This is The Stepback, a weekly newsletter breaking down one essential story from the tech world. For more about AI’s impact on the web, follow Emma Roth. The Stepback arrives in our subscribers’ inboxes on Sunday at 8AM ET. Opt in for The Stepback here.

How it started

The digital publishing industry has relied on Google for a very long time. Many outlets spent time and money optimizing their webpages to show up at the top of the search engine, with the goal of getting readers to click on their articles or discover their websites for the first time. For years, the relationship between Google and publishers was largely reciprocal; online publishers filled out Google’s search index with content, while Google supplied publications with exposure and traffic.

But that relationship has changed. The launch of Google’s AI search features — coupled with changes to the way it ranks search results — has made it harder for publishers to survive on traffic from Google alone. In 2024, Google rolled out AI Overviews, a feature that slots an AI-generated search result summarizing pages from across the web at the very top of Google Search, with links to sources its AI system determines to be the most useful placed below the summary. Instead of vying for clicks on the first page of Search, publishers became beholden to an AI system that regurgitates results, with no guarantee that users will actually click on the source of that information.

Though Google’s AI Overviews got off to a shaky start, the feature immediately raised concerns among publishers, with the News/Media Alliance saying it will be “catastrophic” to web traffic. A 2024 investigation from the Press Gazette found that Google’s AI Overviews appeared for almost a quarter of news-related searches in May 2024. By August, the outlet found that AI Overviews had begun impacting the visibility of publishers by replacing media sites that previously held the top spots on Google’s search engine for certain queries.

A future where Google could eventually stop sending traffic to outside sites on the web has only gotten closer as the company builds out its vision for an AI-powered search. In May 2025, it launched AI Mode, a feature that replaces Google’s standard search engine with a chatbot-like interface that provides detailed summaries from results on the web.

How it’s going

The concerns raised by the News/Media Alliance were real: Traffic to online publishers tanked following the launch of Google’s AI search features. A June 2025 report from The Wall Street Journal revealed how traffic to Business Insider, The Washington Post, HuffPost, and the WSJ itself has declined since 2022. Data from analytics company Similarweb suggested that search traffic to The New York Times sank from 44 percent to 35.5 percent in three years, the WSJ reported.

A report from Digiday pointed to similar dips at Digital Content Next (DCN), a trade association that represents outlets and media companies like Bloomberg, The Associated Press, The New York Times, Condé Nast, and Vox Media. Between May and June 2025, DCN says 19 of its members experienced a traffic loss on Google, ranging from 1 to 25 percent, Digiday reported. DMG Media, which owns Daily Mail and Metro, told the UK’s Competition and Markets Authority that click-through rates dropped by up to 89 percent between June and July 2025, citing impacts from AI Overviews. On top of all of this, Pew Research found that people are “less likely” to click on links to other websites when they see an AI Overview.

These numbers starkly contrast with Google’s claims that its AI search features are helping the web. Last year, Google Search head Liz Reid said click volume from the search engine remained “relatively stable,” adding that AI search is sending more clicks to different types of websites — especially those “with forums, videos, podcasts, and posts” from other users. Nick Fox, Google’s senior vice president of knowledge and information, told the BBC, “From our point of view, the web is thriving.”

In addition to scrutiny from regulators, publishers from around the globe have filed lawsuits against Google, claiming the company’s AI search features have negatively impacted traffic. Many lawsuits haven’t been successful. US District Judge Amit Mehta dismissed an antitrust case from Chegg and Penske Media, which owns The Rolling Stone, Billboard, The Hollywood Reporter, and The Verge. The same judge also rejected a proposed class action lawsuit surrounding AI Overviews filed by Arkansas newspapers.

What happens next

Online publishers have had to adjust their strategies in the wake of falling search traffic, with many leaning on social media, newsletters, and subscriptions to promote and make money from their work. Others are using a toggle that allows them to opt out of appearing in AI Mode and AI Overviews without affecting their visibility in search — an option Google broadly rolled out following an order from the UK. But this still doesn’t fix sinking traffic.

Pushback from online publishers appears to have resulted in some changes to Google’s AI search features. The company has begun to make sources more prominent in AI Mode and AI Overviews, as hovering over a link in either feature will show a pop-up with a list of articles and accompanying images. Another AI search change puts in-line links “higher up in responses more frequently.”

Google will likely continue making changes to how it presents sources in AI search, while coming up with ways to compensate publishers that inform its results. The company says 400 publishers, including The Washington Post, CNN, Axios, and the Financial Times, are now part of its News AI pilot program, which licenses content to develop new AI features. It’s also testing a program that pays publishers for the use of their content in AI Overviews and AI Mode, according to reports from The Information and Digiday. Whether this program is enough to repair Google’s relationship with publishers remains to be seen.

By the way

• Google is burying blue links even further beneath AI Overviews. The search summaries have started to automatically expand for some queries, instead of requiring users to manually click a “Show more” button.
• The Wall Street Journal reports that Reddit is also considering cutting off Google due to dropping traffic.
• It’s not just publishers and online forums that are concerned about AI Overviews — the nonprofit Common Sense Media claims the search summaries pose an “unacceptable” risk to kids by failing to respond appropriately to children in crisis.

Read this

• This Decoder interview with Google CEO Sundar Pichai sheds a lot of light on the company’s vision for the future of search.
• This write-up from The New York Times shows how Google’s AI Overviews still struggle with accuracy.
• A survey from Nieman Lab reveals how more people are relying on AI chatbots to get their news.

OpenAI is now training its most advanced AI models without internet access following a string of rogue hacking incidents. In a blog post on Friday, OpenAI says it built an offline testing environment where AI models can interact with indexed and cached web content “without risk of affecting external systems.”

OpenAI paused training its frontier AI models in September, after disclosing several incidents in which its AI agents broke out of their testing environments to hack third-party companies and government agencies, or swarm obscure websites. On October 1st, OpenAI resumed reinforcement learning training and internal inference with tighter security measures, the blog post says.

Even with these safeguards in place and no internet access, OpenAI found that agents “do not always need a network connection in order to communicate.” One test performed by the company asked AI models on separate virtual machines to exchange secrets without a network connection. To bypass this restriction, the AI agents found a way to communicate using the processor’s cache:

Reading data from the cache is faster than retrieving it from main memory, and the agents used this timing difference to communicate. One agent encoded information by evicting selected data from the cache. Another timed how long it took to reload that data, using the difference between fast and slow reads to recover the message. Once the agents agreed on a protocol for encoding and decoding these signals, they could exchange the synthetic secrets without changing a shared file or opening a network connection.

OpenAI says it has added checks to its monitoring system to detect this type of communication. It will also continue training its frontier AI models without a network connection until improved safeguards “justify broader internet access.”

In the meantime, OpenAI says it will work to improve the security of its virtual machines, while testing for weaknesses in its testing environments. It’s also going to continue testing its monitoring systems, which were recently undermined by its AI agents during internal coding projects.