The Daily Mail's desktop click-through fell 89% the moment an AI Overview appeared above its link. AI Overviews now show on 48% of Google queries, and publisher search traffic is down ~33% YoY. The unwritten deal — you crawl us, you send us traffic — got rewritten without a vote. Your content is still being read. You're just no longer being paid for it. The obvious answer is "build our own AI." But that road has scar tissue. The Washington Post shipped Ask The Post AI; by late 2025 its standards editor's Slack leaked — the system was inventing quotes, misattributing sources, presenting commentary as the paper's own position. The cause was one missing step: citation verification — and the damage was global. So mid-tier publishers are caught between two costs: doing nothing while their distribution engine becomes their biggest competitor, and shipping a hallucinating chatbot under their own masthead. They also can't replicate the 6-to-20-engineer ML teams the FT, Bloomberg, and NYT built before the cliff. What works is two moves at once, not one. One: a conversational engine over your own archive — GraphRAG entity resolution so "Mr. Musk," "Elon Musk," and "the Tesla CEO" collapse to one node, temporal reasoning so a question like "how did this story evolve across a decade" resolves correctly, and citation enforcement so every answer links back to the article it came from. That's the retention play. Two: a parallel licensing strategy — ProRata's 50/50 attribution pool, Cloudflare Pay Per Crawl, Tollbit — to capture revenue from the AI engines you'll never own. That's the leakage play. Most consultancies pick one. Both are correct. And no, the engine doesn't cannibalize subscriptions — Ask FT measured retention lift, not page-view loss. Your archive is the asset. Stop letting it get rented for free. Save this for your next strategy meeting. Which of the two moves is your newsroom missing? #PublisherTech #ConversationalAI #GraphRAG #AIOverviews #NewsMedia