WaPo’s AI Strategy Revealed, Google’s AI Crackdown Begins, and Affiliate Revenues Slide...

Shout-out to my friend Mary Hogarth and the team behind the biennial Future of Magazine (FoM25) Conference in Lisbon, 29-31 Oct. The theme of this year’s event is ‘(Re)Thinking Magazines: Challenges and Opportunities’ and they have announced a call for papers (500 words including references, deadline 19th May). P.S. All submissions are double-blind and anonymised.

Today’s Long Read looks at how the Washington Post is harnessing AI in its newsroom. Phoebe Connelly, WaPo’s Senior Editor of AI Strategy & Innovation, has focused on a collaborative approach after seeing that top-down AI edicts rarely work - their most successful AI initiatives have been when they work alongside journalists.

Happy Easter! Let’s dive in…

• Google Quality Raters Now Assess Whether Content Is AI-generated

  Speaking at Search Central Live in Madrid, Google’s Senior Search Analyst and Search Relations lead John Mueller says Google is directing its quality raters to look out for pages with main content created with generative AI tools – and rate them as lowest quality. Excellent feature - contains in-depth operational SEO hacks.

• Springer Nature Launches New AI Tool to Safeguard Research

  Following on from last week’s story about the growing need for research integrity roles, Springer Nature has created an AI tool that ferrets out irrelevant scientific references and unethical submissions. They’ve also created two other in-house AI tools—Geppetto, for spotting Gen AI fake content, and SnappShot, for detecting ‘problematic’ images. Fauci’s not going to like it.

• Fixing What’s Broken: Programmatic’s Next Chapter

  A surprisingly low percentage of programmatic ad spend reaches the publisher - typically just 60%. Where does the other 40% go? Media veteran, Tom Bowman (MSN, BBC Studios, ex-IAB Board member), outlines why the digital ad supply chain is inefficient — and what needs to change. TL;DR: Curated marketplaces offer a more transparent, premium, and controlled environment for the sell and buy side.

• Everyone Is Building AI Summaries (50min video)

  Last week’s International Journalism Festival in Perugia saw the Guardian, NYT and WaPo disclose their approach to AI. Key takeaway? “Those doing it well are spending lots of time with human editors—fine-tuning takes a lot of human work”. P.S. The Guardian’s Moran shared why they scrapped their AI liveblogging—the editorial juice wasn’t worth the journalist squeeze.

• AI Is Starting To Eat Affiliate Revenue (paywalled)

  Consumers are turning to ChatGPT for product inspiration, while Google’s AI Overviews are hitting traffic. The result? Affiliate revenue is trending down fast, with affiliate publishers now targeting readers directly e.g. email and social media. Plan C? Online retailers are ensuring links pop up in chatbot search results. Publishers are following suit—story next week.

• IMPORTANT: World's First Agentic AI Browser Blows ChatGPT Away

  Ominous developments from Opera whose push into task-completing browsing (“agentic browsing”) is a signal of what’s coming. The risk? AI agents may bypass a publisher’s site entirely—extracting and acting on content in the background—undermining ad revenue, affiliate links, etc. Opera only has a 2% browser market share, but 380M monthly users across its portfolio make it a powerful testbed.

• TIME Reinvents Person of the Year with Gen AI

  TIME has built TIME AI, a platform to enhance reader engagement through personalised & interactive storytelling. It offers personalised summaries, voice interactions, and multilingual support, as well as conversational AI. Media Voices’ Peter Houston says, “Just don't ask it any rude questions unless you want to get told off.”

• You Think You’ve Got A Paywall?

  Paywalls are being bypassed by readers simply pasting the URL into AI and asking for a summary, which can be expanded with a simple prompt. If that wasn’t enough, threads are appearing on Reddit with crafty workarounds for publisher paywalls. Licensing deals with OpenAI might feel like progress but they could be a zero-sum game if they cannibalise subs.

• Superhuman AI is Around The Corner: Brace Yourselves

  Former OpenAI researcher Daniel Kokotajlo and the AI Futures Project have published “AI 2027,” predicting superhuman AI by 2027. Once full AGI is here, they argue that superintelligence will achieve years of tech progress each week— leading to domination of the global economy by 2029. Kokotajlo left OpenAI in 2024 and led the ‘Right to Warn’ open letter—he’s not happy.

---

---

• AI TOOL OF THE WEEK: Reve AI, Inc., a Palo Alto-based startup, has launched Reve Image 1.0, an advanced text-to-image generation model previously known by its codename "Halfmoon." It now tops the global rankings for image quality. I daren’t think what AI imagery will be like in two years’ time.

• DIARY DATE #1: App Promotion Summit, London | 24th April

  The App Promotion Summit brings together app marketing, growth, product, CRM and revenue leaders with a singular goal: to maximise the reach, engagement and revenue of apps. Founder of the Summit, James Cooper, used to be Head of Strategy at The Guardian—it’s a solid event.

• DIARY DATE #2: UPM Caledonian Paper Mill, Scotland | April 24th

  How much do you know about your magazine paper? The International Magazine Centre has organised a visit to the UPM Caledonian Paper Mill in Irvine, North Ayrshire, to find out. An opportunity to visit one of the world’s most celebrated countries on business expenses—nearest whisky distillery? Auchentoshan.

---

---

Long Read…The Collaborative Engine Behind The Washington Post’s AI Strategy

The Washington Post is reshaping its editorial engine—guided by its belief that AI should serve journalism, not substitute it. In a recent WAN-IFRA webinar, Phoebe Connelly, WaPo’s Senior Editor of AI Strategy and Innovation, offered a glimpse into how one of America’s leading newsrooms is leveraging the tech.

Ask The Post: Starting Small, Thinking Big

WaPo’s AI journey began quietly: a pilot chatbot, trained only on the Post’s climate coverage since 2016. The goal? To understand how AI could responsibly answer reader questions without straying from published journalism.

According to Connelly, the topic of climate provided an ideal starting point. “We needed to stress test reader intent, information retrieval and hallucination control,” Connelly explained. “Limiting the scope helped us learn fast.”

That experiment has since grown into Ask The Post—a reader-facing tool now spanning the entire newsroom’s reporting. Built to be model-agnostic (initially powered by Meta’s Llama), the platform prioritises precision. “If it can’t answer confidently, it won’t,” Connelly said.

Humans in the Loop: Critical to AI Success

Connelly’s clearest message was that AI only works when editorial and product teams collaborate deeply. At the Post, AI-generated summaries are reviewed by assigning editors and refined by the copy desk. Every piece of feedback informs ongoing prompt development.

This iterative approach has allowed the team to move swiftly—from their initial ‘A prompt’ to a third-generation version, with better attribution and accuracy, guided entirely by editorial insight.

“Great AI is about the humans using it,” said Connelly. “It’s only powerful when it’s in service of something central to your newsroom or your readers.”

AI Training That Works

Rolling out AI tools is one thing; embedding them into day-to-day journalism is another. Connelly places huge importance on AI education, and the team has approached training not as abstract seminars but as hands-on workshops tailored to each department’s needs.

In one example, the newsroom’s data and curation team collaborated on writing prompts for their regular morning traffic reports. Participants didn’t just learn what an LLM is—they wrote real prompts, tested outputs, and walked away with something they could use immediately.

The impact has been tangible: more experimentation, more idea-sharing, and what Connelly calls a “quiet groundswell” of AI usage inside the newsroom.

“There are a lot of secret AI users out there,” she said. “My job in 2025 isn’t just to promote adoption—it’s to find those people already experimenting and help share their stories.”

What’s Next: From Side Tool to Core Feature

At present, Ask The Post is accessible via its own URL and embedded in site search. The next step is deeper integration—surfacing it directly on article pages to support follow-up questions in context.

Beyond that, the Post is exploring translation workflows, smarter SEO auditing tools, and AI-generated story summaries—all co-designed with editorial teams to ensure trust and relevance.

“The real work of AI adoption happens at the intersection of product and news. Great ideas come when you bring both together from the start.”
— Phoebe Connelly, The Washington Post

Key Lessons for Publishers

1. Use AI to solve real problems.
   The most effective tools at the Post emerged from clear editorial needs.

2. Tie training to day jobs.
   AI workshops focused on real newsroom tasks saw much higher uptake and engagement.

3. Human review is non-negotiable.
   Every AI output, from summaries to prompts, is subject to editorial oversight before publication.

4. Spot and support internal experimenters.
   Some of the best ideas come from reporters experimenting quietly—find them, talk to them, and amplify their work.

5. Acknowledge internal resistance.

   Not everyone embraces AI immediately; concerns about bias, trust, and job security are real and must be addressed openly to build newsroom-wide confidence.

Final Thought

The Washington Post’s AI journey is chiefly a story of the importance of collaboration—of editors, engineers, reporters, and product thinkers coming together to ensure AI strengthens journalism at its heart. And in the process hopefully serve readers better.

---