Are user agents people?
Perplexity and Cloudflare fight over the future of bots. Plus: Interviewing the dead with AI, Trump becomes a copyright victim, and more
The fight that broke out this week between Cloudflare and Perplexity is a fascinating one. Cloudflare accuses Perplexity of using bots improperly, which defends itself by saying its bots are behaving just like people, which makes it OK. I couldn’t help but think of Mitt Romney’s famous “corporations are people” line. Romney was speaking more philosophically, but in the case of bots there’s a real question to be answered of just how much equality we should grant agents acting in our name.
More on that in a minute, but some quick notes about what I’m up to the next few weeks:
My monthly 1-hour AI Quick Start class is back! The next edition is happening Thursday, Aug. 21, and it’s still just $49. Sign up here.
I’m partnering with The Upgrade to offer two six-week AI training courses in the fall: one for journalists, and one for PR & comms professionals. They’re loaded with media- and PR-specific use cases and are designed to transform the way you work. See the ad below for a discount code to save big, and get in touch (or just hit reply) if you’d like to send a group.
For the next few weeks I’m going to be building something special for The Media Copilot, plus I’ve got a vacation planned next week. As a result, I’m dialing back the newsletter to once a week for the rest of August. It’ll go out Tuesdays, and I’ll try to get some news items in as well. The podcast will still go out every other Friday. But come September, expect things to ramp up considerably.
OK, now a little more about those courses, then let’s talk about user agents.
A MESSAGE FROM THE MEDIA COPILOT AND THE UPGRADE
You hear it everywhere now: AI is changing the workplaces across industries. Reduced headcount, enterprise platforms, redefined jobs. If you’re still monkeying around with some favorite prompts in ChatGPT, chances are you’re behind. And if you’re leading a team, you can’t put off AI training anymore.
That’s why The Media Copilot has partnered with our friends at The Upgrade to offer two extremely focused AI training courses—one for journalists, one for PR pros—to show how you can go beyond the prompt and transform the way you work. Kicking off in September, these six-week programs focus on media workflows and real productivity gains, with 1:1 coaching and a final project to cement your skills.
💥 Act fast: 25% discount ENDS August 7.
🔹 AI for Journalists: Use code AIJO25OFF-SEP (checkout link)
🔹 AI for PR & Comms: Use code AIPR25OFF-FL1 (checkout link)
Cloudflare and Perplexity mark the next AI battleground: user agents
When Cloudflare announced In July that it would start blocking AI bots on its network by default, it was clearly serious. The media blitz came fast and furious, including the press release full of quotes from prominent media executives, an interview with the CEO in The New York Times, and plenty of other coverage all over the AI and media internet. The Media Copilot even got in on the action with an interview with Chief Strategy Officer Stephanie Cohen.
Now we know what comes next. Cloudflare just escalated its conflict with Big AI, publicly accusing Perplexity of ignoring the Robots Exclusion Protocol (robots.txt) in order to access websites with its user agent bots. Perplexity fired back with its own blog post, arguing it had done nothing wrong and accusing Cloudflare of mischaracterizing what its bots were doing.
While user agents are a type of AI bot, they are different from bots that perform AI training or search the web generally. A user agent is a bot that acts on behalf of a user. For example, if I tell Perplexity's Comet browser to find out if my local Home Depot has in-store pickup for a new kamado grill, the agent would go and check that information in real time, as opposed to falling back on previously crawled data.
Clearly, Perplexity sees this proxy behavior as something fundamentally different from what robots.txt is supposed to manage, and can effectively ignore the protocol. The logic: Since the agent is simply doing what a human does, and isn't storing or training on the data, it should get to play by the same access rules as humans do.
And it may have a point. Perplexity isn't the only AI player who grants its user agents permission to bypass robots.txt. Google and Meta make this explicit in their terms of service as well (although Cloudflare says OpenAI's agent bots are apparently "good actors" compared to Perplexity's). When I asked a Perplexity spokesperson about its stance, he said, "User agents (of anyone, not just Perplexity; this is part of their definition) do not store or retain data for training purposes, and they are limited only to the actions specifically requested by the user."
That makes agents distinct from other types of AI bots, but from a user standpoint, it's a distinction without a difference. A person using Perplexity or Comet just wants the answer served up or the task completed—they don't care what kind of bot does it. And the site the bot is visiting doesn't care either. It only cares whether the visitor is robot or human, because only one of those things looks at ads.
Like other aspects of AI that simulate human behavor—"reading" a lot of content about as subject and then writing something original about it, for example—this may be a case where what's the technology is doing has generally been tolerated up until now, but the scale and sophistication that AI brings to the table alters the picture considerably.
Think about it this way: It's hard to argue that, if all human web traffic were suddenly replaced by user-agent traffic, the internet would function exactly the same. Before the bots take over, we should at least agree on what they are.
The Chatbox continues below
A MESSAGE FROM THE MEDIA COPILOT
AI is now part of the everyday workflow for media pros. More than 60% of journalists and communications professionals use AI to research, brainstorm, draft, and pitch — and those numbers are only growing. But most are still just dabbling, tossing questions into ChatGPT without much strategy.
AI Quick Start is a one-hour AI power session is your chance to go deeper. No fluff, no tool demos — just clear, practical ways to make AI a real advantage in journalism and PR.
You'll walk away with the strategy, mindset, and confidence to use AI effectively — without losing your edge or your voice.
What you'll learn:
✅ Where AI is actually changing media today
✅ The mindset top pros are using to stay ahead
✅ How to apply AI, from sharper pitching to faster story development
Details:
🗓️ Aug. 21, 1 p.m. ET
💸 $49 — a smart spend to stay sharp
🎁 Includes slides, pro prompts, and a Quick Start AI guide
The Chatbox
All the AI news that matters to media*
The ethics of interviewing the deceased
When former CNN anchor Jim Acosta interviewed the AI avatar of one of the slain victims of the 2018 Parkland high school shooting, he may have thought he was on the bleeding edge of the intersection of AI and media. Instead, he may have inadvertently shown just what a misguided use case the idea of interviewing an AI avatar can be. The father of the victim, Manuel Oliver, created an AI version of his son Joaquin through prompting, context, and imagery, but the limits of the technology were apparent throughout the so-called interview: The avatar's intonations weren't natural, the tone changed jarringly more than once, and it always closed its answers by reflecting the question back on the asker.
But moreover, with full respect to the seriousness of the original incident, it's difficult to mine journalistic value out of this interview. AI systems only know what data they're given, and they only behave according to their instructions. Acosta didn't probe the Joaquin avatar in the way he would have a real person, and there wouldn't have been much point if he did. As a stress-test of the current state of AI visual avatars, the interview exposed that there is work to be done. And that probably goes for AI in journalism, too.
Google disputes study showing AI hurts clicks
Google's last earnings call offered a counternarrative to the recent Pew Research study that found AI Overviews reduce website clicks by nearly half, according to MediaPost. While Pew's data showed only 8% of AI Overview searches led to clicks compared to 15% for traditional results, Google claims its AI features actually drive 10% more queries and billions of daily clicks to websites. The company dismissed Pew's methodology as "flawed" and unrepresentative, arguing the study only examined clicks to the top three links rather than the full range of sources AI Overviews can surface. With AI Overviews now reaching 2 billion monthly users, the stakes couldn't be higher for media companies trying to understand the full effects of AI on their businesses. (AI-assisted)
NYT-Amazon deal shows what news costs in AI age
Amazon's $25 million annual deal with The New York Times, reported by The Wall Street Journal, offers the clearest window yet into how publishers are pricing their content in the AI era. The payment represents nearly 1% of the Times's total 2024 revenue, establishing a benchmark for what premium news content commands from tech giants. While OpenAI's News Corp deal reportedly reaches $250 million over five years, Amazon's arrangement shows how even "first mover" agreements are generating substantial revenue streams for major publishers. Who knows? Maybe strategic partnerships with select AI companies could offset some of the traffic and revenue losses from AI-powered search engines, though if the courts chimed in on one of the major AI copyright cases, it could alter the picture considerably. (AI-assisted)
News Corp makes AI theft personal for Trump
News Corp CEO Robert Thomson's pointed appeal to Trump about AI companies stealing "The Art of the Deal," as reported by The New York Post, illustrates how the copyright battle has evolved beyond industry complaints into a political strategy. By framing the issue around Trump's own bestselling book being "cannibalized" by AI systems like Meta's Llama, Thomson is making the intellectual property fight personal for a president who's already signaled support for AI companies training on publicly available content under the banner of fair use. While News Corp has secured licensing deals with OpenAI and filed lawsuits against Perplexity, Thomson's calculated move to position Trump as a victim of AI theft might end up proving more effective than courtroom battles. The timing is shrewd: as the new administration rolls out its AI Action Plan emphasizing infrastructure investment over content protection, media executives need political allies who understand that creative work fuels the very innovation America claims to champion against China. (AI-assisted)
*Some items are AI-assisted. For more on what this means, see this note.