Research in action: How newsrooms can use AI to integrate “Words That Work” into their strategy

The American Journalism Project launched its Product & AI Studio in 2023 with support from OpenAI and additional support from the Patrick J. McGovern Foundation. The studio explores how generative AI and other emerging technologies can serve local news, sharing insights along the way. In this post, we consider how news organizations can use AI to incorporate new research into their strategy and communications.

Leveraging AI’s capacity to sharpen newsroom messaging is something we explored in several ways last year: metabolizing large amounts of information to improve work products, such as analyzing volumes of donor data to inform strategy or translating articles from Spanish to English. Now we’re also developing ways in real-time to use AI for targeted, specialized needs. Today, we’re exploring how to use AI to ensure newsroom messaging for fundraising purposes is research-led and current with audience preferences.

For fast-paced newsrooms, integrating the latest cutting-edge research into organizational strategy and fundraising communications can seem like one more thing to add to an already long to-do list. Yet, as we’re finding at the American Journalism Project’s Product & AI Studio, it can also serve as an exciting and new frontier for AI applications. 

Press Forward recently released its “Words That Work” research for newsrooms and funders, which examines the language journalists use to reach audiences. The study found that Americans said they support the mission of local news, but they need clearer, more relatable messages about why it matters and how their support sustains it.

The “Words That Work” toolkits provide newsrooms with guidance on the language to use in messaging, and the Product & AI Studio saw an opportunity to turn the toolkits into more than just a messaging guide, thanks to AI. 

By combining the “Words That Work” message frames with generative AI, newsrooms can bridge proven communication research and everyday workflows. Embedding “Words That Work” principles directly into a Large Language Model’s context turns ChatGPT into a writing partner that reflects community-tested language, values and tone. This process can be referred to as “context engineering,” or simply giving the AI clear instructions and reference material up front. It also means news outlets don’t need to start from scratch each time they craft an appeal, newsletter or campaign message.

This approach allows newsrooms, including small teams with limited resources, to tap into fresh insights from new research, as we’re exploring with Press Forward’s “Words That Work” research.

Simple and effective: One newsroom’s recent experience
During its end-of-year fundraising campaign, San José Spotlight put the “Words That Work” framework into practice, showing how AI-assisted messaging can help local newsrooms quickly generate research-informed copy that stays true to their mission and voice.

Jennifer Morrow, audience engagement specialist at San José Spotlight, said in an email that integrating the toolkit into ChatGPT was simple and effective.

“I was able to follow the instructions and upload the files into our ChatGPT account within a few minutes, and from there I was ready to test out the prompts provided and start crafting messages,” she said.

By combining Press Forward’s “Words That Work” toolkit with an LLM, San José Spotlight generated fundraising messages that highlighted journalism as a shared community good, reinforcing trust and encouraging support.

“We wanted to utilize the public-good language recommendations for our Giving Tuesday campaign to emphasize the collective good and importance of our journalism as an essential public service,” said Morrow. “The output definitely delivered on that and stayed on theme. I mainly made edits to cut the copy down and [to] customize it to our newsroom’s specific coverage focus.”

Integrating research into your AI prompts: How to get started

We recommend newsrooms implement the following steps to ensure the best, research-backed outputs possible.

There are two concepts to use “Words That Work” in ChatGPT or a similar LLM to help ensure the AI-generated outputs more closely follow the guidance and framework laid out in the toolkit.

Option one: Singular chat
An easy and immediate way to utilize this prompting is in a singular chat: Open ChatGPT, attach the toolkit, and then prompt the chat box by using the playbook and prompt pack outlined below. The benefit of this approach is that it’s easy and quick to do; however, the drawback is that you will have to repeat the process for every message you write. 

Option two: LLM project or GPT
While you can query in one-off chats with the research and generate results, we suggest a second option that uses the toolkit in an LLM: setting up a “project” or “GPT” in ChatGPT specifically for the “Words That Work” research. This will establish the appropriate context for the project or GPT (i.e., context engineering), so you can revisit the chat over and over, every time you need to write fundraising messaging. In other words, it saves you time in the long run.

Below, we outline how to build a project in ChatGPT, integrating the “Words That Work” research.

Steps to create AI-assisted fundraising messages:

Step One: Prepare the needed inputs.
Gather your key reference materials: download Press Forward’s “Words That Work” toolkit(s), the “Words That Work” LLM Project Instructions PDF, and your organization’s style guide.

Step Two: Create a project so you can provide context to ground message outputs.
Using the steps outlined in the playbook & prompt pack, prompt ChatGPT with everything it needs to create better outputs.

To effectively context-engineer in ChatGPT, you need to give it detailed information and clear instructions from the start. This can include using custom instructions, sharing the toolkit PDFs, or providing multiple examples to shape its responses. The goal is to treat the AI as a collaborator. You supply the context, tone and guidance it needs to complete a task accurately, then refine prompts based on its outputs.

Step Three: Write an AI-assisted message.
After you’ve given the LLM the context needed, ask it to write the fundraising message needed. Remember, the more you give it, the better your output will be. 

Telling ChatGPT who the target audience is, whether there are any audience segments to consider, or any issues or news stories to include, will help give a more detailed output and message.

Step Four: Run a review and accuracy check.
To ensure the message output truly implements the “Words That Work” principles, you can ask the chat to “review the output and tell me how it implements the ‘Words That Work’ toolkit.” This will allow you to review the language choices that the output gave you, and consider if they work or need edits.

Step Five: Edit the AI-assisted message.
Once the AI has generated and reviewed the message, take time to refine it with a human touch. Edit for tone, clarity and alignment with your newsroom’s voice and values. Make sure the message feels authentic, accurate and emotionally resonant — something your audience would recognize as coming from you, not a machine.

Final considerations for using the toolkit in an LLM

There are a couple of things newsrooms should keep in mind when using this approach in an LLM like ChatGPT. 

First, this approach seemed to produce better results. While newsrooms can simply attach the “Words That Work” toolkit as a PDF in the ChatGPT chat window, the Product & AI Studio team found that if that’s the only step that is taken to incorporate “Words That Work” into AI-assisted messaging, the output results are not as robust, and the AI assistant sometimes “forgets” (i.e. hallucinates) key findings from the report. As Press Forward found in their own use of AI, you can instruct the GPT not to use inputs from outside the toolkit, to limit hallucinations.

No matter what process is used, it’s always important to double-check AI outputs for accuracy and context. Keeping the human in the loop helps maintain journalistic standards and ensures that research is translated into messaging that is accurate, ethical and audience-centered. When used this way, LLMs like ChatGPT can be a powerful tool to support newsroom judgment, not replace it.

The American Journalism Project is thankful to Press Forward for producing this research and for granting permission to use and adapt “Words That Work” in this AI-assisted context.

This piece was developed and authored by Maggie Cogar, a consultant working with the American Journalism Project’s Product & AI Studio and Startups Studio. Caroline Porter, a consultant, also contributed to this piece.