Two weeks ago the story was publishers saying out loud that they might leave Google Search. This week the story is quieter and more serious: they have started doing the arithmetic. Blocking Google’s crawlers has moved from conference-panel bravado to boardroom slide, and Digiday’s reporting captures the shift precisely — the threat has entered the spreadsheet stage.
The numbers pushing it there are ugly. Publisher ad supply is down as much as 40% year over year in Q2 — fewer human sessions means fewer impressions to sell, before any pricing effect. And 15.5% of US Google news searches now show Top Stories inside an AI Overview, meaning the answer box has swallowed the shelf space publishers used to occupy even when they “won” the search result. Against that backdrop, Reuters is openly weighing the tradeoff of blocking Google’s crawlers — not leaking it, weighing it, in public.
But the same reporting carries the historical warning label. Axel Springer and a coalition of Spanish publishers tried versions of standing up to Google in 2014 — and reversed within weeks, because the first mover eats the traffic loss alone while everyone else inherits their shelf space. That collective-action problem has not gone away. What has changed is the existence of a middle path: the UK CMA route, under which publishers can opt out of Google’s AI features while keeping their search placement. One executive’s description of that option — a “poisoned chalice” — is the most honest sentence in the piece, because nobody yet has the data to model what accepting it costs.
The numbers in this piece
01Why the full block keeps not happening
The game theory is worth stating plainly, because it governs everything. A coordinated, simultaneous block by major publishers would impose a real cost on Google — its AI answers and news surfaces would visibly degrade. But an individual block imposes almost no cost on Google and a severe, immediate cost on the blocker: the traffic disappears, competitors absorb the queries, and the publisher re-enters the index weeks later having proven only that it could be lived without. 2014 ran this experiment twice and both cohorts folded. Everyone modelling a block today knows that history, which is why the modelling is the story — publishers are not preparing to jump, they are trying to find out what jumping would cost, and what a partial jump might buy.
That is also why the CMA-style partial opt-out matters more than the dramatic version. Declining AI features while keeping search placement is the only move on the board that doesn’t require a coalition. The problem is that its cost curve is genuinely unknown: does opting out of AI Overviews restore clicks that the answer box was intercepting, or does it just remove you from the one surface users increasingly see, while your competitors’ content fills the summaries? “Poisoned chalice” is an executive admitting that nobody knows — and that the first publishers through the door will generate the data everyone else uses.
02The model is the asset, not the decision
Here is the operator’s read: you are not being asked to decide anything this quarter. You are being given a window in which to build the model before someone — a board, a coalition, a licensing counterparty, Google itself — demands an answer. Publishers who have the arithmetic done will negotiate; publishers who don’t will guess.
The model needs three inputs most publishers do not currently produce. First, revenue per Google-referred session, by section — not traffic share. Sessions that bounce off an unmonetised page are not worth defending, and averages hide the fact that Google may drive 60% of sessions but a much smaller share of contribution margin. Second, the brand-term share of your Google traffic — searches for your name or your properties would partly survive any delisting as direct navigation, so they discount the true downside. Third, your subscription-to-programmatic revenue mix, because that ratio is your actual leverage. A subscription business can absorb a traffic shock and keep billing; a traffic-dependent programmatic business cannot. Knowing which you are settles most of the internal argument before it starts.
With those inputs, the scenario worth pricing is not the full block — history has already priced that at “capitulation within weeks.” It is the partial opt-out: AI features declined, search placement kept. Your model’s output should be a single number — the annual revenue at which that trade breaks even for you — because that number is also your reservation price in any licensing conversation.
Nobody can model the poisoned chalice for you — that is the whole point of this week's reporting.
03Why this matters for publishers
| The supply collapse makes this urgent, not theoretical | Ad supply down as much as 40% year over year means the status quo is already expensive. "Do nothing" is not the safe option in this model; it is one of the scenarios, and it has a cost line like the others. |
|---|---|
| The answer box has absorbed the news shelf | With 15.5% of US news searches showing Top Stories inside an AI Overview, winning the ranking no longer guarantees receiving the click. Your Google-referred traffic can decline while your rankings hold — which is exactly why traffic share is the wrong metric to model with. |
| First movers generate the data; everyone else uses it | Reuters and whoever follows through the CMA door will produce the real-world numbers on what opting out costs. Watching those outcomes closely is free; being early is not. |
| The breakeven number is negotiating collateral | When a licensing offer, a coalition invitation or a board question arrives, the publisher who can say "our partial opt-out breaks even at X" is negotiating. The one who can't is being negotiated with. |
04What publishers should do
05The bottom line
Nobody can model the poisoned chalice for you — that is the whole point of this week’s reporting. The history says lone blockers fold, the game theory says coalitions are hard, and the only unilateral move on the board is a partial opt-out whose cost nobody has measured. In that environment, the publishers with power are not the loudest ones; they are the ones holding a finished spreadsheet. Build the model now, while it commits you to nothing. The moment it’s needed — an offer, a coalition, a board meeting — is the moment it’s too late to start.