Business 3 min read
AI Is Breaking the Referral Economy That Built the Web
Publishers are losing the guarantee of a click, while stock-image and freelance platforms are losing the guarantee of an intermediary fee. The companies adapting best are moving toward data, trust and complex AI-enabled work.
For two decades, the commercial web rewarded intermediaries. Search engines routed attention to publishers. Stock libraries routed buyers to licensed images. Freelance platforms routed clients to workers. Each business captured value because it occupied a necessary step in the transaction. Generative AI is making several of those steps optional at once.
The change is easiest to see in referral economics. Cloudflare says AI crawlers consume large quantities of publisher content while returning comparatively little traffic. A user can ask a model a question, receive a synthesized answer and never open the underlying pages. Similarweb estimates generative AI services averaged 9.5 billion web visits per month between June 2025 and May 2026, up around 70% year on year.
Stack Overflow illustrates what happens when user behaviour changes upstream. The viral line that its traffic fell 98.5% is inaccurate; the near-99% collapse is in new questions from historical peak levels. More than 6,700 questions a day at the 2014 high became roughly 42 a day in May 2026. The result is not only fewer page opportunities but less fresh public data being created.
The same compression is repricing content marketplaces. Shutterstock’s second-quarter 2026 revenue fell 17% year on year, with subscribers down to 951,000. Its late-July market capitalisation was roughly $216 million. Getty Images was worth around $171 million, about 92% below its late-2023 level. Generative imagery is a structural threat to commodity stock content, although litigation, merger uncertainty and company-specific execution also matter.
Fiverr is an even more dramatic equity story. Its late-July share price was about 96.6% below the 2021 closing peak, while first-quarter active buyers fell 17.8%. For low-complexity tasks, the customer can now test an AI model before entering a marketplace at all.
Upwork provides a useful contrast. First-quarter revenue rose 1%, gross services volume remained near $987 million and AI-related work GSV grew more than 40%. Demand for AI-related skills rose 109%. That suggests the intermediary does not disappear if it moves up the value chain from matching simple tasks to providing trusted access to complex expertise.
There is an analogous path for publishers. A site with undifferentiated information is vulnerable if an answer engine can summarize it. A publisher with proprietary data, trusted reporting, direct audience relationships or licensable archives has more leverage. The strategic asset is no longer the page view alone; it is the source material and trust behind it.
AI is therefore not destroying the internet economy so much as removing rents from compulsory intermediation. Businesses that can prove rights, quality, reputation and accountability still have something a generic model cannot cheaply reproduce. Those that mainly sold access to a routine digital result face a much harder reset.
