How The Atlantic turned around: subscriptions, AI deals, and running

Executive overview

Legacy media outlets have faced accelerating decline — shrinking ad revenue, reader fragmentation, and AI companies scraping content without compensation. Nick Thompson took over The Atlantic in 2021 when it was losing $20–30 million a year and had just laid off 20% of staff.

The fix was rigorous paywall optimisation and a data-driven subscription model — not archive digitisation or affiliate revenue. Thompson ran 230+ tests on paywall variables in a single year, reached 1.46 million subscribers, and turned the organisation profitable.

The core insight: treating subscriptions as a product to be iterated relentlessly — not a fixed gate — is what drives media economics.

Why Thompson took the job

• Competitive advantage, not ability, drove the decision: he wasn't the best journalist, but was unusually skilled at building business models for serious journalism organisations
• Saw the Atlantic's journalism as genuinely strong — a necessary precondition for any turnaround
• Recognised two prior templates (the New Yorker and WIRED) where he had already executed a similar playbook

How the turnaround worked

• Early bets — archive digitisation, book affiliate revenue — generated almost nothing
• The real lever was paywall optimisation: 230+ A/B tests in a single year on offer structure, referral-source rules, and pricing
• The paywall model: if a visitor is unlikely to subscribe, show the article; if they are, make them pay — and guess propensity using referral source, story type, and behavioural signals
• Tightening the paywall hurt advertising and brand reach but drove subscription hockey-stick growth
• Leading indicators turned positive around 18 months in; public results followed by winter 2024
• Starting point: ~800,000 subscribers, losing $20–30M/year; outcome: 1.46M subscribers, profitable, with bonuses paid to all staff

The OpenAI licensing deal

• AI companies scraped Atlantic content in clear violation of terms of service and built competing products
• Thompson's view: frustration without action is not a strategy — the industry largely chose frustration
• Signed a licensing deal with OpenAI rather than defaulting to litigation; reserved legal rights and joined a separate lawsuit
• Internal backlash was significant; the journalism team ran a piece called "The Deal with the Devil" — Thompson remained unshaken
• Current licensing fees are a transitional structure, not a durable model; the real goal is a value-exchange ecosystem

What a sustainable AI model would look like

• Prorata model: payments proportional to how much a publication contributed to a given AI answer
• Subscriber bridging: Atlantic subscribers authenticate with ChatGPT and receive Atlantic-informed answers
• Exclusive platform deals: only partnering with one AI provider to create competitive differentiation
• Referral model: AI identifies high-propensity subscribers and surfaces Atlantic subscription prompts
• None of these have been implemented yet; Thompson expects one major AI company to move first

How to build leverage with AI platforms

• Block scrapers actively — robots.txt signals are ignored; Cloudflare-level bot blocking creates real friction
• Legislation in Europe and Australia could shift the balance of power
• Winning copyright lawsuits would reset expectations across the industry
• Reputational differentiation: the first AI company to treat publishers fairly gains a meaningful brand advantage

Retaining talent in the creator economy

• Substacks, podcasts, and independent platforms offer journalists higher direct income than staff roles
• The Atlantic's counter-offer: editing, fact-checking, SEO, promotion, reach, and network — benefits independent creators lack
• Flexibility matters: giving writers room for side projects (YouTube channels, etc.) reduces the pull to go independent
• Long-term framing: a strong 25-year-old may be better served building reputation at The Atlantic for a decade before going independent
• Goal is to combine Yankees-style star hiring (e.g. Sally Jenkins) with developing young talent from within

Lessons from running applied to leadership

• Thompson ran 2:43 marathons for 12 consecutive years, believing it was his ceiling
• After his father's death he broke through — running 2:38, then 2:29, then setting an American 50K record in his 40s
• The block was psychological: an unconscious ceiling set before thyroid cancer treatment
• Implication for managing people: the constraint is often internal, not external — coaches and managers should ask "what's the biggest thing you think you could accomplish?" rather than prescribing answers
• Real improvement requires genuine discomfort — not just consistent effort at a manageable pace

Personal brand and the CEO content question

• Thompson continued his daily LinkedIn video series after becoming CEO; it serves as a forced learning mechanism
• Practical business value: CMOs watch the videos and come to ad sales meetings already warm
• Condition for it working: content must be unfiltered — the moment it becomes managed communications, it loses all value
• Not universally advisable: only worthwhile if the CEO has genuine communication skill and the opportunity cost is low