Can AI pick the best stories for my newsletter?

The real bottleneck in a local newsletter is getting the data and finding the content. Writing is the part everyone already knows how to speed up. So I'd rather point AI at choosing what goes in than at polishing sentences.

How do I give the AI my engagement data?

Export the click report: the spreadsheet with the link to each article or event and how many people clicked it. Give that to ChatGPT. Because it can see the content behind each click, it can start learning what your readers care about.

When a guest on my podcast described this, I literally wrote it down mid-conversation. My worry had always been that a click by itself doesn't mean much. Attach the content, and it does.

"My brain is not storing every single one of these and then analyzing it and comparing it to the past six weeks. That is what AI can do that no regular person can do."

I look at a click report and glance at it. This got clicked, good. Comparing every item against six weeks of history would be insanely hard even if you were paid to do it eight hours a day. That's the job to hand the model. You input the reports, explain what its job is, and let it remember.

How do operators use it day to day?

That same guest runs two local newsletters with about 13,000 subscribers combined. He bakes his Beehiiv subscriber and engagement insights into custom ChatGPT prompts. Each day he pastes in 20 to 40 candidate stories and asks the model which ones to include, based on what his subscribers engage with. He told me he writes each issue in about an hour, three sends a week per newsletter. He didn't report a measured lift from the selection step itself, so treat it as a workflow, not a proven result.

A starting prompt:

Here is my click report from the last [number] issues, with the link for each item.
Here are today's [20-40] candidate stories with links.
Based on what my readers clicked, pick the [number] best stories for this issue,
rank them, and explain each pick in one line.

Where do the candidate stories come from?

Selection only works if you have something to select from. A few ways I've seen it done:

• Facebook pages as feeds. One publisher I've talked to gets almost everything from Facebook. He hooks up the chamber of commerce, the city government, the police department and similar pages, and turns them into RSS feeds. Anything posted there arrives like an article.
• Automated lists for recurring features. We always feature a business or a restaurant, and finding one used to mean my team searching Google. We built an automation that finds the local restaurants. Now the editor just picks one that sounds good and hits rewrite, and it writes the copy in our format. Restaurants get a different format than news.
• Search tools, with care. Perplexity can find local news and events. Ask it what the news was in Round Rock last week and it'll pull some. For events it pulls a few, but it gets things wrong, sometimes surfacing items from six weeks ago. It helps. It doesn't do the job for you.

What happens if the AI picks badly?

Readers get annoyed and stop reading. AI collects a lot of dumb stuff. I've seen an AI newsletter full of random events, like a chamber of commerce event clearly aimed at 36 business owners, sent to 7,000 people. Keep doing that and people won't want to read it, because it's full of things they don't care about.

That's why the click data matters. It teaches the model what your readers actually want, instead of everything that technically happened in town. And it's why a person still makes the final call.

"Any AI, you need a human, right? You got to have a human do this stuff."

Where does selection fit in the workflow?

The stack that keeps showing up across my podcast guests: sourcing, then selection, then a template-based draft, then a human pass, then social repurposing. My order for building it:

1. Sourcing and selection first. That's where the time goes.
2. One prompt per template section, each with its own tone.
3. Generate the intro last, since it summarizes what's already in the issue.

Can AI curate a community or Slack group the same way?

Yes. One of the biggest problems with Slack groups or Discord is that if you don't go in and check, there might be great stuff you miss. Tell Claude to find the most interesting and engaging posts or comments and put them in the newsletter under something like "interesting things that came up in the Slack." Now you're pushing it in front of people who never open the app.

Make it a recurring section. The more recurring your formats are, the less you have to think each time.

Whatever the AI picks, a person still reads the issue before it sends: should you send an AI newsletter without reading it.