The value of chatting
We can all agree that writing headlines is the most horrific thing journalists have to endure. Let’s see what we can do to solve that problem!
Start by finding a news article that you’d like to have a headline generated for. It’s important that the story has been published so you can compare the headline that’s generated to the headline that was actually published.1
I’m going to use this piece about drug use in the Bronx. If you don’t have your own story, no worries – you’ll be able to click buttons to use mine instead.
The NYT article has ads and pictures and other non-content partway through. Instead of selecting the full page of text I had to paste it in a few chunks.
Prompting basics
Once you’ve found an article you’d like to use, we’ll start by simply asking the AI to write a headline for us.
Cut and paste the sample prompt below into the chat box to the right, replacing the placeholder text with the content of your story.
Write a headline for the article below:
(paste the text of the story in)
Press enter or click the send button to ask the chat bot to write the headline.
A prompt is any time you text to a chatbot and ask for a response. Take a moment to think about the chatbot’s response to your prompt. How did it do?
Reflection
My result was a little more intense than I’d have liked, but correctly noted the location and one of the major points of conflict in the story:
Bronx Drug Crisis Intensifies Amid Overdose Spike, Highlighting Citywide Dilemma Over Enforcement and Decriminalization.
At first glance this seems great. But should we run right off to publish it?
I find that using generative AI is a lot like going out to eat: even if a result is objectively mediocre, it reflexively seems great since you didn’t have to do the work. It might take some addition effort, but I know we can find some things wrong with it!
No matter how you feel about decriminalization and dramatic headlines, there’s one thing we can all agree on: this headline is way way way too long.
Taking feedback
The greatest strength of AI chatbots like ChatGPT is that they are built around having conversations, and they understand and remember context. That means you can provide feedback and ask follow-up questions, guiding the headline to be closer and closer to what you want.
Unlike many other tasks where you want AI to do the entire job without interruption, headlines are both short enough and important enough that you (probably) want to be fully involved from start to finish!
We commonly use the word “context” to mean “the subject matter and general gist of a conversation.” It actually has a special meaning when it comes to AI! We’ll talk about it more in a later chapter, but right now it’s the amount of the conversation the AI can remember. If you talk for long enough it will forget the beginning!
Exercise: Conversing
Have a conversation with the bot to fix up your headline a bit. Nudge it a few times until you get to something you’re more comfortable with!
Directions to give the chatbot might include revising the length, tone, story angle, and objectivity/emotional appeal.
- Revise the headline no more than 10 words
- Focus on the inclusion and danger to children
- Rewrite the headline in the style of the New York Times
- Rewrite the headline as sensationalist, provocative, and punchy
Consider this a tiny, tiny starting point: there are a lot of other aspects you can fool around with.
Ask to make the headline shorter 🔮
Reflection
Notice that my suggested prompts include two different types of tone changes: one asks in the style of a certain publication, while the other includes adjectives specifically describing the tone.
Instead of “sensationalist, provocative, and punchy” I could have said “in the style of the New York Post.” Would that be a more effective shorthand in communicating to the chatbot, or is a specific description more successful? You only know by experimenting!
Also, it’s important to note that I picked those words to describe the New York Post because that’s how ChatGPT described it when I asked. If you can’t quite put your finger on what makes a certain outlet sound the way it does, ChatGPT is a fantastic resource:
Although note that ChatGPT doesn’t mention the NYT’s strange relationship to prepositional phrases, so you might still be missing characteristic elements.
Conclusion
After a few rounds of edits, I got tired of making demands of the chatbot and ended up with:
Community Faces Dual Challenge: Drug Use and Public Safety
That is… awful, honestly. Far worse than the original one!
But even though I ended up in a place I didn’t like, along the way we developed a list of suggestions that will be in common for all of our headlines. A specific tone, a specific length, things to avoid or pay attention to.
In our next step we’ll use these notes to craft a single prompt that we can re-use for any story.
Footnotes
This idea is courtesy Sisi Wei of The Markup. On a panel at ONA23 she mentioned that she doesn’t use ChatGPT for headlines, but likes to plug stories in after publication and then ask for headlines, just to be able to say “no no no, we could never use that!”↩︎