quotes is the line you never cross, with or without AI. AI makes it technically easier to create convincing fake quotes, which makes our ethical responsibility even greater.

Fabricating quotes is the cardinal sin of journalism. AI does not change this. If anything, the ease with which AI can generate convincing quotes makes this an even more critical standard to enforce. Any journalist who uses AI to manufacture quotes should face the same consequences as one who fabricates them manually.

Our relationship with our audience is built on the promise that when we put words in quotation marks, someone actually said them. Breaking that promise — whether through laziness, deadline pressure, or AI convenience — is an unforgivable betrayal of reader trust.

Copyright law is struggling to keep up with AI technology, and that uncertainty is a risk we need to manage proactively. I recommend treating AI output as raw material that must be substantially transformed by human creativity before it's publishable.

The copyright questions around AI are some of the most consequential legal issues facing journalism today. Until the law is settled, err on the side of caution. Human authorship, substantial transformation, and legal consultation should be standard practice.

Our content is our product, and we need to be sure we actually own it. If there's any question about whether AI-generated elements in our work create copyright exposure, that's a question we need answered before publication, not after a lawsuit.