Raphael Thys
By Dalia Hashim
As AI technology continues to grow in popularity, efficiency, and capability, journalists around the world are increasingly wondering how these new tools can assist them with their age-old profession. Transcribing interviews, personalizing homepages for individual readers, even automating article-writing itself: these are just some of the ways large newsrooms are currently using AI. With so many AI tools available, however, it can be tough for smaller news organizations to know which ones are appropriate for their needs — and what the potential benefits and drawbacks of adopting them might be.
Navigating this space successfully is particularly important for local newsrooms which work with fewer resources despite their essential role as community watchdogs and information providers. A national survey by the Associated Press revealed that while local newsrooms had a strong interest in using AI, few were currently doing so.
Against this backdrop, Partnership on AI (PAI) saw a clear need to demystify the landscape of AI tools for local newsrooms. As a first step, we are developing the AI Tools for Local Newsrooms Database. This is intended as a resource for local news organizations considering AI adoption, listing (among other information) 70 automated tools that could be useful to journalists, how much they cost, and which newsrooms are currently using them. In the future, PAI plans to expand the database and create an accompanying guidebook to walk newsrooms through the process of procuring AI tools in a responsible manner that fully considers the positive and negative implications of these technologies.
Additionally, through analysis of these tools, PAI identified three main categories of AI tools that could support local newsrooms, some of the key opportunities they provide, and some of the most prominent risks they pose. Here, we provide a high-level overview of this AI tool landscape with key takeaways for local newsrooms wondering how AI can support their organizations.
It is important to mention that AI tools are not a panacea that will cure local newsrooms of all that ails them today. Like any technology, AI tools come with their own set of benefits and drawbacks that should be considered alongside adoption. AI tools can help journalists identify important stories, automate routine tasks, and increase readership. Using them can also result in negative interactions with audiences, mishandled personal information, and even the spread of misinformation. With a greater understanding of the tools available to them, AI-adopting newsrooms will be better equipped to discuss these potential tradeoffs as they pursue their goal of informing audiences with a sustainable news operation.
The AI Tools for Local Newsrooms Database
For a local news team strapped for expertise, time, and resources, surveying the full range of available AI tools is not easy. In addition to the sheer number of tools available, it can be difficult for someone without a background in machine learning to glean exactly what an AI tool does.
Over the summer, we surveyed local news teams, established news organizations, social media platforms, and AI developers to get a better understanding of the entire ecosystem. Based on that feedback, PAI saw a need for an easily accessible resource that explained, in plain language, which AI tools could be used to support local newsrooms.
After identifying the AI tools currently being used by other news organizations, tools with similar functionality, and additional tools with local news applications, PAI categorized these tools within a sortable database — an idea that stakeholders responded to with uniform positivity. Importantly, this database is not designed to endorse one tool over another — or the use of any AI tools at all. It is intended as an overview of the wide array of available tools and does not assess their utility, functionality, or associated risks. As the product of an initial survey of the tools being used by newsrooms and ones with similar functionality, this database is also not exhaustive. PAI intends to continue expanding this resource. If we’ve missed any tools, please let us know so we can add them.
With the AI Tools for Local Newsrooms Database, local newsrooms can compare AI tools’ functionality, use-cases, cost, and more.
All of these tools can play a role in easing the burden on newsrooms when it comes to the amount of effort and time that goes into aspects of news generation, freeing up resources to work on more in-depth pieces, investigations, or overall news creation. At the same time, these tools require that we take a deeper look into the opportunities they present and the risks that need to be taken into consideration when implementing them. Below, we break down the available tools according to three main use-cases (which largely correspond to phases in the news production cycle), highlighting key opportunities and risks along the way:
- Lead Generation and Investigative Journalism
- Content Creation and Distribution
- Audience Engagement
1. Lead Generation and Investigative Tools
“Lead generation tools” are ones that provide advance notice of trends, developing stories, or witness leads on breaking news. One kind of lead generation tool flags “trending topics,” discussion subjects that are gaining momentum on social media. Such topics can start trending because users within a particular area are posting about accidents, disasters, or other dangerous situations. In such cases, trending topics can alert journalists that a story is developing and potentially enable them to be the first ones on the scene. Such tools can also identify potential sources for stories by scanning social media and public records or transcribe municipal meeting minutes and pull out important keywords.
AI tools that analyze social media trending topics or breaking news, however, often focus on nation-wide or even state-wide trends. Trending topics can mask local news stories that are important to readers interested in learning more about their city or town. If local newsrooms are not careful they could overlook their biggest value add of providing local insights in favor of covering trending stories. Additionally, any algorithmic tool that draws on large datasets (like social media posts) poses questions about consent that may go beyond the established ethical codes of journalism.
“Investigative tools” support fact-finding and data analysis, especially when trying to make sense of large datasets or a large number of documents. Using algorithmic tools, journalists can run data analysis on large datasets, uncovering patterns and otherwise hidden connections. This analysis can then be transformed into insights for in-depth and investigative stories.
In the case of the Pandora Papers, for example, reporters used machine learning to sort and cluster 2.94 terabytes of offshore financial documents, enabling a consortium of more than 600 journalists to approach the data in “more manageable groupings.” Such tools can also detect changes in websites or legislation or scan through thousands of financial transactions to detect irregularities suggesting fraud or embezzlement.
Investigative tools, while very sophisticated, are still prone to error if the data sources are not carefully interrogated. The data sources, parameters provided to the algorithm, or assumptions made by the reporter need to be interrogated to ensure that the data analysis is not biased and that the conclusions are sound. In addition, it is important that these stories are provided with appropriate context as to how the data was obtained and the methodology through which the findings were reached.
2. Content Creation and Distribution Tools
“Content creation tools” are used to simplify and automate the news-writing and reporting process — in other words, to help create content. Currently, automated reporting tools are used primarily in areas like real estate and sports news. By utilizing open datasets, these algorithms can pull data on the latest house sale or high school game and turn it into a short article ready for publication. This frees up much needed time and resources to write longer, more in-depth news analysis. For example, the AP uses Natural Language Generation to automate thousands of college basketball game previews and corporate earnings stories. Similarly, the BBC has used AI-generated writing software to write 7,000 hyper-localized news stories on British shopping trends.
When working with such tools, however, errors in data can result in much larger errors in the news articles produced. For example, a bot once misreported a soccer game result as a “humiliating defeat” with a 10–1 result, when in fact it was a 1–0 loss. As a result, it is important to consider the types of news being auto-reported and if errors can be tolerated in that reporting. Being transparent with audiences when stories have been written by a bot and ways to report any errors is an important mitigation method.
Automated reporting can also potentially lack very important context or information that gives the reader a complete picture of the story. Consequently, newsrooms may be providing audiences with incomplete or disjointed pieces of information, inadvertently resulting in spreading misinformation or letting readers fill in the gaps in reporting with their own biases. One newsroom, for example, used automated reporting to generate articles from open datasets on police reports. The result was a number of articles providing information on burglaries, gunshot sounds, or other crime reports without verifying if those calls were actually true as reported.
“Distribution tools” ensure that a single piece of content can be shared in many languages or in a variety of formats. Using AI translation tools, reporters can generate news content in multiple languages in a matter of minutes. Newsrooms can also make their reporting more accessible by automatically generating written descriptions of images and transcripts of audio and video content.
Auto-translation (though fast and potentially helpful) needs to be reviewed for context-specific translation, accuracy, and cultural sensitivity. So it’s important to keep a human-in-the-loop, specifically a native speaker, to review those translations prior to publication.
Additionally, newsrooms can use distribution tools to transform articles into tweets, audio into written text, and text into videos complete with synthetic voiceovers or background music that work natively on Instagram or Tiktok. Some news organizations have begun illustrating their articles with AI-generated art, a practice that brings in its own host of ethical considerations.
3. Audience Engagement Tools
“Audience engagement tools” are tools focused on managing audience interactions, newsletters, and comments. Such tools, for example, can help newsrooms manage and grow the comments sections that can be a significant resource drain on small newsrooms, facilitating healthy discourse by surfacing helpful engagement and removing spam.
Audience engagement tools can also collect data and give publishers a much more insightful understanding of their audience’s behaviors, interests, and preferences. In addition to organizing publishing on multiple platforms, such tools can suggest headlines and descriptions that can resonate with audiences. These strategies collectively have proven to be very effective in increasing click-through rates, readership, and subscription/advertisement rates.
Going further, these tools can even tailor content to individual news consumers’ interests, customizing their homepage or newsletter experiences to show them stories they’re likely to be interested in. Personalization tools like these can also suggest pricing schemes based on data collected on the reader, suggesting discounts when appropriate. Such tools can both promote a newsroom’s public interest goals by reaching a wider audience and bolster business sustainability.
One audience engagement tool used by the South China Morning Post was able to personalize the recommended articles section on their website. The tool was able to recommend both articles similar to the one the reader was scrolling through and articles from different sections in the newspaper. The team reported that by adding the AI tool to their site, readers were 75 percent more likely to click on a second article which significantly boosted their content consumption.
Personalized news delivery, however, relies very heavily on collecting data from individual users. It can be unclear who has access to the data, how it’s stored, or if it’s being sold to other marketing platforms. There is often limited clarity on whether the audience has provided informed consent to their data being collected and shared in that manner. Further, in order for the news personalization algorithms to work, they must make certain assumptions about audiences and their interests, assumptions that might be biased or incorrect without the newsroom knowing.
Over reliance on audience data can also provide an incomplete picture of audience behaviors or preferences. This can be detrimental to the public interest goals of a newsroom in the long run, either exposing their audience to a very limited number of news categories or creating news echo chambers. This could skew the audience’s understanding of reality, reinforcing unfounded biases instead of challenging them.
Next Steps
AI tools for newsrooms present enormous potential for the growth of local news and the better allocation of news resources to writing and reporting. However, it is important to consider both the potential and risks when considering adding AI tools to your newsroom. As we illustrate above, there can be tensions between the financial, time-saving, and efficiency benefits of some AI tools and ethical risks like inaccurate information and data misuse.
The AI Tools for Local Newsrooms Database and the landscape overview above are intended as first steps in developing a broader framework for responsible AI procurement in local news, one that helps newsrooms fully consider how to embed ethics and responsible practices into the use of AI tools. We would love your feedback and input on the database and the analysis provided. Any additional tools we’ve missed? Examples you’d like to add? Please let us know by contacting dalia@partnershiponai.org
Our next steps include launching a multistakeholder steering committee for our local news work and creating a responsible AI procurement guidebook to walk news organizations through the steps of procuring AI tools — with all its considerations. If you’d like to be part of the process, sign up to our mailing list for the latest updates.
It is important to note that the measure of success for any given AI tool will vary from one newsroom to the next and will depend primarily on what the newsroom is trying to augment or solve for using the additional technological resource. AI tools are most effective where they can automate the most mundane and repetitive tasks in a newsroom. That might be writing prompts for social media or pulling the latest sports stats from last night’s high school game. These goals will differ depending on the size of the newsroom and where AI can support it in the news creation process. PAI’s AI and local news team will have more to say in detail in our forthcoming AI procurement guidance.
This work was generously supported by the Knight Foundation.