# The second reader

Why the web's audience changed, what that does to media, and the company we are building for what comes next.

Fores Media · London, October 2026

## A web written for one reader

For thirty years every page on the web was written for a person. Layout, headlines, images and advertising all assumed a pair of eyes at the other end, and the economics of media followed: attract the eyes, count them, sell them. Search engines sat in the middle as brokers, sending readers to pages in exchange for the right to index them.

That arrangement depended on one thing: the reader had to arrive. A search result was a promise that a person would click through, see the page, and be counted.

## The second reader arrived early

By early June 2026 bots made up about 57.5% of requests for web pages on Cloudflare's network, overtaking people.[1] Only three months earlier, Cloudflare's chief executive had predicted the crossover for 2027.[2] Imperva, measuring all web traffic, had already put automated traffic at 51% in 2024 and 53% in 2025.[3]

The fastest-growing slice is not the old crawler. Daily requests from AI agents, software acting on someone's behalf, grew more than 1,700% on Cloudflare in the year to September 2026.[4] On publisher sites, AI-bot visits went from one for every 200 human visits in the first quarter of 2025 to one for every 31 in the fourth.[5]

This reader does not browse. It arrives with a task: compare these companies, check this figure, summarise this market, book this, buy that. It reads ten sources in the time a person reads a headline, and it never sees an advert.

## What the second reader wants

An agent wants what a careful analyst wants, only faster and at far greater volume. Facts, not prose. A figure with its source, its date and its currency. The same company named the same way every time. Terms that say what it may do with what it reads. And an interface it can call, rather than a page it has to scrape.

The standards for this arrived in the last two years. The Model Context Protocol lets agents call tools and data sources directly; its official SDKs together are downloaded close to half a billion times a month.[6] Cloudflare now lets sites charge AI companies each time their content is used in an answer, not only each time it is crawled, and since 15 September 2026 it blocks AI training and agent crawlers by default on ad-carrying pages of newly onboarded sites.[7] Machine-readable licences such as RSL are backed by more than 1,500 publishers, brands and technology companies.[8]

## Why media breaks

When the agent answers, the person does not click. In a Pew study of US adults' Google use, people clicked a traditional result on 8% of searches that showed an AI summary, against 15% of searches without one, and a link inside the summary on just 1%.[9] Google Search referrals to publishers fell 40% between July 2025 and July 2026, after a 22% fall the year before. AI chatbots sent 0.01% of traffic.[10]

Europe has it worse. In the first half of 2026 its publishers received one human referral from AI apps for every 179 AI-bot visits, more than three times worse than North America.[11] The archive is being read more than ever. The business built on counting readers is being paid less than ever.

The response so far has been defensive: block the bots, sue the labs, sign one licensing deal and hope it renews. All three treat the second reader as a threat. None treats it as a customer. We do not expect the second reader to send traffic back. We expect it to cite, and to pay.

## The source economy

The money is already moving. Licensing is shifting from one-off training datasets toward ongoing feeds and live access, paid for continuously.[12] Google and Microsoft now pay publishers according to how their content is used in AI answers.[13] Axios is launching data feeds for investment firms and plans feeds for AI models and for readers' own agents.[14] PitchBook, Crunchbase, CB Insights, Dealroom and Preqin have all opened MCP servers to their private-market data, each behind a subscription.[15] Advertising has followed the answers: ChatGPT's ad business passed a $1 billion annualised run rate less than 200 days after launch.[16]

Investors are pricing the layer that feeds agents. Exa, a search engine built for AI, was valued at $2.2 billion in May 2026; Parallel Web Systems at $2 billion three weeks earlier.[17] Both build their own indexes of the open web. Neither owns verified, structured truth about a domain: who raised what, from whom, when, and according to whom. That is the scarce input, and it is what a newsroom with a data desk produces every day.

This has been done before. Crunchbase began in 2007 as TechCrunch's database of the companies it covered. Mergermarket turned M&A reporting into a data business that sold for £1.35 billion in 2019.[18] Bloomberg built a newsroom to make its data indispensable to traders. Fores Media runs it the other way: a newsroom whose reporting becomes the data for private markets, read by people and called by agents.

## What we are building

Fores Media is a media company built for both readers. Every product has a page for people and an endpoint for agents, and both read from the same records. Every figure links to where it was published and carries a confidence grade, so when an answer engine quotes us, the citation travels with the claim.

We started where the facts are hardest to get right and most valuable to machines: technology companies, the capital behind them and the people who build them. tech.eu has reported on European technology since 2013. The Tech.eu Funding Explorer turns that reporting into a cited dataset of more than 52,000 European funding rounds, free for any agent to query through an API or an MCP server, without a key. Where incumbents keep their agent access behind subscriptions, ours is open, and every figure carries its source.

Next comes the work the second reader makes possible: the private-market graph of the world, licensed data feeds that AI companies pay for each time they use them, and every conference stage indexed into the same graph.

## 2030

By 2030 most reading on the web will be done by software, and the media companies that matter will be the ones software trusts. They will be cited, licensed and paid per use. Their stories will be living records rather than finished pages. Every reader will get their own edition, and small teams will cover the world.

We are building that company now, in public, with both doors open.

## Notes

1. Cloudflare Radar data posted by Matthew Prince, 3 June 2026, via Forbes. https://www.forbes.com/sites/josipamajic/2026/06/04/bots-now-outnumber-humans-online-and-the-internet-was-never-built-for-this/
2. Matthew Prince at SXSW, March 2026, via Search Engine Land. https://searchengineland.com/bots-overtake-human-web-usage-2027-472109
3. Imperva Bad Bot Report 2025 and 2026, Thales. https://community-cybersec.thalesgroup.com/blogs/tim-chang/2026/05/05/the-2026-bad-bot-report-is-here
4. Cloudflare, The Internet has a second audience, 30 September 2026. https://blog.cloudflare.com/agentic-web/
5. TollBit State of the Bots, Q3 to Q4 2025, via The Register, February 2026. https://www.theregister.com/software/2026/02/04/ai-bot-traffic-closing-in-on-human-web-visits-study-finds/4921090
6. Model Context Protocol blog, The 2026-07-28 Specification. https://blog.modelcontextprotocol.io/posts/2026-07-28/
7. Cloudflare, Pay Per Use, 30 September 2026, and AI crawler defaults, 1 July 2026. https://blog.cloudflare.com/pay-per-use/ · https://blog.cloudflare.com/content-independence-day-ai-options/
8. RSL Collective, RSL 1.0 specification release, December 2025. https://rslstandard.org/press/rsl-1-specification-2025
9. Pew Research Center, 22 July 2025. https://www.pewresearch.org/short-reads/2025/07/22/google-users-are-less-likely-to-click-on-links-when-an-ai-summary-appears-in-the-results/
10. Chartbeat 2026 Publisher Playbook, via Search Engine Journal, September 2026. https://www.searchenginejournal.com/google-search-decline-publishers-chartbeat/590681/
11. TollBit State of the Bots, H1 2026, via Digiday, August 2026. https://digiday.com/media/european-publishers-are-getting-hit-harder-by-ai-bot-scraping-report-finds/
12. Rob Kelly, Media and the Machine: AI content licensing deals, June 2026. https://mediaandthemachine.substack.com/p/ai-content-licensing-deals-june-2026
13. Google pay-per-use programme via Digiday, 14 September 2026; Microsoft Publisher Content Marketplace, February 2026. https://digiday.com/media/google-rolls-out-pay-per-use-ai-licensing-program-to-publishers/ · https://about.ads.microsoft.com/en/blog/post/february-2026/building-toward-a-sustainable-content-economy-for-the-agentic-web
14. Digiday on Axios Direct, 18 September 2026. https://digiday.com/media/axios-preps-new-axios-direct-feeds-for-ai-models-agents-as-revenue-tops-2026-goal/
15. PitchBook and Perplexity, March 2026; Crunchbase, Dealroom and Preqin MCP documentation. https://pitchbook.com/media/press-releases/pitchbook-announces-new-essential-mcp-integration-with-perplexity-expanding-access-to-ai-powered-verifiable-market-intelligence · https://data.crunchbase.com/docs/mcp-overview · https://developers.dealroom.co/agent-apis/mcp · https://preqin.com/our-products/preqin-mcp
16. OpenAI, 31 August 2026. https://openai.com/index/expanding-access-to-ai-with-chatgpt-ads/
17. Exa Series C, May 2026; Parallel Web Systems Series B via TechCrunch, April 2026. https://exa.ai/blog/announcing-series-c · https://techcrunch.com/2026/04/29/parallel-web-systems-hits-2b-valuation-five-months-after-its-last-big-raise/
18. Crunchbase and Mergermarket (Acuris) company histories. https://en.wikipedia.org/wiki/Crunchbase · https://en.wikipedia.org/wiki/Acuris