NinjaAI.com
Welcome to another episode of the NinjaAI Podcast ā your go-to source for staying seen, cited, and successful in the age of AI-powered search.
Todayās episode is all about visibility.
But not the kind youāre used to with Google rankings or PPC ads.
Weāre talking about AI visibility ā specifically, how ChatGPT, Claude, Gemini, and Perplexity actually access your website⦠or donāt.
You see, most business owners think if they publish a page on their site, itāll just magically be picked up by all the major AI models.
But hereās the truth: Thatās not how it works.
We recently ran a controlled experiment to find out exactly which AI platforms are fetching your webpages in real-time ā and which ones are faking it.
What we discovered?
Well⦠it changes everything you thought you knew about AI search.
āø»
So hereās the setup.
We published a brand-new page on our website.
No links pointed to it.
It hadnāt been submitted to Google Search Console.
It wasnāt listed in the sitemap.
It was, for all intents and purposes, a digital ghost ā invisible to traditional search.
Then we prompted five of todayās most popular AI models to summarize the pageās content.
We used:
⢠ChatGPT-4o, OpenAIās flagship
⢠ChatGPT-3.5, aka GPT-o3
⢠Claude Sonnet, from Anthropic
⢠Perplexity Sonar Pro
⢠And Gemini 2.5 Flash, from Google
After prompting each model, we checked our bot tracker ā watching closely to see whether any of them actually visited the page.
Not what they claimed.
Not what they hallucinated.
But what they actually loaded.
Hereās what we found.
āø»
Letās start with Perplexity Sonar Pro.
This one blew us away ā in a good way.
Every time we asked a question, the Perplexity bot loaded the raw HTML of our page. Not the JavaScript, not the images, just the content.
It did this consistently, every time we asked.
Thatās real-time AI search in action.
If youāre looking to show up in AI summaries right away ā especially for fresh or not-yet-indexed content ā Perplexity is the real deal.
But remember: it only sees HTML. No JavaScript, no dynamic elements. If your content relies on scripts to render, youāre invisible to Perplexity.
āø»
Next up: Gemini 2.5 Flash.
This one showed up in our logs using the basic āGoogleā user agent ā yep, just āGoogle.ā
Kind of hilarious.
But like Perplexity, Gemini also loaded just the HTML. No JavaScript. No frills.
Every time we asked it to summarize the page, it made a fresh pull from our server.
So again, if your site is static and optimized for quick rendering, Geminiās seeing what it needs to.
But if your page is built with React or Vue and needs client-side rendering? Geminiās missing the party.
āø»
Then thereās Claude Sonnet, Anthropicās model.
Claude was a bit different.
It loaded the page ā once. Thatās it.
From then on, it used a cached version for follow-up queries.
Which makes sense. Claude seems to be trading freshness for efficiency.
But hereās the catch: if you update your page, Claude wonāt see the new version unless it decides to refresh the cache.
And we donāt control when that happens.
So if your business depends on fast content updates, like promotions or time-sensitive announcements, Claude might not be your best bet for real-time results.
āø»
Now letās talk about ChatGPT-4o, OpenAIās multimodal supermodel.
You might expect that with all its power, itād be doing some real-time magic.
But hereās the bombshell: it never even visited the page.
Not once.
Instead of loading the URL, ChatGPT-4o tried to guess what was on the page by searching Google for the words in the link ā stuff like āblind spot,ā āSEO,ā āprompt fix.ā
It was essentially treating the link like a Google search query ā not a destination to crawl.
Because the page wasnāt indexed yet, GPT-4o concluded that it didnāt exist.
Let me say that again for the folks in the back:
If your content isnāt indexed by Google, ChatGPT-4o canāt see it.
That means itās not crawling your site.