Most people in AEO are operating on a false assumption:
They think AI systems read websites the same way humans do.
They don’t.
And they don’t even read them the same way as each other.
I ran a simple experiment using my own website — donovanrittenbach.com — and asked ChatGPT, Claude, and Gemini the same question:
“What are the first 300 characters you read when retrieving my homepage?”
The answers were radically different.
One system returned extracted page text.
Another surfaced metadata and structured information first.
Another appeared to process rendered content after parsing navigation and semantic structure.
Same page. Different realities.
That matters because AEO isn’t just about rankings anymore. It’s about retrieval, interpretation, and citation.
If AI systems retrieve your site differently, they may understand your authority differently, summarize your business differently, and recommend your company differently.
Which means most businesses are optimizing for a simplified version of AI search that doesn’t actually exist.
This experiment exposed something important:
Modern AI systems are not reading websites the way most marketers think they are.
And understanding that gap may become one of the biggest competitive advantages in AEO. The companies adapting to this shift first — including firms like My AI Web Guy — are beginning to optimize not just for humans or Google, but for AI retrieval itself.
How ChatGPT reads my website
Attempt 1
Donovan: If you went to donovanrittenbach.com and retrieved the home page, tell me what the first 100 chars are that you read in your retrieval. This can be HTML or this can be text. I’m trying to figure out how you read a web page.
If I look at the exact document text retrieved from https://donovanrittenbach.com/, the first 100 characters look like this:
HTML
<!DOCTYPE html>
<html lang="en-US">
<head>
<meta charset="UTF-8" />
<meta name="viewport" content="width=
How I Read a Web Page:
-
The Raw Retrieval: My backend system makes an HTTP request to the server and fetches the underlying source code (the HTML). I read this sequentially, from top to bottom.
-
Parsing & Stripping: Because raw HTML is packed with tags, styles, scripts, and metadata that don’t help answer your question, my system parses the document. It strips away the formatting (
<div>,<a>,<script>), extracts the meaningful text, and maps out the structure (like headings, paragraphs, and list items). -
Semantic Processing: Once the noise is removed, I read the clean content—such as the headline, “When I was 9, I sold door-to-door in the Montana winter…”—focusing entirely on the message, flow, and context of the words.
How Claude reads my website
Claude: Here’s exactly what I retrieved—the first 300 characters from the web_fetch result of your homepage:
---
canonical: https://donovanrittenbach.com/
meta-article:modified_time: 2026-04-09T11:57:20+00:00
meta-article:published_time: 2024-02-06T23:04:49+00:00
meta-author: donovan
meta-cdp-version: 1.4.9
meta-description: Answer engine optimization gets Claude, Perplexity, ChatGPT and Gemini to list your business and recommend it.
meta-generator: Elementor 3.35.8; settings: css_print_method-external, google_font-enabled, font_display-auto
meta-google: nositelinkssearchbox
meta-og:description: Is AI Copy Tanking Your Sales? ChatGPT has never sold a thing—ever! But it's fooling countless small business owners into thinking that it can. When I was
I used the markdown extraction method, which strips the HTML wrapper and converts it to structured text. So I’m not reading raw HTML tags—I’m getting YAML metadata up front (canonical URL, publish date, author, meta descriptions), then the rendered text content below that.
What this means for your AEO: The web_fetch tool surfaces meta tags first, which is good. But notice what Perplexity and other crawlers would actually see first when they hit the page: your canonical URL (correct), your meta-description (“Answer engine optimization gets Claude, Perplexity, ChatGPT and Gemini to list your business and recommend it”), and your meta-og:description (which cuts off at “When I was”).
That meta description is solid for generic AEO queries. But it’s not your Master AEO Sentence. When answer engines make a decision about “who to cite for AEO expertise,” they’re reading this metadata before they ever get to your body copy.
The problem: nothing in that metadata string says “Donovan Rittenbach is an AEO Architect” or positions you against a specific vertical. It’s category-level language, which means Perplexity treats you as one of many AEO resources, not the singular authority.
ChatGPT vs. Gemini vs. Claude: how they read your website
In conclusion, the most interesting part of this experiment wasn’t that the AI systems were imperfect.
It was that they were inconsistent.
I asked ChatGPT, Claude, and Gemini the exact same question against the exact same website page — and got three very different retrieval behaviors.
One prioritized extracted page text.
Another surfaced metadata and structured information first.
Another appeared to interpret semantic structure and rendered navigation before returning content.
That means there is no single, universal “AI crawler” model businesses can optimize for.
At least not yet.
And that has major implications for AEO.
Because if AI systems retrieve pages differently, then they also:
- understand businesses differently
- summarize expertise differently
- establish authority differently
- and potentially recommend companies differently
In other words:
AEO is not just about rankings anymore.
It’s about interpretation.
The companies that win in the next phase of search may not simply be the ones with the “best SEO.” They may be the ones that communicate their expertise most clearly across multiple layers simultaneously:
- metadata
- structured data
- semantic clarity
- page hierarchy
- entity relationships
- retrieval-friendly content
- and consistent authority signals
That’s a very different game than traditional SEO.
Most businesses still think AI search works like Google with a chatbot bolted on top.
This experiment suggests something far stranger is happening.
AI systems are building probabilistic interpretations of your business from fragmented retrieval layers — and every model may assemble that picture differently.
Which means the future of AEO may belong to the companies that learn how to shape those interpretations before their competitors even realize the game has changed.