Dow Jones Factiva's Emma O'Brian on why publishers hold 'the ace card' in the AI era

Full analysis

Emma O'Brian, who runs Factiva at Dow Jones, spent an episode of The Current podcast arguing that publishers hold "the ace card" in the AI era: AI companies have scraped everything and now need a steady supply of fresh, high-quality journalism, so content licensing becomes a revenue rail as big as ads or subscriptions. Factiva's twist is a royalty model. Publishers get paid each time their content is actually used inside Factiva's AI research tools, not a flat check like the OpenAI-style bulk deals. She says Factiva has 8,300-plus sources cleared for generative AI use.

What's being decided for an operator: how much to bet that AI licensing becomes real, durable publisher money, and whether that money leaks out of the programmatic pool you trade in. This is easy to undo. Nobody has to commit anything today. The deadline, such as it is, is set by how fast AI assistants keep eating click-through traffic.

The Market Analyst. Follow who profits from the framing. Factiva's whole business depends on convincing publishers to license through Factiva instead of withholding content or suing. So "publishers hold the ace card" is exactly what O'Brian is paid to say. The contradiction is right there in her own parent company: News Corp both sued and cut deals. If the ace card were that strong, you wouldn't need lawyers. For a generalist: the vendor selling you the shovel is telling you there's gold. A royalty rail exists and is growing, but it's a rounding error against ad and subscription revenue today, and O'Brian offered no numbers to say otherwise.

The Skeptic. For this to work, per-use royalties have to actually clear real money at scale, and O'Brian gave zero dollar figures. 8,300 sources cleared, 160 countries, 33 languages: those are input metrics, not revenue. Factiva is a B2B research tool for banks and law firms, a small, high-value enterprise niche. That is not a replacement for consumer ad CPMs; it is a nice adjacent line for the handful of publishers whose archives lawyers pay to search. The Stagwell brand-safety study she leans on to argue news is safe? Unnamed, no methodology, no sample, and Stagwell profits from selling more inventory categories. Take it as marketing.

The Operator. Try to execute this Tuesday. You call your legal team to "establish an AI licensing strategy" and discover your archive rights are a mess across a decade of contributor agreements, wire deals, and photo licenses. Now try to grant "training vs. RAG vs. API vs. grounding" tiers per counterparty, per use. That is a rights-management project measured in quarters, not a term sheet. The one genuinely useful warning O'Brian gives: if your rights only cover old-style search and retrieval, your royalties shrink as usage moves to agents and MCP connections. Renegotiate rights to cover formats that don't exist yet. That is real, boring, and expensive.

The Customer / End User (the advertiser and the buy-side). If you buy media, this episode changes nothing you do this quarter. It matters at the edges. Every question an AI assistant answers without sending a click is one less publisher impression to trade. That is a slow contraction of open-web programmatic inventory, and it is already happening independent of anything Factiva does. The buy-side argument to test is O'Brian's pitch that premium, human-reported news is a safer bet than blind programmatic. Maybe. But agencies block news for reasons of headline risk, not because they think readers hate the ads. A study saying "consumers don't mind" doesn't move the trade desk that blocks the war coverage.

The CFO. Model it honestly and licensing is a margin line, not a volume business. High value per unit, tiny volume, near-zero marginal cost once the content exists. That is attractive precisely because it isn't CPM. But it does not backfill lost ad revenue if AI traffic erosion hits your consumer site, because the enterprise research market is a fraction of the size of the consumer ad market. The right frame: treat AI licensing as incremental found money on assets you already own, not as the plan that saves the P&L when clicks fall. Anyone selling it as the second thing is selling something.

Where the council splits

Two real disagreements. First, is "the ace card" leverage or wishful thinking? The Market Analyst and Skeptic say the leverage is oversold by someone paid to oversell it, and News Corp's own lawsuits prove the negotiation isn't as clean as the pitch. The Operator says there's a genuine, unsexy job underneath: get your rights in order or lose royalties as formats change. Both can be true. The strategic framing is inflated; the operational advice is sound.

Second, does licensing replace ad revenue or just sit beside it? Nobody on the council thinks it replaces it. The CFO's version is the useful one: incremental margin on owned assets, not a lifeboat.

What this actually hinges on

Whether per-use AI royalties ever clear enough dollars to matter against ads and subscriptions. O'Brian asserts they will and shows no numbers. Until a publisher discloses licensing revenue that's a real slice of the top line, this is a promising side rail, not a pillar. What to verify before acting: get the Stagwell study's methodology before you cite it to a client, and audit your own content rights before you promise anyone tiered AI permissions you may not legally hold.

Prediction: Through the end of Q2 2027 earnings, no publicly reporting news publisher (News Corp, The New York Times, Gannett) will disclose AI content-licensing revenue that reaches 10% of its total revenue, and licensing will remain a minor line beside advertising and subscriptions.

Confidence: Medium. The rail is real and growing, but the base is tiny and no operator has shown the money.

Why: The episode's entire "ace card" case rests on a vendor, Emma O'Brian of Factiva, whose business needs publishers to believe licensing is big, and she offered input metrics (8,300 sources, 160 countries) but not a single revenue figure. Factiva-style royalties come from a small B2B enterprise research market, a fraction of the consumer ad and subscription pools that fund newsrooms, so even healthy growth off a tiny base stays small in the mix. The opposite outcome, licensing crossing 10% of revenue at a major public publisher within a year, would require enterprise AI-research demand to scale faster than any disclosed deal suggests, and publishers eager to prove the thesis would already be reporting the number if it were near that line. The silence is the evidence: if the ace card were paying, someone would show the hand.

Revisit by 2027-08-15: We're right if no major public news publisher reports AI licensing at 10%-plus of total revenue by Q2 2027 earnings. We're wrong if any of them does, or breaks it out as a material standalone segment at that scale.