GEO Basics: Make Your Site Visible to AI Search

Ben cohen

It, it, it hit me when I realized almost nobody opens ten Google tabs anymore when they are looking for B2B software or a local service. They just go straight to ChatGPT or Perplexity or Claude and they ask a single question, and, and whatever three companies get cited in that generated answer... those are the only ones that even exist to that buyer.

Ido

Right, those three companies get the customer before the buyer ever sees a traditional search results page. The whole blue link page with ten search results is basically getting bypassed.

Ben cohen

It is totally bypassed! We are moving from search engine optimization to generative engine optimization, or GEO. And, and the weird part is, most websites are completely invisible to these large language models because of how they are built.

Ido

Wait, why invisible? If a human can open Chrome, go to your site, and read the text, why can't Claude or ChatGPT read it?

Ben cohen

So think about a fancy restaurant menu. To a human walking in, you have the custom lighting, the heavy leather cover, the decorative borders, right? That is your website's JavaScript, your Elementor page builder, your heavy CSS divs. But an AI crawler like GPTBot or ClaudeBot, it does not want the leather cover or the mood lighting. It has a strict token budget. It has to pay compute costs to process every single token of code. So when it hits a site crammed with visual templates, it literally wastes tokens trying to parse where the actual information starts.

Ido

So it is like hand delivering a restaurant menu where ninety percent of the page is decorative wallpaper, and the bot just gives up or de-prioritizes it because the token cost is too high.

Ben cohen

Yes! Exactly! You need to hand the model a clean markdown representation or an llms dot txt file. An llms dot txt file is basically a plain text roadmap sitting at the root of your domain that tells AI bots, here is who we are, here are our core product facts, and here is where the essential documentation lives, without all the HTML junk.

Ido

An llms dot txt file, okay. So you are basically running a site for three different audiences now. You have human visitors who want the slick design, you have interpreting language models like Gemini or ChatGPT that want structured markdown, and then you have autonomous purchasing agents coming down the line that just want raw data to execute a transaction.

Ben cohen

That is the precise tension! And here is something that trips people up: page speed and clean render time are not just nice features for user experience anymore. They are Layer 1 retrieval signals. If your WordPress site takes three seconds to respond because of unoptimized images or missing cache plugins, the AI scraper times out or deprioritizes your content during retrieval augmented generation. If it cannot extract your text fast and clean, you do not get cited in the answer. Period.

Ido

You simply do not exist in the generated response.

Ben cohen

You do not exist.

Ido

Okay, so if someone is sitting there looking at their WordPress or Shopify site right now, what is step one? What is the binary on off switch that determines if an AI can even read them?

Ben cohen

Step one is embarrassingly simple, but so many people miss it. You have to open your robots dot txt file at yoursite dot com slash robots dot txt and check if you are blocking AI crawlers. You need to make sure GPTBot, ClaudeBot, PerplexityBot, and Google Extended are explicitly allowed. A lot of default security plugins or old SEO setups block those bots by default because people thought they were just web scrapers stealing content. If you block GPTBot in robots dot txt, you are voluntarily taking your brand off ChatGPT entirely.

Ido

You are literally locking the front door and then wondering why the AI never cites your product.

Ben cohen

Exactly! Lock the door, key thrown away. Now, once the door is unlocked, step two is schema markup. Schema is like the official ID badge your content wears for AI. Specifically, for GEO, you want Article schema on your posts, but you must ensure the dateModified attribute is active and updating. AI models favor fresh information. Rank Math or Yoast can inject this, but you have to verify it using Google's Rich Results Test tool.

Ido

Wait, dateModified tells the language model that this knowledge is fresh, not something written three years ago and abandoned.

Ben cohen

Right! And the biggest gold mine inside schema is FAQ schema. When you format content as question and answer pairs, you are literally giving the model pre structured answers on a silver platter. Pages with FAQ schema are significantly more likely to be cited in AI answers because the LLM can extract the exact quote without having to re-summarize a giant wall of text.

Ido

So instead of writing a heading like "Our Thoughts on Cloud Storage Security," you change that header to an actual question like "How Does Cloud Storage Encryption Work?" and then put a clear two sentence answer right under it.

Ben cohen

Yes! That formula is Q A E E X: Question header, direct two sentence Answer, Explanation, Evidence, and Example. When Perplexity or ChatGPT searches the web for a user query, it scans for headings that match the user's prompt. If your H2 header is an exact question people ask, and the first two sentences directly answer it, the model can grab that two sentence snippet instantly and cite your domain as the source.

Ido

That makes so much sense. You are structuring the prose so a machine parser can pull a clean vector embedding without extra noise. Now, how do you actually measure if any of this is working? Because traditional rank trackers for Google blue links do not track what Claude is saying about your brand.

Ben cohen

You need dedicated AI visibility platforms. Something like LLMagnet is built specifically for this. It tracks actual AI bot traffic hitting your site, shows you which models are crawling which pages, generates your llms dot txt file automatically, and tracks the exact prompts where your brand gets cited across ChatGPT, Gemini, Perplexity, and Claude.

Ido

LLMagnet gives you an actual AI Visibility Score, right?

Ben cohen

Yeah, it gives you a single metric for your brand's footprint in language models, plus it tracks sentiment and prompt positioning over time. Tools like Otterly dot ai or Scrunch dot ai do similar prompt tracking. But you can also do a free manual audit every week. Take five to ten core prompts that your ideal buyers would type into ChatGPT, run those exact prompts across ChatGPT, Perplexity, and Gemini every Monday, and log in a spreadsheet whether your site was cited, which URL was referenced, or if a competitor was picked instead.

Ido

So you track share of voice inside the AI answer, not just position four on a search page.

Ben cohen

Exactly. AI models are stochastic, meaning they generate responses probabilistically, so rank position fluctuates. What matters is retrievability, brand mentions, and whether your structured data makes you the easy, authoritative source for the model to quote.

Ido

Unblock the bots in robots dot txt, add FAQ schema with dateModified, write direct question H2s with short answer blocks, and track your prompt citations on LLMagnet. That is a pretty clear playbook.

Ben cohen

That is the whole playbook for the generative era. Good chatting with you, Ido.

Ido

Talk soon, Ben.