Read It Before You Air It: The One Rule for AI in the Newsroom

Every newsroom I train asks the same question, usually within the first ten minutes: "What's your policy on AI?"

They're expecting a document. A framework. Fifty pages of guidelines with a flowchart. And those things have their place — your station probably needs one, and your legal department definitely thinks so.

But when I strip away everything negotiable, everything that varies by station and market and tool, one rule is left standing. It fits in six words, it covers ninety percent of the ethical ground, and it's the rule I'd tattoo on every newsroom wall if the GMs would let me:

Read it before you air it.

That's it. That's the rule. Let me show you why it's not as obvious as it sounds — and why it carries more weight than most of the policies I've read.

The rule we already had

Here's what I want you to notice first: this isn't a new rule. It's the oldest rule we have, wearing new clothes.

Before AI, no responsible newsroom aired copy nobody had read. Wire copy got a producer's eyes. A reporter's script got an EP's eyes. The rookie's first VOSOT got the veteran's eyes. Verification before air isn't an AI policy — it's journalism, and it predates every tool in your newsroom including the teleprompter.

What AI changes isn't the standard. It's the temptation.

When a tool hands you clean, confident, well-formatted copy in four seconds, something in your brain whispers that it's already been checked. It reads finished. It looks edited. And on deadline, "looks edited" starts to feel like "is edited."

It isn't. AI output is a first draft written by a very fast writer who wasn't in the room, doesn't know your market, and will state a mistake with the same smooth confidence as a fact. Treat it exactly the way you'd treat copy from a brand-new intern who types 400 words a minute: gratefully, and with your red pen out.

What "read it" actually means

Reading isn't skimming, so let's be specific. Before AI-assisted copy goes to air, a journalist verifies four things:

The facts survived. Every name, number, date, location, and charge matches the source material you fed the tool. Rewriting for tone should never touch substance — but should never is a promise, and journalists check promises.

The attribution survived. "Police say" still says police say. The qualifier that made a claim accurate — alleged, preliminary, according to — is still attached. A dropped qualifier isn't a style change; it's a factual error with a lawyer attached.

Nothing was invented. The output contains no detail, however small or plausible, that you didn't put in. If a sentence makes you think "huh, I didn't know that" — stop. You didn't know it because you didn't provide it, and the tool doesn't know it either.

It still sounds like your station. Tone, sensitivity, community context. The tool doesn't know what your viewers went through last month. You do.

Four checks. On a normal script, this takes under a minute — less time than you'd spend fixing wire copy by hand. That's the trade AI actually offers: it doesn't remove the journalist from the process; it moves the journalist's minute from typing to judging. That's a promotion, not a replacement.

The byline never transfers

Here's the sentence I make every workshop repeat back to me: the tool has no byline.

When a script airs, a journalist's name and a station's license stand behind it. Not the software's. If an AI-assisted script gets a name wrong on air, "the tool made a mistake" is not a correction strategy, not a legal defense, and not something you'd accept from that fast-typing intern. The person who approved the copy owns the copy. Full stop.

This is why I get uneasy when newsrooms frame AI adoption as "letting the tool write it." The tool doesn't write anything, in the sense that matters. It drafts. Writing — in the professional sense, the sense that carries responsibility — ends with a human decision to publish. Keep that verb where it belongs and most of the scary questions answer themselves.

What this rule is not

I want to be honest about the boundaries, because a six-word rule can't do everything.

"Read it before you air it" doesn't answer disclosure — whether and how your station tells viewers AI touched a script. It doesn't answer data — what happens to the copy you paste into a tool, and whether your vendor trains on it. It doesn't answer scope — which tasks your newsroom permits tools to touch at all. Those are real questions. They belong in that longer policy, and they deserve better answers than most stations currently have.

But notice something: every one of those questions is about process around the journalism. The six-word rule guards the journalism itself — the accuracy of what actually reaches the viewer. Get that right and you've protected the thing your license and your credibility actually rest on. Get it wrong and no disclosure policy will save you.

Why I built tools anyway

People sometimes find it strange that the guy preaching caution also builds AI tools. I'd argue it's the only honest order to do it in.

I built SmartRundown because I believe in the rule. Every tool in the suite is designed as a drafting partner, not a publishing pipeline — output that arrives expecting your eyes, your edits, your judgment. The tools clear the mechanical work: the reformatting, the restructuring, the sixth version of the same story. What's left is the part that was always the real job — deciding what's true, what's fair, and what your community needs to hear.

AI in the newsroom isn't a question of whether. It's already here, in your station or the one across town. The only question is whether it arrives with the standard attached.

Six words. Read it before you air it.

Everything else is commentary.