AI Is Reshaping Journalism and PR: Francesco Marconi on Surviving ‘Machine Consumption’
Francesco Marconi, an early AI newsroom pioneer, explains how artificial intelligence is transforming journalism and PR. He also outlines why communicators must structure content for AI systems that now read and rank information before humans.
Newsrooms are increasingly embracing large language models (LLMs) to streamline production, support reporting, and reshape how journalists work.
One pioneer who worked at major news outlets led early efforts to integrate artificial intelligence into editorial workflows, helping lay the foundation for how generative AI is used in modern journalism.
Francesco Marconi, a computational journalist who founded AppliedXL, helped The Associated Press and The Wall Street Journal bring AI into their editorial processes before these tools went mainstream.
So has AI use in the newsroom become the norm?
Recent global surveys show that a strong majority of news organizations have already experimented with generative AI tools for tasks such as transcription, summarization, data handling and drafting.
About 73% of those news organizations said AI platforms such as ChatGPT present new opportunities in the journalism industry.
But AI is not only being used as a tool in newsrooms. It is also reading content and deciding what gets surfaced online.
In fact, humans are often second in line when it comes to reading online content like press releases, Marconi said in an exclusive interview with EIN Presswire.
That’s why Marconi says communicators should structure content with the most helpful information at the top — for what he describes as the first reader: AI systems. He expands on that subject later in the interview.
Speaking at events and on podcasts, Marconi continues to share how he has helped spearhead AI-driven newsroom tools.
Some of those innovations include automated systems that generate routine, data-based stories such as earnings updates and sports results, and machine-learning models that identify trends in large datasets.
He’s also known for leading early experiments in AI‑assisted reporting.
Still, Marconi maintains that AI should augment journalists, not replace them, by handling routine tasks so reporters can focus on deeper, more investigative work.
Today he leads AppliedXL, a company built on the idea that early signals in public data can be detected long before they become news. Its clients include Bloomberg, Dow Jones, Washington Post Intelligence and more.
His career — spanning newsroom R&D labs, MIT Media Lab research, and books on AI in journalism — has made him one of the most influential voices on AI in modern journalism.
EIN Presswire caught up with Marconi to further explore how AI is reshaping the news industry, why machines are now reading content before humans, and how communicators can “survive machine consumption” by writing structured content.
Q: How did your time at The Wall Street Journal and the Associated Press shape your views on AI and the future of news?
At the AP, we were generating thousands of earnings stories a month from structured data before most newsrooms had a coherent AI strategy. Not as an experiment, as production infrastructure.
At the Journal, we took a step further to build data mining and event detection tools that monitored massive troves of data from regulatory fillings, transcripts and press releases to find signals and break stories against the competition.
What that work made clear is that the constraint in journalism has never been writing. It’s knowing what to look for, in which data, at what moment. The harder problem is encoding that judgment, the investigative instinct, the research discipline, the sourcing logic, into systems that can apply it at scale. That’s what we do at AppliedXL. We turn the insight of journalists and domain experts into AI that can run those same steps across millions of data points continuously. The outcome is detection. Signals that would take a team of researchers weeks to surface, found in seconds, before they become headlines. We’ve benchmarked this directly against the leading general-purpose AI systems. General-purpose models produce complete responses. They do not produce high-signal ones. That’s the difference between a system trained on everything and a system built to know one domain deeply.
Q: You recently wrote that AI agents are becoming the primary consumers of news. What does that mean?
Most news is still produced as if a human will read it first. Increasingly, that’s not true. For every human a bot sends to a publisher, that bot has already read the publisher’s work up to 70,000 times. The machine isn’t a distribution channel. It’s the primary audience. And it reads differently. A 2,000-word article is less accurate than a 200-word list of facts if the audience is an AI. Agents don’t read. They retrieve fragments and discard the rest. Every sentence that isn’t a clean, extractable claim is noise the machine uses to mis-rank and mis-summarize the source. Narrative flow is exactly what breaks the pipeline. Organizations that understand this early have a structural advantage. The ones that don’t will find their content readable but not usable.
Q: You’ve introduced the concept of “pre-news.” How would you explain that idea to communications professionals in simple terms?
Before a story runs, there’s a data trail. A clinical trial registry update. A regulatory filing. A carbon credit issuance. These are public records, just not yet news, because no journalist has connected them to a narrative. The gap between when that signal appears and when it becomes a headline can be hours, days, sometimes weeks.
And it goes beyond detection. We build risk scores and prediction models on top of those signals. Our clinical trial prediction model calls trial outcomes with 88% directional accuracy, higher than any peer-reviewed model in the literature. That’s what happens when you encode genuine domain expertise into the system rather than applying a general-purpose tool to specialized data.
Q: Your company works heavily with public data and signal detection. How can that help communications teams or agencies today?
Some of the most sophisticated investors and hedge funds have been doing this for years. Mining regulatory filings, tracking activity patterns in job postings, running linguistic analysis on public disclosures to find risk signals and decode what companies are actually doing versus what they’re saying. That infrastructure exists. It’s being used on your clients right now.
Between 50 and 70 percent of S&P 500 companies in healthcare, energy, utilities and other regulated sectors have material signals sitting in public data right now that their own IR and communications teams aren’t seeing. Not because the data doesn’t exist, but because accessing it requires the right infrastructure. Hedge funds built that infrastructure years ago. Most corporate communications teams are still working without it. We give them access to the same types of signals, the same patterns sophisticated market participants are already reading, so they can walk into any conversation knowing what the data already says about them.
Q: As AI systems increasingly scan press releases, filings and online content before humans do, how should communicators adapt?
Structure matters more than prose now. A material fact buried in paragraph six may never get weighted correctly by an automated system. Putting the most important signal at the top, with precision and not marketing language, has always been good practice. Now it has operational consequences.
The deeper shift is about what information needs to become. To survive machine consumption, content needs to be atomic, structured, and retrievable. Not just the facts, but the context that tells the machine what the facts mean. Ontologies. Structured claims. Instructions an agent can follow without distorting the source. Almost none of what gets published today is built that way. The communicators who figure it out first will have a significant advantage over those still optimizing for human readers who may no longer be first in line.
Q: You’ve said the “signal before the story” may become more valuable than the story itself. What does that mean for journalism and media companies?
Every era of media gets defined by what it monetizes. Print monetized distribution. Digital monetized data. Social monetized attention. The next era will monetize truth, because demand for it is about to outstrip supply by orders of magnitude.
AI-generated content went from roughly 7 percent of new web articles in 2022 to over 50 percent today. At the same time, US newsrooms have lost roughly half their journalists since 2008. The investigative desks, the regulatory beat reporters, the people whose job was to say “this is true and here is why” have been cut, consolidated, or eliminated. The supply of things to verify has exploded in three years. The workforce capable of verifying them has been shrinking for fifteen.
That asymmetry has a price. A market trading on unverified information misprices risk. A court admitting it as evidence reaches the wrong outcome. A clinical decision shaped by it harms a patient. No hiring wave closes that gap. The math doesn’t work.
Meanwhile, Google is already deciding which information surfaces. OpenAI is deciding what counts as a credible source. None of them have editors. None of them issue corrections. What’s true is becoming an engineering decision. A century of journalism built the infrastructure to verify truth. Nobody has figured out how to sell it at the speed it’s now needed. That’s the market that’s opening up.
Q: What skills do you believe journalists, PR professionals and communicators will need most in an AI-driven industry?
Data literacy, not programming, but the ability to read a database, parse a filing, know where primary sources live before they enter the news cycle. Domain depth, because generalist skills are being commoditized faster than specialist knowledge. A communications professional who genuinely understands pharmaceutical regulation or carbon markets is not interchangeable with a general-purpose tool.
It also helps to understand which kind of journalism you’re in. There are two. The first is utility journalism: information whose job is to inform decisions. For that kind, the article is increasingly the wrong container. The machine retrieves fragments and discards narrative. Structure is everything. The second is community journalism, news you feel, identity, belonging, storytelling as the product rather than the packaging. That one doesn’t need to change. The professionals who struggle most in the next few years will be the ones who haven’t figured out which of those two they’re actually doing.