If you read financial news, sports scores, weather reports, or local crime roundups online, there is a real chance some of what you read was not written by a person. It was generated by software. This has been happening for longer than most people realize, and it is expanding fast.

Automated journalism, which the industry calls “natural language generation,” has been around since at least 2014. The Associated Press started using automation software from a company called Automated Insights to produce quarterly earnings reports that year. Instead of having a reporter write up the numbers from each company’s quarterly filing, the software read the data and generated a standard news article in seconds. The AP went from publishing a few hundred earnings stories per quarter to producing thousands. From a business perspective, it was an obvious win.

The business case is clear. Certain types of news are fundamentally data-driven and follow predictable formats. A company earned X per share, compared to an expectation of Y, and the stock moved Z. That sentence structure does not require a journalist. It requires data and a template. Same goes for weather forecasts, sports box scores, real estate transaction reports, and local crime blotters. These are genuinely good use cases for automation, and news organizations have been quietly using them for years.

The problem comes when this model starts being applied to content that actually requires judgment. In 2023 and 2024, several news outlets were caught publishing AI-generated articles that contained factual errors, outdated information, and in some cases sentences that seemed to be confused hallucinations from different stories blended together. The outlets had automated production in areas where automation was not ready for it, and in several cases had not told their readers.

There is also a subtler issue around scale. When one newsroom publishes a story, it is one perspective. When software generates thousands of near-identical articles about the same underlying data and dozens of outlets publish them, it creates an illusion of independent confirmation. Multiple sources say the same thing. That must mean it is true. Except all those sources drew from the same automated pipeline and none of them actually checked anything.

PaxPoint goes into this shift in some depth. The volume of AI-assisted content in local news specifically has risen sharply as the economic model for local reporting has collapsed, and the communities affected are often the last to know their local news is being generated rather than reported.

The transparency question is real. Most outlets using automated content do not label it. Readers assume a byline means a human being researched and wrote the story. That assumption is no longer reliable.

The simple thing you can do as a reader: pay attention to bylines and datelines. Many outlets now use phrases like “this article was produced with AI assistance” or simply have no byline for automated content. If you notice either of those signals, bring extra skepticism. If a claim in a data-driven piece matters to you, find a human-reported source before acting on it.