What Metrics Matter? Two Case Studies. And a Survey.
Is generative AI cratering traffic? Not according to nonprofit newsrooms Mongabay and the Journalist’s Resource.
Today’s post comes from Adiel, the Tow-Knight Center’s Program Director. Don’t worry (or worry, if you prefer), Gina will continue to post regularly.
Is generative AI cratering traffic? Not according to nonprofit newsrooms Mongabay and the Journalist’s Resource.
Both saw referrals from major chatbots jump last year, and overall traffic flat or even increase. Mongabay’s Google Search traffic was up. Most intriguing, the best-read stories on both sites from ChatGPT, their largest AI traffic source, included decade-old pieces.
What’s going on here? And what does this mean for web traffic in general, and more importantly, how we should think about metrics more broadly?
(Plus, we also want your feedback and input — scroll to the bottom to take our survey; we’ll share what we learn).
The data
Generative AI is changing web traffic, and for news outlets where page views have long been a key metric of success — for advertising, donor reports, etc. — the heralded AI-driven death of Google Search traffic has been seen as an existential threat. New Chartbeat data published in a Reuters Institute report today shows an average 33% drop in organic search traffic last year.
But the impact of chatbots isn’t a linear plummet to zero click-through traffic. It’s complex, and evolving, and differs widely across newsrooms. Two case studies in 2025 traffic data can give us a glimpse into how, and some theories about why.
Rhett Ayers Butler, founder and CEO of Mongabay, an international environmental newsroom that publishes in multiple languages with a primary audience outside the U.S., told me that:
Traffic was up compared to 2024, even from Google Search. Overall traffic was 45% higher in 2025.
Mongabay saw major increases in AI referral traffic, with ChatGPT bringing by far the most traffic.
Visits from Google Discover, which uses AI to curate content feeds, and Google News both roughly doubled.
AI referral traffic had the highest engagement quality of all sources.
Carmen Nobel, editor-in-chief of the Journalist’s Resource, a four-person team based at Harvard’s Shorenstein Center that publishes reporting and research guides for journalists, educators and academics, didn’t see traffic crater either. But her year-end data had some fascinating differences from Butler’s:
Overall traffic stayed flat from 2024. Google Search dropped some, but other traffic with less clear sourcing rose to fill the gap.
AI referral traffic grew from most major AI platforms, except ChatGPT, which fell in half from 2024.
Engagement time from AI was far below other sources, like Google Search, schools and other news outlets.
And for both sites, neither of which blocks AI crawlers:
Old articles received some of the highest AI referral traffic — half of the most referred stories from ChatGPT were older than two years, and both had more than decade-old stories in their top 10.
AI referrals are still a small portion of the traffic, but both suspect there’s a lot of uncategorized AI traffic leading to changes in their metrics.
Both pulled their top ten links that ChatGPT referred to in 2025. Mongabay’s list included a 2013 article on Brazilian satellite monitoring reducing Amazon deforestation. The Journalist’s Resource list included two research roundups from 2014, and a third from 2015. The most ChatGPT-referred story for Mongabay was from 2023 (about Philippine wetlands) and for Journalist’s Resource from 2018 (about school uniforms).
None had been updated since they were published.
That all runs counter to what we’ve been hearing about recency bias in AI citations. Frankly, Gina and I don’t know what to make of it, and neither do Butler or Nobel. Butler did a little more digging, finding 82% of Google Search traffic went to 2025 links, compared to just 47% of ChatGPT traffic, which underscored the contrast.
Are older articles coming up because they’re the most comprehensive, credible sources on their topics, regardless of date? Is it something about how they’re structured — like Nobel’s research roundups that synthesize multiple studies? Butler’s most-referred articles varied, including reporting on research, but also longer explainers with anecdotal leads and hard news on weather disasters, so format alone doesn’t seem to explain it. Is it because of the audience — Nobel’s readers are mostly journalists and educators?
We had those same questions in trying to understand the differences in engagement from AI referrals.
Butler’s AI engagement is what first piqued my curiosity. He wrote about this last month, calculating his own “engagement quality score” as an index of metrics, including time spent on page and scroll depth. When looked at together, that showed traffic from major chatbots outperformed other sources by 15-20%, significantly exceeding Google Search and well above social media (Substack was the only thing close).
His interpretation: “Readers who arrive via chatbots — including ChatGPT, Perplexity.ai, Google Gemini, Microsoft copilot, and Claude.ai — spend significantly more time with our articles than those coming from other platforms. They appear to be checking the source, reading closely, and staying.”
There are reports of increased traffic quality from AI, for individual metrics like time on site and subscription rate, but Butler’s score is quite comprehensive. Mongabay may be an outlier in general, though. Its 2025 traffic from all sources was up nearly 50% from 2024, including Google Search traffic, which rose 16%. That bucks basically every trend and prediction we’ve seen (congratulations Mongabay).
Its traffic also runs counter to what the Journalist’s Resource saw in 2025. Nobel doesn’t use Butler’s quality score, but both shared with me how their average engagement time for Google Search and ChatGPT stacked up this year. For Mongabay they were roughly even. For Journalist’s Resource, which has high engagement time (above 2 minutes) from many traffic sources — especially news and education sites — ChatGPT engagement time was just 22 seconds, less than half the Google Search time.
What metrics actually matter?
Without better windows into AI prompts and responses, we’re stuck theorizing. But what does seem clear is that website traffic patterns are changing, sometimes wildly, and we don’t yet know how they correlate with content, audience, and strategic decisions about AI. As information becomes increasingly individualized and intermediated by AI systems, we’re facing fundamental questions about measurement.
Butler and Nobel have been running their newsrooms for years — since 1999 for Butler and 2018 for Nobel. As nonprofits, they measure success by impact, not traffic volume. But traffic has long been the best proxy for reach — and that proxy is getting muddier. Their data raise questions beyond “what do these changes mean?”: If your mission is to unleash facts on the world, does traffic even matter? If it doesn’t, what do you measure instead? And if it does, how do you measure when AI citations don’t result in clicks, or when “direct” traffic hides AI’s influence?
The AI information age requires re-evaluating our metrics — which matter most, and how to find or create new ones for all news business models, ones that hopefully better measure the value and service news organizations seek to provide. And which ones may lead to more dollars coming in the door. It’s not that traffic is dead, but that traffic is changing, and the more we understand that change, the better our ideas will be about what can come next.
Now we want your help. These were just two newsrooms that graciously shared data. To see emerging patterns — and to spot curious and counterintuitive trends — we want to hear from more of you. How has your traffic changed due to AI? What has surprised you? Fill out our brief survey to help us understand how audience engagement with news sites is changing. We’ll let you know what we learn.
Because I like a good methodology end note, here are details on data sources:
Journalist’s Resource data came from Google Analytics. Mongabay engagement quality score is an index calculated in-house from multiple analytics platforms (more details are in Butler’s original post). Both newsrooms used Google Analytics for their list of top referred links from ChatGPT and the engagement time comparison between Google Search and ChatGPT.
That’s all. Please take the survey. It’s four questions.