Scott Brodbeck, the founder of Virginia-based media company Local News Now, had wanted to launch an additional newsletter for a while. One of his sites, ARLnow, already has an automated daily afternoon newsletter that includes story headlines, excerpts, photos, and links sent to about 16,000 subscribers, “but I’ve long wanted to have a morning email with more voice,” he told me recently in a text.

Though it could expand his outlet’s reach — especially, in his words, as email becomes increasingly important “as a distribution channel with social media declining as a traffic source” — Brodbeck didn’t think creating an additional newsletter was an optimal use of reporter time in the zero-sum, resource-strapped reality of running a hyperlocal news outlet.

“As much as I would love to have a 25-person newsroom covering Northern Virginia, the reality is that we can only sustainably afford an editorial team of eight across our three sites: two reporters/editors per site, a staff [photographer], and an editor,” he said. In short, tapping a reporter to write a morning newsletter would limit ARLnow’s reporting bandwidth.

But with the exponential improvement of AI tools like GPT-4, Brodbeck saw an opportunity to have it both ways: He could generate a whole new newsletter without cutting into journalists’ reporting time. So last month, he began experimenting with a completely automated weekday morning newsletter comprising an AI-written introduction and AI summaries of human-written stories. Using tools like Zapier, Airtable, and RSS, ARLnow can create and send the newsletter without any human intervention.

“A newsletter written by a knowledgeable local journalist with an engaging writing style and good taste in what to highlight for readers in the intro would absolutely outperform the AI model in its current iteration,” Brodbeck noted in a follow-up email. But in practice, he believes AI-generated newsletters could allow other small outlets to reach more readers while conserving staff time for reporting. “The hope,” he said, “is that people find the AI-written newsletter ‘good enough’ for their local news reading purposes.”

The automation to send out the newsletter is code-free, but Brodbeck said that coding the newsletter’s HTML and CSS to work across various email clients was one of the thorniest challenges of making the AI-powered newsletter work. After attempting to use ChatGPT to generate a solution, he ultimately hired a person to troubleshoot (the machines haven’t replaced us just yet).

“The AI part was pretty easy by comparison,” he said. “It just required patient experimentation to get it to produce a pretty consistent, desired output.”

The new AI-generated email may have a voice, but Brodbeck is the first to admit that voice “is super corny.” (Case in point: “Welcome to your Terrific Tuesday, Arlington! 🎉 As we slide into April 4th, enjoy the warm embrace of this fine spring day.”)

ARLnow is experimenting with a new 100% automated, AI driven email newsletter.

The intro is written by AI, the articles are AI-written summaries of existing ARLnow stories, and the whole thing runs automatically, generating a new morning email daily with no add’l intervention. pic.twitter.com/BAmobo1QTe

— Scott Brodbeck (@scottbrodbeck) April 4, 2023

“It’s been a tedious process to try to get it to be engaging and topical but not cringe-y,” Brodbeck said. “If you want it to show personality, you really have to push it, at times, to a personality that’s not akin to Michael Scott from ‘The Office.’”

“To be honest,” he added, “I suspect most people would just skip the intro anyhow…it’s the news that people are here for.”

As for the article summaries, GPT-4 “does a good job of summarizing stories in an interesting, engaging way — somewhat like a cross between Axios and Morning Brew,” Brodbeck said.

But even working from the human-written articles, the AI can “miss factual nuances in stories,” according to the April 20 post introducing the experiment to readers. The ARLnow team cautioned readers that “the blurbs should not be entirely relied upon for decision-making purposes.”

For now, ARLnow’s AI newsletter only has around 100 subscribers. No human editor checks it before it is circulated, but Brodbeck said he’ll add a “human editing step” if, with an official rollout, the newsletter surpasses 1,000 subscribers.

Beyond newsletters

Brodbeck is already exploring ways to use the same set of tools across different media, such as to potentially develop a daily update on YouTube. On Friday, for the first time, the AI-generated newsletter also included a two-minute AI-powered audio news summary. Voiced by a personality a shade more robotic than Siri nicknamed (brace yourself) ARLy, the recap clearly summarizes the day’s stories and transitions between them with passable fluidity.

“It’s amazing what AI and no-code tools allow someone with some technical background but minimal coding skills to do,” Brodbeck said.

That said, ARLy needed some human tweaking and monitoring to ensure it stuck to the facts. Earlier last week, the audio summary automation malfunctioned and did not pass the day’s stories over to the model. So when the AI was asked for an ARLnow summary, “in the absence of stories it just made stuff up. And it sounded pretty convincing!”

Separate from its GPT-4-written newsletter, ARLnow is also experimenting with using AI to look for typos and other errors in newly published articles; categorize articles into positive, neutral and negative buckets for potential social media purposes; and drive a chatbot to help clients write sponsored articles.

The only explicit negative feedback Brodbeck has received from readers about AI-generated content so far, he said, has come not from the daily newsletter but from another AI application ARLnow uses: The evening “Thought of the Day” and “Haiku of the Day” posts generated by GPT-3 or GPT-4 APIs and inserted into ARLnow’s evening Daily Debrief website posts.

“It just seemed like a fun thing to include that an AI model should be able to generate reasonably well,” Brodbeck said. He also experimented with a Quote of the Day, but again ran into the AI’s notorious hallucination tendency — “it would often make up quotes and attribute them to well-known people.”

Readers were unimpressed by some of the AI’s daily musings. “People pointed out, correctly, that the Thought of the Day was, in many instances, nonsensical word salad (at least to our mortal intellects),” Brodbeck said. On April 13, for instance, the spiritual AI counseled, “In learning to accept oneself, one finds freedom of muted self-critique — rehoming comfort and anchoring intimacy in the imperfect magnificence of human nature.”

The Haiku of the Day, meanwhile, “has proven somewhat repetitive, even when you ask it to write something totally new,” Brodbeck added. An attempt to link the haikus to daily weather forecasts and timely holidays exacerbated that repetitive tendency. The AI seemed to gravitate toward blandly pleasant lines often mentioning rain, moonlight, and tranquility. But in one instance, “after the automated prompt produced a bunch of gibberish for unknown reasons,” Brodbeck manually instructed ChatGPT to generate something more on the nose for the April 17 edition.

Low-stakes haikus and word salad of the day aside, Brodbeck believes current AI capabilities mean the technology can be most useful to news organizations as a tool for repackaging content, while humans should still report and write original stories.

“Until the models can incorporate past articles and better write in our editorial voice,” he said, “We’re going to largely stick to summaries rather than AI-driven original reporting, even at the most basic, rewrite-this-press-release type level.”

The bigger picture

That’s a view Joe Amditis shares. Amditis, an assistant director for products and events at the Center for Cooperative Media at Montclair State University, recently released a free handbook detailing applications and considerations for AI use by local news publishers. The handbook guides prospective users through detailed, color-coded tips for crafting the most effective prompts; describes some of the technology’s strengths and limitations; and includes guidance for executing specific uses, from cleaning transcripts to creating outlines to critiquing story ideas.

Since releasing the handbook, Amditis has heard that many publishers and reporters “seem to really appreciate the possibility and potential of using automation for routine tasks,” he told me in an email. Like Brodbeck and others, he believes “AI can save time, help small newsrooms scale up their operations, and even create personalized content for their readers and listeners,” though he raised the widely held concern about “the potential loss of that unique human touch,” not to mention the questions of accuracy, reliability and a hornets’ nest of ethical concerns.

Even when instructing AI to summarize content, Amditis described similar challenges to those Brodbeck has encountered. There’s “a tendency for the summaries and bullet points to sound repetitive if you don’t create variables in your prompts that allow you to adjust the tone/style of the responses based on the type of content you’re feeding to the bot,” he said.

But “the most frustrating part of the work I’ve been doing with publishers of all sizes over the last few months is the nearly ubiquitous assumption about using AI for journalism (newsletters or otherwise) is that we’re out here just asking the bots to write original content from scratch — which is by far one of the least useful applications, in my opinion,” Amditis added.

Brodbeck agrees. “AI is “not a replacement for original local reporting,” he said. “It’s a way to take what has already been reported and repackage it so as to reach more readers.”