Anthropic is in talks to raise a new $50bn round at a $900bn valuation. That's a number that would have sounded ridiculous twelve months ago.

Meanwhile The Wall Street Journal reported this week that OpenAI is missing key revenue and user targets as it sprints toward its IPO. The company set a goal of one billion weekly active users for ChatGPT by the end of last year. It's currently sitting somewhere between 920 and 950 million.

The bigger story is internal. Sarah Friar, OpenAI's CFO, has reportedly told other company leaders she's worried OpenAI may not be able to pay for future computing contracts if revenue doesn't grow fast enough. She's questioned why Sam Altman is still investing heavily in data centre growth when user growth has slowed. And according to the reporting, she's been left out of key meetings.

A CEO and CFO who don't see eye to eye on the IPO plan is not a small problem. Altman wants to plough ahead in Q4. Friar has reservations.

OpenAI is burning cash at the rate it needs to compete. Anthropic is being valued like the market thinks it can win. That's the gap to watch over the summer.

I made it for the Friday only. Mike, my co-founder and Chief Customer Officer at Obsero, was on a round table about the future of AI search and the impact on consumers, which I caught the back end of, and I sat in on most of the morning Auditorium One talks.

A few things stood out.

The platforms owned the stage. The agencies didn't. Profound, Peec, ahrefs and SEMrush all had polished stands. The legacy SEO tools were there in force. The agencies were in the room in numbers, just not on stage (from what I saw) and not sponsoring. That tells me there's a gap, and it's a gap Obsero is going to look at next time around (owning both the agency and tech relationship).

Community kept coming up. Not just the Reddit-as-a-citation-source angle that everyone's covered. Brands building their own communities, being helpful inside spaces where their category is being discussed, rather than parachuting in to sell. That's where the long-term retrieval signals get built, both for humans and machines.

Samanyou Garg from Writesonic talked about how they use agents to power their own marketing. It was a more useful talk than I expected, partly because it showed what platforms should be doing on stage: telling a narrative that helps the audience, not running through the same citations-and-content-structure deck.

Ryan Law from ahrefs was the standout. Clear, well-structured and extremely useful on how AI search works in practice. Training data, grounding, the principles models lean on when they're not confident. Insightful, engaging and worth your time.

The main gripe: a few main-stage talks were ruined by slides crammed with tables and data nobody could read in a 700-seat room. I get there are nerves and wanting to make sure you don’t miss anything, but on a stage that size, less always wins.

If you were there, hit reply or leave a comment. Keen to hear what you thought of it.

Ty Magnin, the CEO of Animalz, posted on LinkedIn this week pointing at one of Profound's headline customer case studies. He pulled the ahrefs data for Airbyte's data-engineering-resources’ subfolder, flagged that those 427 pages are being graded as "Very High" for AI Risk in ahrefs, and argued the content effort itself is the problem: walls of text, low information gain, full of AI-isms, outside the core of what Airbyte does.

His advice at the end was constructive. Walk before you run. Keep a human in the loop. Tens of pages per month, not hundreds.

I want to build on that.

The headline reading some people will land on is "AEO killed Airbyte's traffic, AEO means AI content at scale, AEO doesn't work". That isn't what Ty argued, but it's where the comments tend to drift. And it's wrong.

Look at what Profound's own case study describes Airbyte doing. It isn't a content factory. It's work on existing pages: rewrites into clearer answers, schema, llms.txt (which is a waste of time), server-side rendering. Whatever you make of the outcome, the activity Profound is selling isn't "publish hundreds of AI pages and hope".

Two things can be true at the same time. Airbyte appear to have a content programme inside their data engineering resources subfolder that Ty's pointed at fairly. And AEO done well, by whatever name you want to give it, doesn't have to come at the cost of your organic search presence. It isn't one or the other.

We don't know the full story, and I'm not going to pretend we do. We don't know how much of the AI-flagged content predates the Profound work, runs alongside it, or is unrelated to it.

The warning still stands. If you're scaling hundreds of AI-generated pieces with no original research, no opinion, no live experience, no video, you've created commodity content at scale. That isn't an AEO failure. That's a content strategy failure.

Both can coexist. They have to. But you can't go all in on commodity content and expect your established organic traffic to survive. There's a responsibility in what you produce: it has to resonate with your ICP, it has to be a content experience built for humans, not just shipped because it's cheap.

It is also worth highlighting: this isn't proof AEO doesn't work. Anyone with a big SEO following and a grudge against AI search will use cases like this to shout that the whole category is broken. It isn't. Bad strategy is. We built Obsero because there's a real discipline here, not a shortcut. GEO done well, with differentiated content underneath it, still moves the needle, both in traditional and AI search.

None of this stops me using AI in content workflows. I use it daily. But the workflow matters. Subject matter expertise, fact-checking, a unique angle. If those aren't in the loop, you're shipping without thinking.

I've been using Claude Design more this month and one thing kept driving me mad: there's no clean way to export to video. You can't drop the output into Canva easily either, which I'll come back to in a future edition.

I found a fix on Reddit. A developer called fermatf has built a free tool called Claude to Video. You paste your Claude design URL into the field, and it exports to MP4. That's it.

Useful if you're running anything programmatic, dropping designs onto social, or just want a video version of something you've made.

I'd assume Anthropic will close this gap in Claude Design itself eventually. Until then, this works.