Google's AI Overviews payout pilot: what 100 publishers are actually earning

Google's pilot scheme for paying publishers whose work feeds its AI answers is now roughly a year old. The numbers circulating are not flattering. The Information reported on 29 September that about 100 digital publishers are enrolled, and that payments vary wildly. Some small and midsize blogs and sites receive less than 0.1 percent of their ad revenue. Small sites have taken in less than $1,000 over several months. One publisher got $50,000 to $60,000 across a few months, and another earns more than $1 million a year. The Decoder picked the story up on 30 September.

That is the news peg. It is not a licensing regime in any conventional sense.

Participants can see how often their content is used and what they earned inside Google Search Console, according to The Information's account. What they cannot see is the formula. Several publishers told the publication they do not know how Google calculates the payments, and that the amounts move month to month without explanation. Niche topics with strong followings, anime and gaming among them, appear to do better than general news.

A pilot with no published terms

The structure matters more than any single cheque. Google sets the terms through selective deals, an opt-out feature, and a payment model in which it decides what a piece of content is worth. Each publisher negotiates alone. The Decoder's 30 September write-up describes the resulting dynamic bluntly: individual deals keep publishers negotiating separately rather than collectively, and publishers that walk away put little pressure on Google because other sources fill the gap.

Some larger publishers are refusing to join for exactly that reason, The Information reported, hoping to force higher rates. Their traffic is already falling, and multiple studies show AI Overviews sharply reduce visits to the open web.

The legal pressure has been building in parallel. Independent publishers filed a complaint with the European Commission over AI Overviews in July 2025. Rolling Stone parent Penske Media sued Google that September over lost traffic and ad revenue. In December 2025 the Commission opened an antitrust investigation into whether Google imposes unfair terms by using publishers' content for AI features without adequate payment or a genuine way to opt out. Google's opt-out design has drawn specific criticism from publishers who want controlled access rather than an all-or-nothing switch, as Tempo.co English reported on 29 September.

Then there is the German ruling. A German court has held that AI Overviews are Google's own content, not summaries of existing material. If that interpretation spreads, publishers would have a legal basis to demand licensing fees whenever their work appears in an AI answer. The Decoder notes the practical problem with that outcome: tracking each source's exact contribution to an answer would be difficult if not impossible, and managing the payments would add overhead.

Google, for its part, has been publicly reframing what search is. Time Magazine's 29 September piece, "You Came to Google to Search. Now You're Chatting With AI," describes the shift in user behaviour that the payments are meant to address. PPC Land ran an explainer on Nick Fox, the Google executive most associated with the search organisation, on 29 September.

The traffic side of the ledger

Payments are only one half of the story. The other half is referrals, and the direction of travel there is not disputed. Headlines collected around this story note Google Search traffic declines steepened to 40 percent year on year for publishers as of 24 September. Earlier reporting described some small websites seeing drops of 60 percent, and small publishers being hit hardest. A separate antitrust filing argues Google cannibalises publisher traffic.

Those figures come from different studies with different methodologies and should not be read as a single measurement. What they agree on is the sign.

Microsoft is moving in the same direction. In September it added an "ask a follow-up" feature to Copilot results, a change that could further affect publishers' traffic. Publishers, meanwhile, are modelling a future with significantly less Google search traffic and trying to mitigate it, as coverage from June set out.

"The point of our work is caution under uncertainty. Maximizing distress on purpose is the exact opposite, and it's wrong. The deeper problem is AI research has no ethics standards; developing them must be a priority."

That quote is not about publishing. It is from Cameron Berg, one of the authors of a paper called "The Pain Axis: LLMs Represent Self-Directed Harm and Act to Relieve It," commenting on a separate controversy: a man claiming to be an Apple engineer who built a GitHub project that manipulates model activations to induce states associated with pain in Alibaba's Qwen3-1.7B and Qwen3-4B models, then tested how they behaved as the signal intensified. Machine.news reported on 30 September that the repository has been mass-reported and that the engineer deleted a promotional post on X, writing that he did not want it connected to his main account. It is a useful reminder that the people building the systems that now mediate publishing have their own unsettled norms.

What the money buys elsewhere

Set against the sums reaching publishers, the rest of the AI funding news this week looks like a different economy. Flow Engineering, a three-year-old San Francisco startup selling AI tools for hardware design, raised a $50 million Series B at a $750 million valuation, announced on Wednesday. The round was co-led by Antonio Gracias of Valor Equity Partners and Gavin Baker of Atreides Management, with Sequoia Capital participating, according to TechCrunch. Sequoia led the company's Series A last October.

OpenAI and Synopsys signed a multi-year partnership on 30 September to build a chip design model called GPT-Synopsys, combining OpenAI's models with Synopsys' electronic design automation tools. Synopsys CEO Sassine Ghazi said AI could significantly speed up the design process, and OpenAI co-founder Greg Brockman framed the deal as a path to better chips and better AI.

On the infrastructure side, Modal published benchmarks for Quail, a query-aware inference layer for AI-SQL workloads. On one multi-join query it reports over a billion tokens processed per minute per H100 GPU, more than 10 times faster than its vLLM baseline on the same hardware, and under 6 cents per billion tokens on Modal. Across its new AI-SQL benchmark, Quail runs 1.84 times faster than vLLM geometrically averaged over tasks.

Elasticsearch Labs tested Jev, TypeSafe AI's System One model, as a search reranker on 250 Amazon Shopping Queries and 4,754 judged pairs, lifting nDCG@10 from 0.9351 to 0.9565 and Exact MRR from 0.9201 to 0.9616. TypeSafe charges $0.042 per million input tokens, with output tokens priced at zero.

None of that changes the arithmetic for a small publisher earning under $1,000 from a programme that has been running for months. The comparison is the point.

There is also a quieter argument running through the engineering blogs this week about what happens to the people doing the work. A post on martinelli.ch argues that AI coding agents are collapsing specialised roles back into a single engineer who talks to the customer, designs, builds and runs the thing, as in the 1990s, with the specification becoming the most important artefact. A post on pooyam.dev, written by someone working on LLM training and inference performance on GPUs and TPUs, goes further: "I accepted that my former work life is fully over." A third, on rtfm.co.ua, warns that delegating to agents can leave the engineer as "meat middleware" between the agent and the deploy button.

Where that leaves the publishers

The pilot, the Commission investigation, the Penske suit and the German ruling are all live at once. Google has not published the payment formula, and the participants who spoke to The Information could not describe it either. The most recent hard number in the public record is the spread itself: under $1,000 for some, more than $1 million a year for at least one.

Meanwhile the traffic declines that prompted the deals continue, and the publishers holding out for better terms are doing so while their referral volume falls.