The Atlantic CEO Nick Thompson on subscriptions, Google Zero, rage bait, and journalism's AI dilemma

The Atlantic's Nick Thompson on subscriptions, print, and journalism's AI dilemma

Nick Thompson runs The Atlantic as a subscription business, and the logic is straightforward: journalists do great work, readers develop loyalty, and loyalty converts to paid subscriptions. With between 1.5 and 1.6 million subscribers, roughly half of whom receive print, the model is holding even as the digital advertising economy deteriorates around it.

Google Zero

Thompson pushes back on the fatalism embedded in the "Google Zero" framing. Google traffic to The Atlantic is declining, but subscription conversions from Google are actually up year over year. The readers being lost are low-intent visitors who arrived via generic queries and never subscribed anyway. The readers who do subscribe arrive with a prior relationship with the brand. Thompson's target isn't zero Google dependency — it's what he calls "Google one," meaning someone can still search "how to subscribe to The Atlantic" and find it. He estimates The Atlantic is heading toward something like 25% of its former Google traffic, with subscriptions intact.

Print as a moat

Print now runs 12 issues a year, up from 10, and Thompson would push it further if editorial bandwidth allowed. His case isn't nostalgia — it's disintermediation. Print arrives via the US Postal Service, which has no algorithm to game and no incentive to cut off The Atlantic's distribution. Every other channel, from Google to social platforms, carries the risk that a policy change or algorithmic shift severs the audience relationship overnight. Print doesn't.

“We are getting less traffic from Google. But the people we are losing are not the most loyal people... The number of subscriptions that we get from people coming from Google is up year over year... We have a very firm policy that we won't [use AI to write]. As do the New York Times, as do other places.”

Talent and the Substack pipeline

The Atlantic has experimented with structured Substack acquisition, running a program that brought in six writers and offered a hybrid model combining institutional editing with upside on personal subscriptions. The program didn't work. The current approach is simpler: identify strong writers on Substack, bring them into the Atlantic's core editorial structure, and accept that they'll write less but with more depth. Thompson draws a clean distinction between writers who thrive on high-volume, direct-audience formats and those who produce better work with editors, fact-checkers, and institutional backing. The Atlantic is recruiting the latter.

The maximum time on record for a single piece is roughly a year, for reporter Caitlin Dickerson. Thompson doesn't want that ceiling to rise. His ideal is writers who file regular features and also have something authoritative to say when a story breaks in their domain.

Video

Thompson's read on the economics of journalism video is blunt. Short video performs on social platforms, not media websites, and monetizing it on social is difficult. Long-form editorial video, the kind that would align with The Atlantic's voice, is too expensive to produce against achievable advertising revenue and doesn't obviously support a subscription paywall. He points to the New York Times as a cautionary data point: the video investment is building brand and audience on vertical platforms, but the economics weren't clearly working as of its last earnings call.

The Atlantic is moving into short-form and vodcasts carefully, without making a large bet until Thompson has an economic model he trusts. He's watching what's happening to 60 Minutes and tracking the Times closely.

One unconventional idea raised in the conversation: YouTube creators who spend an hour reading Atlantic pieces aloud, adding context and commentary, are already building audiences on top of Atlantic journalism. Thompson hasn't built a formal affiliate structure around them, citing caution about anything that looks like paying influencers to promote editorial content. He leaves open the possibility that a business model could exist there if the audience behavior scales.

Rage bait and AI

Thompson is unambiguous on rage bait: good for short-term traffic, bad as a long-run strategy. His more interesting bet is that AI-mediated media consumption may reduce the incentive for it. Social media algorithms pushed publishers toward extremes because engagement rewarded outrage. AI assistants, in his view, tend toward the moderate and consensual rather than the provocative. If AI displaces social as the primary discovery channel, the commercial logic for rage bait weakens.

On AI in the newsroom, The Atlantic's policy is firm: AI can be used for editing, research, and thinking, but not for writing copy. Thompson calls not using AI for research and editing "crazy." He's running nine agents in the background of day-to-day work. The enforcement concern he raises is real — other publications have accepted freelance pieces from fake AI-generated personas — and The Atlantic invests in verifying that contributors are real people who are not using AI to write.

Sentiment on AI

Thompson's read on public AI sentiment follows something like a damped sine curve: early excitement, then backlash over hallucinations and job displacement fears, a brief period of normalization as people actually used the tools, and now a new wave of backlash driven by data center opposition and the broader cultural reaction to tech companies that, as he puts it, marketed themselves as possibly capable of destroying humanity. He expects sentiment to recover again, but the current moment is negative.

The single story he most wants to commission right now: a major piece on spatial intelligence.