My ChatGPT Visibility Experiment: Does It Use Google Snippets?

A few weeks ago, I came across an experiment by Aleyda Solis that sparked debate across SEO circles. Her claim sounded almost like an OpenAI ChatGPT visibility experiment: ChatGPT may be picking up and reusing Google SERP snippets to generate its answers.

While not an official ChatGPT visibility study by OpenAI, it raised important questions about how do AI tools like ChatGPT gain visibility and what experiments are conducted to measure AI tool visibility.

Many SEOs now ask, “What were the results of the ChatGPT visibility experiment?” or even “How does the ChatGPT visibility experiment affect user engagement?”

Others wonder “How was the ChatGPT visibility experiment conducted?” or even “What are the implications of the ChatGPT visibility experiment?” This blog takes those questions head-on with a controlled test, live data, and clear takeaways.

This blog takes those questions head-on with a controlled test, live data, and clear takeaways.

Naturally, I decided to test the ChatGPT visibility trial myself to reveal how visibility is measured in AI products.

I designed a controlled methodology, verified each step through Google Search Console’s API, and compared ChatGPT’s output using screenshots, validated timestamps, and indexing status. I also used Google’s URL Inspection Tool and Fetch as Google functionality to ensure accuracy.

This blog walks you through:

• What the ChatGPT visibility experiment reveals
• How I replicated the test (with screenshots)
• What ChatGPT reused word-for-word—and what it didn’t
• What these results tell us about LLM retrieval behavior
• Why this matters for featured snippets, on-page SEO, and the broader framework of Generative Engine Optimization (GEO)

Let’s dive into the findings.

My ChatGPT Experiment Reveals Google Snippet Dependencies

While not framed as a formal ChatGPT exposure study, the results echo many concerns raised in the SEO community: ChatGPT’s reliance on snippets makes indexing and snippet optimization critical for visibility.

This experiment uses two of my published blogs to see what really happens when a page is indexed vs not indexed.

So I picked two pages from my own website:

• One that had never been indexed by Google.
• Another that was already indexed and ranking.

The experimental design isolated two variables: indexing status and ChatGPT’s content access capabilities. Building on Aleyda’s initial findings, I designed a controlled test to verify ChatGPT’s dependency on Google’s indexing system. Here’s what I did, and what I found step-by-step:

1. I Prompted ChatGPT Using a URL That Was Not Yet Indexed by Google

I started with Prompts vs Keywords in GEO, a live page from my site that had been published, but never submitted for indexing via Search Console or any search engine.

My goal was to test how ChatGPT would respond when I asked it to fetch content from a live, but unindexed URL.

2. I Confirmed the Page Was Not Indexed on Google

Before prompting, I checked the indexing status of the page in Google Search Console.

As expected:

• No referring sitemaps
• No crawl activity
• Status: URL is not on Google
  

The page existed, but as far as Google’s systems were concerned, it was invisible.

3. I Asked ChatGPT to Fetch the Content Using That URL

Then, I opened ChatGPT and asked:

“Can you fetch content from: https://wellows.com/blog/prompts-vs-keywords/.Tell me what are the main differences between prompts and keywords using this link.”

The response revealed generative search behavior patterns. ChatGPT said it couldn’t locate the exact URL, and that the page appeared to be not publicly indexed or outdated. It did offer a general answer based on other content it had “seen” before—but not from this specific link.

This confirmed one thing: ChatGPT couldn’t fetch or reference a page that wasn’t indexed by Google.

This initial result confirmed my hypothesis about indexing requirements. Next, I tested the opposite scenario with a fully indexed page.

4. Then I Switched to a Fully Indexed Page: “LLM Seeding”

To validate my hypothesis, I repeated the test using a second URL:
 https://wellows.com/blog/llm-seeding/

This page had been live for a few days, and I had already submitted it via Google Search Console.

5. I Verified It Was Indexed in Google Search

Before testing it in ChatGPT, I again used Search Console and verified the indexing status.

• The page was indexed
• It appeared in Google’s search results
• A clean snippet was showing on the SERP

6. I Asked ChatGPT to Fetch the Indexed Page’s Content

Then, I used the same prompt format:

“Can you fetch content from: https://wellows.com/blog/llm-seeding/.Tell me what it says and what is LLM seeding.”

This time, ChatGPT returned a structured summary of the article, complete with formatting and headings pulled straight from the page.

It didn’t link to the URL, but the answer was nearly identical to the Google Search snippet.

7. ChatGPT Pulled Its Answer from Google’s Snippet

When I compared ChatGPT’s output with the actual snippet shown in Google, the match was undeniable. It wasn’t hallucinating the content—it was summarizing the Google snippet.

ChatGPT even explicitly said:

“I was able to fetch a snippet from the publicly indexed version of the page… likely from search engine indexing.”

So the evidence was clear:

•  ChatGPT uses Google Search snippets when it can’t access full content directly.
•  It cannot extract anything from unindexed pages.

These contrasting results establish clear patterns in LLM content retrieval that have broader implications for content strategy, underscoring the need for more formal ChatGPT visibility research by OpenAI in this space.

What Were the Results of the ChatGPT Visibility Experiment?

This experiment showed me something fundamental: ChatGPT doesn’t crawl or cache your content independently. It relies heavily on existing public indexes—especially Google’s—when forming its answers.

If your page isn’t indexed, it’s invisible. If it is indexed, your snippet becomes your voice—because that’s what ChatGPT may end up quoting.

1. Is the ChatGPT Visibility Experiment Focused on SEO?

Yes. The test confirms that SEO is no longer just about rankings or backlinks. It’s about snippet optimization. ChatGPT often summarizes Google’s SERP snippet instead of accessing the full page.

That means your title, description, and first few lines matter more than ever. If those aren’t optimized, ChatGPT may surface a vague or incomplete version of your content.

2. Does the ChatGPT Visibility Experiment Measure User Awareness?

Indirectly, yes. Because if your snippet is visible in Google and ChatGPT surfaces it, that snippet becomes the user’s perception of your brand.
The experiment proved that repeated inclusion in AI answers builds credibility—even without clicks. In generative search, your snippet is your brand’s AI-facing elevator pitch.

What does this tell us about LLM Retrieval Behavior?

This wasn’t just a test of whether ChatGPT “knows” a link. It revealed something deeper about how Large Language Models (LLMs) actually retrieve and assemble answers.

Let’s break down what this means:

1. LLMs Don’t Browse the Web Like Humans

When you ask ChatGPT to fetch content, it‘s not using traditional web crawling like Googlebot or browser rendering engines. Instead, it relies on search API integrations, cached snippet databases, and pre-processed content indexes.

These are maintained by search engines including Google Search, Bing Web Search API, and indexed content repositories. If that page isnt indexed, it might as well not exist. LLMs rely on pre-digested content pipelines whats already been surfaced and structured by search engines like Google.

This makes indexation a prerequisite for generative visibility, something we also explain in detail in Why LLMs Need Context.

2. Google Is the Primary Retrieval Interface

The experiment confirms it: ChatGPT is tapping into Google Search results, not crawling the web directly. It doesn’t build its own link graph or maintain a live index like Googlebot. Instead, it leans on Google’s SERP structure—especially featured snippets and preview text—to construct its answers.

When LLMs say “I found this snippet,” what they often mean is:

“I pulled this from Google’s cached understanding of the page.”

3. SERP Snippets = The New Citations

Your meta description, headline, and first few lines of copy?
That’s the part ChatGPT might “quote” in place of the real thing.

If your snippet is incomplete, vague, or cut off mid-sentence, that’s what the LLM will absorb—and reflect in its output. It’s not pulling your full page; it’s reconstructing meaning from the snippet.

So the way Google summarizes your page becomes your brand’s AI-facing elevator pitch.

4. Retrieval is Memory-First, Search-Second

Unless explicitly forced to trigger a web search, ChatGPT will almost always answer from memory. That memory includes content that was:

• Trained into the model (static memory)
• Seen during prior citations (cached memory)
• Pulled from search snippets (live assist)

The result? Unless your content lives inside one of those three zones, you’re out of the loop.

5. Visibility is Conditional—Not Guaranteed

Getting indexed by Google isn’t a win—it’s a prerequisite. But it’s not enough on its own.

You must:

• Write content that summarizes well in snippets
• Be clear enough to stand alone in 2–3 lines
• Use headings, schemas, and snippet-friendly formatting

Because if Google doesn’t frame it clearly, ChatGPT won’t either. Understanding these retrieval patterns enables strategic optimization for generative search visibility.

Why GEO Strategy Requires SERP Snippet Optimization?

If ChatGPT is reconstructing answers using Google SERP snippets (not your full page), then your Generative Engine Optimization (GEO) strategy needs to reflect this reality. The game isn’t just about being indexed. It’s about being snippeted, surfaced, and selected by AI.

Here’s what to do:

1. Make Your Snippets Count

That little preview text Google shows in SERPs? That’s what LLMs are lifting. So:

• Write compelling first 40–60 words that clearly summarize your core idea.
• Make sure meta descriptions are not just clickbait—but actually reflect the content.
• Use semantic headers (not “Overview”—think “Async Onboarding Frameworks for Remote Teams”).

Why it matters: If your snippet doesn’t explain the value, LLMs have nothing useful to cite.

2. Audit Which Pages Are Indexable

Use tools like Search Console to:

• Confirm index status of key GEO-targeted pages
• Check how they appear in “site:” searches
• Ensure your canonical URLs are set correctly

Why it matters: If it’s not indexed, ChatGPT won’t see it. Full stop.

3. Use Structured Data to Train the Snippet

Help Google help ChatGPT. Use:

• FAQ, HowTo, and Article Schema
• Structured sections (headings, bullets, tables)
• Highlight key takeaways early in the page

Why it matters: Structured content tends to produce better snippets—and LLMs trust that formatting.

4. Strengthen Your Internal Linking

LLMs often pull from site clusters, not just single pages. So:

• Build topic hubs with clear subpages
• Internally link deep content to overview pieces
• Use descriptive anchor text (not just “click here”)

Why it matters: The more semantically connected your content ecosystem, the easier it is for AI to cite you meaningfully.

5. Monitor What ChatGPT Can & Can’t Fetch

Use a test prompt like:

“Summarize insights from [your-page-URL]”

If ChatGPT says it can’t access or only finds a snippet—you’ve got work to do. Consider:

• Refreshing the content
• Improving snippet clarity
• Resubmitting via Search Console

Why it matters: This is your live check on whether you’re GEO-ready.

6. Don’t Just Publish—Seed

Go beyond your site:

• Share in Reddit threads (where ChatGPT listens)
• Mention in Quora or community roundups
• Get cited in niche blogs or trusted media

FAQs

Why it matters: LLMs often reference the most visible version of your content, not necessarily the original.

Bottom line? GEO isn’t just about what lives on your site—it’s about what AI can access, summarize, and trust. That’s why you must optimize like your snippet is the whole story—because, to the AI, it probably is.

This experiment confirms what many SEOs suspected, but the ChatGPT visibility experiment now proves it: ChatGPT pulls from Google’s indexed content and often surfaces the snippet directly in its responses.

Instead of crawling your page live or pulling from your sitemap, ChatGPT relies on what Google already ranks and displays. As a result, if your content isn’t indexed—or if your snippet doesn’t reflect the core value of your page—you remain invisible in generative answers.

Therefore, this shifts the game for content strategy. If you want to appear in AI-driven answers, you must treat snippet visibility as a priority. In today’s search, the snippet is no longer just a preview—it is the content that fuels visibility.

This is your opportunity to get ahead of the curve—because generative search isn’t just about content depth anymore. It’s about what gets surfaced, summarized, and stitched into answers. And that starts at the snippet.