Newsletter

Mike Masnick on protocols for publishers

An interview with @masnick.com (of TechDirt), the author of the influential 2019 paper "Protocols, Not Platforms," reflecting on how his ideas apply to publishers in 2026.

PfP Team

03 Aug 2026 — 8 min read

Mike Masnick headshot alongside art from the original article by Sara Wong

Last newsletter we reviewed a seminal paper from 2019: Protocols, Not Platforms: A Technological Approach to Free Speech which argues that "protocols could serve to do a better job of protecting both user privacy and free speech" amongst other benefits. This week have the author himself, Mike Masnick of TechDirt.com, reflect on his paper from seven years ago, and how the strategic landscape has changed.

On protocols

🙋

PfP✨: The paper's focus was on free speech. Composable moderation and interface competition makes sense for individuals but publishers are a specific kind of speaker with specific constraints (brand, liability, revenue, audience trust). As a publisher yourself, how do you talk to publishers about why they should care about the protocol layer? How does a protocol-based world change the relationship between publishers, the networks they publish to, and ultimately readers?

Mike Masnick: Right now, two key issues for any publisher is distribution and discovery. While the internet made some elements of both a lot cheaper, it came with other costs, including often having to hitch your wagon to specific platforms and to live by their rules and their algorithms. In a protocol world, publishers and readers gain back significant amounts of control. And while some brush this off as "too confusing" or "too much work" it need not be. We're already seeing intermediaries that embrace open protocols making it easier for publishers and their communities, but in an open way that creates better incentives for everyone, without bogging them down with lock-in and a loss of control.

On agency

💁

PfP✨: Publishers have spent decades being burned by platform dependency (Facebook traffic, Google News, Apple News, now AI crawlers). How can we ensure "protocols not platforms" is a better deal for them, not just inheriting a different kind of dependency?

MM: News aggregator platforms are, once again, a form of walled garden, where publishers (and readers) become totally dependent on the platform, whose best interests may be at odds with the publishers, the readers, or both. Certainly, over time, their interests are quite likely to diverge. With a protocol system, the tools are entirely open, which means anyone can build on them and you're never locked in. Without that kind of lock-in, intermediaries have much more incentive to keep their interests aligned with publishers and readers, or they will lose them to alternatives. It also means that publishers gain much more control over how those intermediaries work. Indeed, it's entirely possible for publishers (or readers!) to effectively build their own aggregation tools as well.

AI crawlers are a bit different, and perhaps a more complicated subject. Right now, there's a debate over whether AI crawlers should be blocked, with many leading publications choosing to do so. The risk, then, is that as more people still use AI tools for news, that low quality publications and shoddy information sources get overweighted as compared to more credible publications. But, at the same time, there's reasonable concerns about the large AI frontier model companies creating a similar form of centralized lock-in of the platforms today. But there is some (potential) good news there. AI lock-in appears to be way less endemic to the setup. In the agentic coding world, we've seen that the model is entirely separated from the context and the harness, such that you could use ANY model with your content and context (stored generally in markdown files). Meaning that it is currently quite easy to switch models without losing anything. At the same time, we're starting to see more and more sophisticated open weight and local LLMs that again mean that the giant centralized players will have less control and less power.

On moderation

🙋

PfP✨: The paper was written when the main complaint about platforms was too much moderation. Now Meta and X have essentially abandoned moderation. Is "composable moderation" fighting the last war? Publishers moderate comments sections, letters, and user-generated contributions. In the composable moderation world how do you think about the publisher's own moderation layer? Should they become labeling services? It seems to me many newsrooms would like someone else to take care of this problem. What are you seeing?

MM: There's wide world of possibility here where I don't think there's one single "right" answer. In a world of composable moderation, there are many different possibilities, boosted now by the power of some interesting LLM-powered tools (such as Zentropi's CoPE, which I've seen at least one publication using to nudge readers if their comments likely would violate a site's policies. There is a need for better curation, and publishers running their own custom feeds or custom communities can help enable that. Surf, from the company behind Flipboard, has been working with companies like Vox to create protocol-powered communities for all their sites, enabling the publishers to build better communities, but without having to manage every aspect of the infrastructure, instead, connecting into the wider open social space.

On money

🙍

PfP✨: At ATmosphereConf 2026 in Vancouver (Mike's recap | PfP✨'s recap) there was a lot of "we can just do things" energy with people building fun apps and experiments, interoperating and iterating off one another. That's exciting for developers. But publishers are not often known to be early adopters. What's the path from "enthusiastic builders" to "this is infrastructure publishers can actually rely on and build business models on”?

MM: Some of that will definitely take time, as it's still a time of experimentation. But in the ATprotocol world, we're already seeing some fascinating protocol-based experiments, including the standard.site lexicon, which is being adopted rapidly, and was developed by three different companies (Leaflet, Pckt, and Offprint) looking to provide tools to publishers (similar to a more centralized platform like Substack). But with those tools, there's no lock-in, so it's easy for publishers to begin to experiment, and if they choose to change or move elsewhere they won't lose out (or lose readers). So one nice thing here is that the "cost" of betting on one particular intermediary is minimal, since switching costs are low. We're also seeing early experiments in direct monetization, which should accelerate as the Atmosphere adds permissioned data into the protocol. It absolutely is still early, but establishing a foothold now does not take that much effort and given the open nature, it's much easier to iterate and adapt as the open social world grows and changes.

🙇

Thanks to Mike Masnick for the original paper and for this response! — PfP✨ Team

🍹 Some Summer Protocol Reading

Vint Cerf Retires From Google With a Warning: Al Agents Need Real Protocols. Protocols > Platforms. https://www.techtimes.com/articles/320264/20260712/vint-cerf-retires-google-warning-ai-agents-need-real-protocols.htm

From NiemanLab: Search traffic has declined so much that some publishers are considering opting out of Google entirely. “We want a technical solution that allows you to be discoverable without having to give your content away for free.” https://www.niemanlab.org/2026/07/search-traffic-has-declined-so-much-that-some-publishers-are-considering-opting-out-of-google-entirely/

Also from NiemanLab: "Despite concerns about AI Overviews, some nonprofit news outlets see a surge in search traffic" https://www.niemanlab.org/2026/07/despite-concerns-about-ai-overviews-some-nonprofit-news-outlets-see-a-surge-in-search-traffic/

Reddit stock sank 8% after it was reported the company is considering not renewing a deal with Google to use its content for AI use. "A lot has changed since those first deals were signed" https://www.msn.com/en-us/money/general/reddit-stock-sinks-on-report-it-may-not-renew-google-ai-content-deal/ar-AA28sVJm

A tale as old as… 2025? Movie data site Numbers, founded in 1996, is crashed by AI crawlers. "Google crawls about five pages for every visitor it sends you. OpenAI crawls over 1,000. Anthropic crawls over 38,000 pages for every single visitor it refers." https://stephenfollows.com/p/what-just-happened-to-thenumberscom-should-worry-us-all

OpenAI profiles 15+ newsrooms (e.g., Associated Press, Axios, The Atlantic, Le Monde) usage of AI for transcription, data analysis, personalization, translation, and interactive storytelling. This showcases OpenAI's support for the Lenfest Institute for Journalism's Collaborative and Fellowship program. https://openai.com/index/how-news-organizations-are-using-ai/

Google released the first of a large-scale survey called Al & Economy ATLAS about how people are using AI at home and at work, finding usage is broad but shallow. Most people are not automating every aspect of their lives like Silicon Valley salesman says they will. Also see this podcast which argues most people aren't tool builders. https://blog.google/innovation-and-ai/technology/research/understanding-the-ai-economy/

Columbia Journalism Review urges “radical transparency” around newsroom AI tool use covering spellcheck, transcription, research assistants, and fact-checking to maintain trust. https://www.cjr.org/ask-anika/how-transparent-should-i-be-about-ai-clearing-confusion-how-why-disclose-use-fact-check-spellcheck-thesaurus-transcription.php

Cloudflare has made two moves for Pay-Per-Crawl. First, it formalized a move from a pageview economy to an answer economy. Last year Cloudflare said AI companies would have to pay whenever a bot crawled a page, but now they pay publishers only when content actually appears inside an answer. The second move is starting September 15, 2026 crawlers classified as Training and Agent will be blocked by default on pages that display ads. https://ppc.land/cloudflare-stops-charging-ai-per-crawl-and-starts-paying-per-answer/

A payment middleman wants to by another payments middleman? Stripe is in talks to buy OpenRouter, a tool that allows companies to easily switch between AI models on a per-request basis to optimize costs. https://finance.yahoo.com/technology/ai/articles/stripe-talks-acquire-openrouter-potential-215104525.html

How the Mozilla Data Collective is helping communities fed up with Big Tech set up data collectives for control, including farming, fishing, genetic and environmental data in addition to the language data aggregated by Common Voice as discussed at PfP✨ London https://restofworld.org/2026/ai-data-collectives-mozilla/

The AT Protocol is much more than Bluesky. Internet researcher Erin Kissane is putting together "A Cloud Atlas" to make "this fizzy explosion of projects" much more legible to those not following the protocol community closely. https://www.wrecka.ge/launching-a-cloud-atlas/

A four minute introduction to the Atmosphere (what we call the network of apps built on atproto) written, illustrated, and published by Leaflet (a publishing app), but presented here in Standard Reader (a reading app). https://standard-reader.app/a/did:plc:btxrwcaeyodrap5mnjw2fvmz/3mrq6ma5dgs2x?q=JFZRchqzmB

"The easiest way to tell if a platform is open or not: do you need to ask permission to integrate with it? Open: RSS, IndieWeb, Bluesky, Mastodon, Micro.blog. Closed: Threads, Instagram, LinkedIn, YouTube." https://www.manton.org/2026/07/28/dave-winer-blogging-about-a.html

Case study of how Jacobin magazine connected its Ghost instance to the ActivityPub-powered Bonfire for more intentional community management. https://bonfirenetworks.org/case-studies/jacobin/

Persona is a new framework for creating agentic front-end experiences with vanilla JS and Google's new WebMCP standard, which has been proposed to the W3C. https://www.persona-chat.dev

Ghost CMS added support for llms.txt and markdown output to help with AI "GEO" discovery. https://ghost.org/changelog/geo/