Paying for AI content: how much publishers get and what the courts will not fix

Google pays some publishers a fraction of a per mille of ad revenue, Cloudflare has launched content usage fees in beta, and a US court dismissed the lawsuits against Google. For companies investing in content, this is a signal that licenses will not quickly replace lost traffic.

Companies that have invested for years in content marketing, expert blogs, and industry websites have a new problem. Their texts are increasingly ending up in chatbot answers and AI summaries, and increasingly less often bringing clicks. In recent days, three pieces of news have emerged that show how much can be recovered from this so far: money from Google's pilot, Cloudflare's new payment models, and a US court ruling. We put them together to check what options a website owner has today.

Google pilot: rates meaningful only for a few

According to The Information, cited by Search Engine Journal, Google is currently paying about 100 digital publishers as part of a pilot program compensating content contributions to AI answers. The program has been testing for less than a year. The range of payouts is huge: some smaller websites earned less than $1,000 over several months, while one early participant receives more than $1 million a year.

For several small and mid-sized websites, these payouts amount to less than 0.1 percent of ad revenue. A publisher who joined a few months ago has earned $50,000-60,000 so far and admitted that it is a small part of their business. According to the same reports, content on topics that attract strong user interest but are covered less often, such as anime or games, appears to earn more.

The question remains what exactly Google is paying for. Content is supposed to earn money when it “significantly contributes” to an answer in the Gemini app, in AI Overviews (AI summaries above search results), and in AI Mode. Fact-checking or a link added after the answer is generated does not count. Google has not explained what “significantly” means. Participants see a monthly amount in Search Console (Google's free dashboard for website owners), but without any methodology, and some of them say the payouts change from month to month without explanation. Some large publishers have not yet joined the pilot because they considered the offered amounts too low, and are hoping for higher rates from Google and other AI companies.

The difference between a small and a large publisher is clear here. A large publisher has negotiating leverage and can wait for a better offer. A small website gets a rate it does not know in advance and cannot verify.

Cloudflare: the price list is set by the buyer or the seller

On September 30, Cloudflare introduced two products in beta. Monetization Gateway makes it possible to charge an AI agent for each query to an API, tool, dataset, or page. The price is set by the seller, settlements range from $0.001 to $100, and payments are made in the USDC stablecoin (a cryptocurrency pegged to the dollar). For now, this is a closed beta exclusively for US companies, with additional requirements, including an account older than 60 days.

The second product, Pay Per Use, reverses the roles. The AI company decides what content usage it will pay for and how much, and the publisher can accept or reject the offer. Usage is reported by the buyer, and Cloudflare only checks whether the report matches the registered publisher, then settles monthly. Cloudflare has not disclosed the names of AI companies buying in this program or any data on the number of transactions or payouts. Industry estimates circulating of $50,000-200,000 per month refer to the earlier Pay Per Crawl, cover only a narrow group of websites with huge traffic, and do not come from the company itself.

Payment for registered content usage is also provided for by Publisher Content Marketplace, which Microsoft launched on February 4, 2026, but there the licensing terms are set by the publishers themselves. The pilot included participants such as Associated Press, Condé Nast, and Vox Media. Each of these companies defines “usage” of content differently, however, and there is no common measurement method.

The court acknowledged that publishers have a problem, but dismissed the lawsuits

On the same day, September 30, Judge Amit P. Mehta of the US District Court for the District of Columbia dismissed the antitrust lawsuits by Chegg and Penske Media (publisher of Rolling Stone and Variety, among others) against Google. He rejected five categories of allegations, including tying the search engine to AI Overviews and the alleged “content for traffic” agreement. In the reasoning, he wrote that publishers merely had an expectation that Google would send them traffic in exchange for free access to content: “But an expectation is not an agreement” - an expectation is not an agreement.

The judge did not find that publishers suffered no harm. He expressly acknowledged that he does not dismiss their situation or the consequences for journalists and creators whose content Google uses without compensation. The lawsuits failed because they did not meet the requirements of US antitrust law. According to Forbes, this is already the third time this judge has dismissed similar allegations concerning Google's AI features, and he himself suggested that responding to changes caused by new technology is a task for lawmakers. In the Penske Media case, the dismissal was without prejudice to refiling, and the ruling is final and can be challenged on appeal. For companies, the takeaway is practical: it is not worth counting on a quick court resolution.

Block, sell, or measure

The scale of the phenomenon is growing. According to data from TollBit, a company that intermediates content licensing, the share of clicks from AI platforms to its partners fell from 0.8 percent in the second quarter of 2025 to 0.27 percent in the fourth. Importantly, it was also falling among publishers with signed agreements with AI companies.

The simplest response, namely blocking bots in the robots.txt file (instructions for web crawlers, compliance with which is voluntary), is not a universal solution. Researchers from Rutgers Business School and Wharton analyzed 30 large news websites: after blocking language models, their traffic fell by 23 percent according to SimilarWeb and by 14 percent according to Comscore. For some smaller websites, the effect was the opposite, but the authors treat these results as preliminary.

That is why it is better to precede the decision with measurement. Google recently added UTM parameters (source tags in the URL) to outgoing links from Gemini, making it easier to distinguish this traffic from direct visits in Google Analytics. However, this works mainly on desktop: in the Gemini mobile app, the parameters appear in only about 9 percent of cases, and Android assistant calls remove the source information altogether.

Our conclusion: for most companies investing in content, payments from AI are currently not a source of revenue but an experiment with opaque rules. A sensible strategy for now is to measure traffic from chatbots, take a selective approach to bots (training bots separately, search bots separately), and treat licensing offers as an extra rather than a business plan.