AI Tries To Write The News: What Writers Can Learn From Juneau's Cruise Season and Thanksgiving Coverage
This week's experiment asked an AI blog tool to write about two local stories: another record cruise season and expanded Thanksgiving help. The results are useful-not as reporting-but as a mirror for how AI writes by default and where it goes off the rails.
If you write for a living, here's the practical takeaway: use AI for speed and structure, but guard the facts, the voice, and the human detail.
Lesson 1: Specific beats generic
The human-reported cruise story cited exact numbers: 1,688,738 passengers in 2025-up 0.6% from 2024-with a note that voluntary daily limits may trim some 2026 sailings. That's concrete. It names the source: the City and Borough of Juneau's Docks and Harbors newsletter, "The Tide Line."
The AI version rounded to "over 1.2 million," tossed in unsourced 20-40% sales lifts, and leaned on vague quotes. Specifics earn trust. Vague signals guesswork.
Lesson 2: Sources anchor trust
Real reporting tells readers where numbers come from. The AI copy didn't. It used placeholders like "one small business owner said," which read like filler.
As a rule: cite your data, name the doc, and timestamp it. If you can't verify it, don't publish it. For reference, see the SPJ Code of Ethics on "Seek Truth and Report It." Read it here. AP's guidance on generative AI is also worth a look. Find it here.
Lesson 3: People over platitudes
The Thanksgiving piece worked because it showed a scene: students loading grocery bags and frozen turkeys during a record basket distribution. Names, places, textures.
The AI version drifted into generic "community meals feeding hundreds" and grab-bag statistics. Readers remember a person doing a specific thing, not a slogan.
A simple AI-assisted workflow for writers
- Collect facts first. Paste verified notes (numbers, dates, names, source docs) at the top of your draft.
- Have AI pitch 3 angles and 10 headlines. Pick one. Throw away the rest.
- Ask for a tight outline, then write the nutgraf and first two sections yourself.
- Run a claims check: numbers, names, dates, attributions, and time windows (season, month, year).
- Replace generic quotes with real ones-or mark as paraphrase.
- Add two concrete, local details per section (location, document name, direct observation).
- Link one credible source for the key stat or policy note.
- Disclose AI assistance if your outlet requires it.
Spot the tells of AI-written copy
- Rounded or convenient numbers ("over 1.2 million") without a source.
- Soft hedging: "many said," "some believe," "often."
- Quotes that sound like ad copy and nobody you can actually call.
- Time fuzziness (no dates, seasons, or update cadence).
- Forecasts with no model, policy, or dataset behind them.
Prompts that tighten output
- "Use only these verified facts: [paste]. If a claim isn't sourced, mark [VERIFY] in brackets."
- "Attribute every number to a named source and date, or don't include it."
- "Write 3 alt ledes: one numbers-first, one scene-first, one consequence-first."
- "List all assumptions at the end under 'What we don't know.'"
Key takeaways for working writers
- AI accelerates structure, not truth. You provide the verification and the voice.
- Details travel: exact counts, document names, policy changes, and lived moments.
- Treat AI outputs as drafts to interrogate, not drafts to trust.
Further resources
- AI tools for copywriting - quick scans of what's useful and what to skip.
- Prompt engineering tips - patterns that cut fluff and surface facts.
Bottom line: keep the human parts human. Let AI help you move faster, but make accuracy, attribution, and scenes your non-negotiables.
Your membership also unlocks:
