AI Search Is Rewriting the Rules of Visibility

Ido

Welcome to Be Visible. I'm Ido, this is Ben, and the news this week is about how AI systems find your pages in the first place. Let's start with Search Engine Journal on October 20th. Cloudflare says its network signals helped OpenAI index hundreds of millions of new pages for ChatGPT search. That's a big number.

Ben cohen

It is, and I want to be careful about it. That's Cloudflare's claim, and it came out of a pilot. But if it's true, discovery isn't just crawling anymore. A company that sits in front of a huge share of the web is telling an AI company which pages exist and which ones changed. That's a different pipeline, and I think it's a little worrying.

Ido

I'd push on worrying. Why isn't it an opportunity? If your site sits behind that network, your pages can surface faster. That sounds like a win to me.

Ben cohen

For the sites behind it, sure. What about the sites that aren't? We don't have evidence yet of a divide. I'm saying the question is fair. If discovery drifts toward partnerships between infrastructure companies and AI labs, then where you host starts to matter for visibility. Nobody used to think about that.

Ido

Fair, but I'd call that a reason to watch it, not a reason to panic. One pilot doesn't make a standard. Here's the one I find more exciting. Personal agents. Meta's Muse and OpenAI's dots, per Search Engine Journal. They reach websites through connectors or cloud browsers.

Ben cohen

And that changes what robots.txt can tell you. A classic crawler announces itself, and you can allow it or block it. An agent acting for one person through a cloud browser isn't that. So your rules and your analytics may not describe what's really visiting.

Ido

Right, and that's the part people miss. If an agent reads your pricing page for a buyer, is that a visit? Does it count as a session? Your dashboard might call it a human.

Ben cohen

Source reporting says it changes what analytics can tell you. I'd go further as interpretation. If you can't separate agents from people, your engagement numbers get muddy, and so does your conversion story. The fix isn't blocking them. It's making your pages readable when everything visual is stripped away. That's exactly what I built my AI Content Reader to show.

Ido

Okay, here's where I think we agree. Be readable, be clear. Now DeepSeek. It opened its in house search engine to developers on October 19th. Every result carries a page level authority tier, and DeepSeek says it filters GEO manipulation.

Ben cohen

And then the hands on test is my favorite detail this week. The tester found DeepSeek's date filters dropped every undated page. That was 11% of results. Eleven percent, gone, just because there was no date.

Ido

Gone silently, too. Nobody gets a warning. So the boring answer is a publish date and a modified date, clearly marked, on every page.

Ben cohen

Yes. I'll say it plainly. This one is cheap to fix, and nobody should lose visibility over it. That's the metadata hygiene story.

Ben cohen

Now the data that should wake people up. Ahrefs found that AI Overviews now appear for 83% of branded queries. Branded. These are searches where someone already typed your name.

Ido

And Ahrefs says that's potentially reducing clicks to brands' own websites, with third party sources like Wikipedia and YouTube filling the space. Notice the careful word, potentially. They didn't measure the lost clicks directly in what we have.

Ben cohen

Fine, but look at the buyer side. Forrester found 94% of business buyers use AI search when buying, and they rate it above vendor websites, product experts, and sales reps. NielsenIQ puts it at 42% of consumers researching products with AI.

Ido

So your homepage isn't the first impression anymore. Somebody else's summary of you is.

Ben cohen

Which is why I'd say this is the biggest shift of the week. The brand search used to be the safe one!

Ido

I'd say it's the biggest opportunity, not the biggest threat. If third party sources shape the answer, you can work on those sources. Also, Ahrefs studied 66.6 million logged out GPT 5.6 Luna responses and 15 million fan out queries. 14.4% of those responses used fan out. So your page has to answer the sub questions, not just the headline question.

Ben cohen

But that cuts both ways, and here's the dark side. A paper reported by Search Engine Journal posted about a made up concept on sites AI search tools cite. Four tools on the English web cited it within a week.

Ido

Though other topics mostly didn't land. That qualifier matters. It doesn't mean any lie gets cited.

Ben cohen

Agreed. But it does show that being cited and being true are different things. And that makes DeepSeek's claim about filtering GEO manipulation interesting. It's a claim, and we haven't seen proof it works.

Ido

One more piece of plumbing. Search Engine Land created a fake SEO term, blocked Bing, and asked ChatGPT about it. The answer, they say, proves it's using Google Search.

Ben cohen

So the engines are tangled together. Optimizing for one assistant quietly means optimizing for what feeds it.

Ido

And Meta shows the policy side. Its new commercial terms let it crawl sites tied to business users and train on the content. Then on October 19th Meta removed training from the crawling clause, but kept the permission to crawl.

Ben cohen

That's the one I'd read twice. Training is out of that clause. Crawling is still in. If you run ads or have a business account, check what you've agreed to.

Ido

So where do we land? I say stay calm and fix the basics. Dates, readable pages, and a presence on the third party sources that shape your brand answer.

Ben cohen

I'll take that list, but I won't call it calm. Discovery is moving to infrastructure deals and personal agents. Check your site the way an AI sees it, this week, not next quarter.