Cloudflare's Matthew Prince on Saving the Web From AI: What Builders Should Know

The Verge interviewed Cloudflare CEO Matthew Prince for a podcast episode in a two-part series on the future of business. It asks whether he can "save the web from AI". Prince last appeared on the show about two and a half years ago, and The Verge's framing is that the internet has moved again since then. The episode URL also carries the phrase "google zero", the idea that search referrals to publishers could shrink toward nothing as AI answers absorb the click.

We are working from the episode's public summary and Cloudflare's own published policy, not a full transcript. So this piece looks at the position Cloudflare has put on record and what it means for people who build software on top of web data.

What Cloudflare has already committed to

On July 1, 2025, Cloudflare announced what it called "Content Independence Day". Its stated policy is to block AI crawlers unless they pay creators for their content. The company also says it plans a marketplace where AI companies and content creators can negotiate directly. Cloudflare argues that the old search bargain, where traffic was the payment, no longer holds for AI engines.

Cloudflare's post also puts numbers on the imbalance. It says it is "750 times more difficult" for creators to gain traffic from OpenAI than from legacy Google search, and "30,000 times more difficult" with Anthropic. These are Cloudflare's own claims, so treat them as one side's argument.

Why this matters to builders

A large share of the current AI toolchain assumes the open web is readable by default. That includes agents that browse, RAG systems that scrape reference pages, and research bots that pull market commentary. If a big edge network blocks crawlers by default, that assumption becomes a dependency you need to manage.

• Agents that browse: expect blocked or challenged requests, and design fallbacks instead of retrying blindly.
• RAG pipelines: prefer licensed feeds, official APIs and content you have rights to, and log the provenance of every source.
• Trading and fintech data: market and news inputs already come from contracts and APIs. Scraped commentary was always the weakest link, and it is now the likeliest to break.
• Compliance: in regulated products, being able to show where each input came from is becoming part of the engineering job.

The open question

It is not yet clear whether a pay-per-access model can work at scale. Small publishers may gain leverage, while AI companies may shift toward licensing deals and first-party data. It is also unclear how much of the web's traffic problem an infrastructure provider can fix from the network layer. That is the tension the interview appears to be probing, and we would rather flag it than guess at Prince's answers.

A practical checklist

1. Identify every place your agent or pipeline fetches public web pages.
2. Respect robots.txt and crawler policies, and identify your user agent honestly.
3. Replace fragile scraping with APIs or licensed data where you can.
4. Add graceful degradation: when a source blocks you, the agent should say so instead of inventing an answer.