Beyond the Blue Link: Making Your Brand Visible to AI

Ben cohen

Picture this. Someone runs a small agency and needs a new project management tool. They open an AI assistant, describe their team, their budget, and the three things that drive them crazy about their current setup. Ten minutes later they have a shortlist of two, a side by side comparison, and a pretty firm opinion. And they never visited a single website.

Ido

And if you were the third tool on the market, you didn't just lose a click. You lost the whole conversation, and you never even knew it happened. There's no rank to check. There's no analytics line that says, hey, you were considered and dropped.

Ben cohen

Which is the question for this whole episode. How does a brand stay visible when the thing standing between the customer and the web is an assistant?

Ido

Welcome to Be Visible. I'm Ido Navarro. I come from the product and website building side. WordPress, Elementor, hundreds of websites, and lately agentic experiences where people create and manage a site just by talking to it.

Ben cohen

And I'm Ben Cohen. I come at this from growth. Where do customers come from, and what happens to that when discovery changes? I also built a free tool called the AI Content Reader at GetAISO. It shows you what a website looks like to systems like ChatGPT, Claude, Perplexity, and Gemini, by stripping away the styling and the JavaScript. Fair warning, it can be humbling the first time you see your own homepage that way.

Ido

That's a good place to start, actually. So, what changed? For twenty years, visibility meant being found. Someone typed a query, got ten blue links, and your job was to be as high on that list as possible. It was fairly deterministic. You could say, we're position three for this keyword, and everyone understood what that meant.

Ben cohen

And now the answer is written on the spot.

Ido

Right. The system reads a bunch of sources, synthesizes something, and sometimes recommends a brand by name. And it's probabilistic. Ask the same question twice, you can get two different answers with different brands in them. So being found isn't enough anymore. You need to be understood correctly, considered credible, selected, summarized, and ideally recommended.

Ben cohen

There was a line in a video we looked at about AI visibility tracking tools. The speaker said, and I'm quoting, 'Rank isn't really something for us to look at. We're really looking more at like brand mentions and citations.' Now, that's one practitioner talking about tools, not a law of nature. But it matches what I see. People stop asking where do I rank and start asking, does the AI even mention me?

Ido

And the buyer can do a lot of evaluating inside the interface. Compare, vet, shortlist, without clicking through. That's the zero click part. Not every purchase works that way, obviously. But the more research happens in the assistant, the more your website becomes source material rather than a destination.

Ben cohen

Okay, but I'm going to play the skeptic, because someone listening is thinking it. Is this really different from SEO? Or are we renaming SEO so we can sell new courses?

Ido

Fair challenge. The overlap is huge. A site that's fast, crawlable, helpful, and trustworthy is the foundation for both. If your basics are broken, no new acronym will save you. So SEO isn't dead, and GEO isn't its enemy. GEO adds layers on top.

Ben cohen

Name the layers.

Ido

First, entity clarity. Does the system know who you are, what you do, and who you serve? Second, extractability. Can a model lift a clean answer out of your page without guessing? Third, corroboration. Do other trustworthy places say the same thing about you?

Ben cohen

That third one is where I want to be careful. The idea is that brand mentions across trusted sources shape how AI systems understand a company. It's plausible, and a lot of people in this space believe it. But nobody outside the labs knows exactly how those signals are weighted. So I'd call it a sensible bet, not a proven formula.

Ido

Agreed. Let me make the first layer concrete. Imagine a homepage whose main headline says, empowering your financial journey. It sounds nice. It tells a machine nothing. Now imagine another one that says, we're a bookkeeping and tax firm for freelancers, and here's what we charge. A person might feel the first one is more inspiring. A model can actually use the second one.

Ben cohen

And that's the business problem. Your website has to say what the company does, who it serves, and why anyone should trust it, in plain words, somewhere the machine can find it. Not buried in a hero video.

Ido

Same with content. Direct answers. Clear definitions. Real examples. And original data, if you have any. If you're the source of a number, you're hard to replace. If you're the fifth site paraphrasing someone else's article, you're easy to skip.

Ben cohen

Here's a question I keep asking myself. Are we optimizing for the model, the crawler, or the user?

Ido

The user. Always the user at the center. The machines are intermediaries trying to serve the user. If you make your site worse for humans to please a crawler, you've missed the point. The trick is that the same clarity helps both. Write each section so it still makes sense if someone lifts it out alone.

Ben cohen

Alright, let's get into infrastructure. Where do you start?

Ido

With the least glamorous thing. Your robots.txt file. AI companies run their own crawlers. GPTBot from OpenAI, ClaudeBot from Anthropic, PerplexityBot, and Google Extended for Google's AI use. If you block those, you may have removed yourself from the room before the conversation started.

Ben cohen

And often by accident. A security plugin or a firewall rule gets aggressive with bots, and nobody remembers turning it on.

Ido

Yes. The GEO Lab's WordPress guide specifically warns about security plugins that block bots by default. Go to yoursite.com/robots.txt this week and read it. One caveat, though. Some publishers block these crawlers on purpose, because they don't want their content used for training or answers. That's a legitimate business decision. Just make it knowingly, with the visibility cost in mind.

Ben cohen

Now explain markdown to people who've never opened a code editor, because this is the part that clicks for non technical listeners.

Ido

Okay. A traditional website is a beautifully designed restaurant. The lighting, the furniture, the music, the plating. It's all there for humans, and it matters. Clean markdown is the printed menu. It tells you exactly what's available and what it costs, without making you inspect the decor to figure it out.

Ben cohen

And the real page is the restaurant plus the kitchen plus the electrical wiring.

Ido

Pretty much. The HTML of a modern page carries navigation menus, scripts, tracking code, nested layout containers, cookie banners, repeated interface pieces. A model has a limited amount of text it can hold at once. Every bit of noise takes space away from the actual content. One of the tools in that tracking video, Salespeak, talks about this as token efficiency. The speaker said, 'If you can optimize the tokenization process and limit the amount of work that these tools have to do, that can be a positive signal.' Again, that's a belief about retrieval from one speaker, not something proven.

Ben cohen

This is exactly why I built the Content Reader. You see the page as text. If your pricing lives inside an image, or your main service sits in a slider that needs JavaScript to load, the text view tells you very quickly. It's not about shame. It's about seeing what the machine actually gets.

Ido

I have to confess something here, because the GEO Lab guide lists heavy page builders like Elementor, Divi, and WPBakery as plugins to avoid for blog content, because of code bloat. And I build with Elementor, so I'll take that a little personally.

Ben cohen

Take it personally. We'll wait.

Ido

Ha. But honestly, I think the useful version of that advice isn't about brand names. It's that the output matters. Some builds produce clean, simple markup, some don't, and it depends on how the page was made. The guide's suggestion to use builders for landing pages and cleaner setups for articles is reasonable. The real test is looking at the text version of your page and asking, would I understand this business from this alone?

Ben cohen

So test it, don't guess.

Ido

Right. And this leads to a bigger idea. Eventually the same content may be served in different forms. A rich visual version for humans, and a clean markdown version for machines. The llms.txt proposal even suggests that a page could have a twin at the same address with dot md added to the end.

Ben cohen

Hold on. Isn't serving different things to machines and people exactly what search engines have punished for years? Cloaking?

Ido

Good challenge, and it's a real worry. The principle has to be same facts, different format. The moment the machine version says something materially different from what a human sees, you're in dangerous territory. And to be clear, widespread machine specific versions are still a hypothesis about where things are heading. Today it's an early proposal that some documentation sites follow.

Ben cohen

Which brings us to the file everyone asks about. Does adding an llms.txt file suddenly make a website visible?

Ido

No. And I'll say why in a second. First, what it is. Jeremy Howard of Answer.AI proposed it in September 2024. It's a markdown file at the root of your site. It starts with a title, then a short summary, then sections of links with a line of description for each one. There's also a section literally called Optional, for things a model can skip if it's short on room. Think of it as a curated table of contents, written for a model, by the person who knows the site best.

Ben cohen

And the second file, llms full dot txt?

Ido

That's the full content. Instead of pointing at pages, it puts the actual text of the documentation into one markdown file. So the first file is the index, and the second is the whole book in one place. The index is faster to read. The full version saves a model from fetching page after page.

Ben cohen

So where does it actually work today?

Ido

Developer documentation, mostly. Mintlify added support in November 2024, which switched it on for a lot of docs sites, and Anthropic and Cursor are among the names that publish these files. In coding tools, you can feed that file straight into your environment. It's useful there because a person or a tool deliberately fetches it.

Ben cohen

And for a regular business site? A dentist, a boutique, a SaaS landing page?

Ido

Honest answer, it's unclear. The explainer we read from the Towards Data Science archive notes that public chat assistants don't automatically discover these files the way search engines crawl pages. Often you'd have to hand the file to the model yourself. So I wouldn't promise anyone a ranking lift. It's experimental for general websites.

Ben cohen

So the right framing is, cheap option, not a ranking lever.

Ido

Exactly. It's low risk, it forces you to decide what your most important pages are, and it positions you if support grows. That's a nice exercise by itself. A small site can write one by hand. Platforms can generate and keep it updated. LLMagnet has a generator for that, for WordPress and Shopify sites, and there are open source generators too.

Ben cohen

And the generating isn't the hard part. Deciding what deserves to be on that list is.

Ido

That's the whole thing. A file full of every page you own is just a sitemap with extra steps.

Ben cohen

Okay, schema. People hear structured data and their eyes glaze over.

Ido

The GEO Lab guide has a nice line from its pocket guide, that schema markup is the ID badge your content wears. Instead of a model inferring that this block of text is an article, by this person, published on this date, you state it explicitly. Article, author, organization, product, FAQ, service, review, event, local business. Each one gives a machine context it would otherwise guess.

Ben cohen

And the rule you keep repeating is that it has to match what's on the page.

Ido

Always. If your schema says a product costs forty nine and the visible page says seventy nine, you've created a conflict. Systems that notice conflicting signals may trust neither. The exact mechanics aren't public, but it's the kind of mistake that only creates risk.

Ben cohen

The guide also pushes a few specifics. Include the date modified in article schema, not just the publish date, as a freshness signal. And it says pages with FAQ schema are significantly more likely to be cited. I'd treat that as the guide's claim, not a proven fact. But the logic is believable. A clear question with a short clear answer is easy to lift into a response.

Ido

And the author piece is underrated. The guide calls out a very common WordPress mistake. Posts published under admin. To a person, that's weird. To a system trying to figure out who's an expert, it's an empty name. Real name, real bio, and the same name on your LinkedIn, your bylines, and your schema.

Ben cohen

So what goes wrong with schema in practice?

Ido

Four things. Generic, where every page gets the same boilerplate. Duplicated, which happens when two SEO plugins both add schema. Inaccurate, where it describes things that aren't on the page. And automatically generated without anyone checking it. You can validate with Google's Rich Results Test or validator.schema.org, and you should.

Ben cohen

And the big caveat. Schema doesn't rescue weak content.

Ido

It's an ID badge on a person with nothing to say. The badge helps people understand who they're talking to. It doesn't make the conversation any better.