To mark World Press Freedom Day earlier this year, the United Nations’ Regional Information Center for Western Europe published a piece entitled “Artificial intelligence and the future of journalism: Risks and rewards.” It explored how AI can be a double-edged sword. It’s simultaneously creating a sea of slop and misinformation while also providing tools that help journalists do their jobs more effectively. Two bespoke tools developed by The Washington Post show how AI can help rather than hinder journalism.
Haystacker: Custom AI that helps editors tweak stories
Back in 2024, The Washington Post published its first story built on the work of a new AI tool called Haystacker. This allows journalists to sift through large amounts of data, whether it’s video, text, or photo, and glean newsworthy trends or patterns. This is particularly helpful to the Post’s visual forensics and data journalism teams since their work requires them to analyze large amounts of data. It can also summarize video footage, which can be particularly useful in the case of long videos like public-body meetings.
In that first story, the Post used Haystacker to analyze over 700 presidential and down-ballot campaign ads that mention immigration from the first half of the year. Haystacker extracted stills from video files, transformed them into on-screen text, and labeled any objects present (e.g., American flags). At the same time, the reporters were careful to review Haystacker’s output. The key takeaway from this investigation was that nearly 20% of the ads surveyed use footage and photos "that are outdated, lack context, or are paired with voice-overs and text that do not accurately depict what is shown on the screen.”
Without Haystacker, this investigation would have been a lot more difficult. Doing the preliminary review of every ad would have taken a long time, and allowing Haystacker to do the prep work simplified and streamlined the process.
Bandito: Analyzing tons of data at speed
Unveiled in 2016, Bandito lets editors experiment with different headlines, blurbs, and photos. It’s a bit like the armory system that’s been incorporated into many video games in order to allow players to switch between roles without needing to manually reallocate their skills, attributes, or gear. Once the article goes live, Bandito can track which version gets the most clicks, and the most popular option is then prioritized on the Post’s website.
This kind of A/B testing is nothing new, but Bandito automates it. This "set and forget" capability means journalists can focus on reporting while the system optimizes reach.
The results validate the approach. At the time of its release, Eric Rich of the Post said that "Compelling presentation is really important in the digital world, and often there is more than one way to write a headline or more than one image available.”
"Bandito gives us immediate feedback from readers and helps give important stories their best chance to be read widely. Early results are very promising with some stories showing more than double the response rates.”
A/B testing can be a pain in the neck, but Bandito makes things much easier by automating the process allowing the Post’s staff to focus on other tasks.
Why these capabilities matter
Bandito and Haystack demonstrate how AI can serve as a force multiplier for journalists. Haystacker enables investigations that would otherwise require a lot of time and effort, while Bandito makes content optimization a seamless process.
Crucially, both preserve tools require careful human oversight. The article variants that Bandito deploys are crafted by humans, and Haystacker’s output is evaluated to make sure it hasn’t hallucinated.
These tools may eventually percolate through the industry. At the time of Haystacker’s debut, the Post’s chief technology officer Vineet Khosla told Axios that “I'm pretty sure this, or some variation of it, is going to make it back to the industry at large. There is no intention of keeping it just for us.”
Conclusion
The rise of AI has made journalists more important than ever. They play a vital role in helping the public see balance and nuance in a world rife with digital noise and misinformation. Haystacker and Bandito show how AI can help with this. Haystacker accelerates investigative work that would otherwise be a time sink, while Bandito quietly boosts audience engagement without demanding more from already stretched editorial teams. Crucially, both systems rely on human oversight; the Post isn’t sitting back and letting the robots do everything. Haystacker and Bandito offer a compelling model for how AI can strengthen journalism in the years ahead.
About the author,
José Mauricio Duque
