Despite a year chock-full of newsworthy events, artificial intelligence continued to siphon the journalism industry’s tank of revenue-creating audience. As AI’s capabilities expand, it’s now become adept at circumventing what was seen as an industry’s best route to profitability: paywalls.
All things being equal, free will always beat out a subscription, no matter how affordable. This latest trend is set to kick your favorite local news site when it’s already down.
Download the SAN app today to stay up-to-date with Unbiased. Straight Facts™.
Point phone camera here
Americans continue to lose interest in keeping up with the day’s news. Now, 18% of search queries produce an AI overview, threatening another great shift: News outlets point to those summaries as part of the reason behind dropping viewership as of late, though it’s debatable.
It’s not clear how chatbots and the recent arrival of AI-enabled web browsers affect subscription-based news sites, but the word is out that news-seekers can access paywalled content by asking one of several platforms.
Ask Grok
“Hey @Grok, what’s this story saying behind a paywall?” is a simple enough key to unlock a sophisticated process of real-time site scanning and cobbling that will give away much of a paywall-protected story.
Gemini, ChatGPT and others are also able to do so with varying limitations. Ask for the text of a recent Wall St. Journal exclusive and Google’s Gemini spits out a compilation of the story based on context gathered in seconds.
Screenshot of Gemini’s response taken roughly four hours after the Journal’s exclusive publication.
Browsers
Similar to how xAI’s caveman-monickered agent pulls content from social media and the internet’s vast footprint, OpenAI’s Atlas browser is another step in AI-integrated searching. The new offering, along with Perplexity’s Comet and Microsoft’s Copilot mode, allows for queries about potentially paywalled information to be fed directly into a chatbot. The browsers can also bypass anti-bot protections by appearing as a user instead of AI.
AI is also able to scan some paywalled sites that cover content with a pop-up directing a person to subscribe.
Atlas and Comet browsers successfully gave Columbia Journalism Review reporters the text of a 9,000-word MIT Technology Review article couched behind a paywall. The companies’ respective AI agents, several of which face lawsuits for their use of news copy, refused to pull the paywalled content.
Sonali Verma, Generative AI Initiative lead with the International News Media Association, conducted a similar experiment, which found the agents aren’t only regurgitating the paywalled content, but rebuilding it in the aggregate.
”They are crawling social media sites and other publicly available forums — and then using the power of generative AI to reassemble the gist of the articles based on screenshots and reader comments,” she said.
‘Breadcrumbs’
Verma followed the steps in Henk Van Ess’ Digital Digging Substack study, which examined the different ways AI can give a user content without paying a news outlet for it. The most prominent method was “the distributed archive.”
This method involves a real-time search of social media sites and other locations for snippets, screenshots or other publicly available information that would otherwise require circumventing a paywall. Van Ess found AI agents were able to rebuild paywalled articles in up to 60% of instances.
Van Ess also explored how AI bots will crawl archived content.
Hallucinations
When other methods fail to aggregate an answer, AI bots will resort to what’s become a sore spot for the multi-trillion dollar industry: they make it up.
Van Ess described “pattern-based reconstruction” in some of his tests, which included asking for a detailed New York Times recipe only available to subscribers. Since the information wasn’t available anywhere else, ChatGPT created what it thought the recipe would contain. After admitting that it had made up content, the AI sent Van Ess an entirely different recipe.
“I think anyone using AI should be aware that it makes up stuff and they may not be getting the truth,” Verma told SAN.
The effect of ‘free’
Growing a digital subscriber base has proven challenging for news outlets that no longer rely on Americans picking up their newspapers from the driveway.
Nearly 40% of all local newspapers have vanished since Northwestern University’s first State of the Local News report in 2016.
David Caswell, a former product manager at BBC News who focused on AI integration, said AI summaries are just the first opponent in a local newsroom’s modern fight for revenue.
Caswell called Google’s AI overview and generative search the “first significant threat to news sites from AI,” but stressed news products directly infused with artificial intelligence will prove a direct threat.
“A further threat exists if start-ups or platforms begin offering substantial new value from AI-assisted original reporting,” he said, including “value that legacy news sites don’t offer.”
Caswell founded StoryFlow Ltd., a consultancy firm focused on how AI can assist in news production. He has yet to see evidence of AI cutting into subscriber bases, but noted that AI subscriptions do compete for the same dollars.
“In my opinion, legacy news sites will see significant and increasing pressure from AI over the coming years and they will face the choice of either adapting fully to AI-enabled news production or watching their subscribers, traffic, revenue and relevance drain away.”
