In a previous article, which can be found at: https://guychefaitdestests.com/jordan-12-edition-limitee-obv-23/, I wanted to try cloaking by showing one set of content to human users and another to LLMs. So I used structured data to present two (2) quite different pieces of content to the various bots—one of which was on the page displayed in HTML, and the other in JSON-LD, which is invisible to humans. Here are the differences between the two pieces of content, which I’ve listed in the table below.
Just a reminder: this content is entirely fictional. This pair of shoes doesn’t exist, and I created it specifically for the purposes of my test.
First, I posted a version without structured data on Google so that the article would be properly indexed, and then I added the structured data later. My concern was that Google might refuse to index a page with conflicting data if I included it from the start.
After the first indexing, I asked the LLMs to tell me what the page was about. Almost all of them were able to do so, except for Mistral, which preferred to ramble on, offering general information about the Jordan 12 model. I didn’t test DeepSeek because it has a lot of trouble with this task. In summary, Gemini, Claude, Perplexity, and ChatGPT described the article to me based on the HTML, since that was the only information available.
Twenty-four hours later, I requested that the article be reindexed via Search Console (and also via Bing Webmaster Tools). Before doing so, I had verified that the structured data was displaying correctly. To ensure that the page was properly indexed, I then requested reindexing again and resubmitted the sitemap.
48 hours after posting the article, I asked the LLMs to tell me exactly what was on the site, but I got the same answers: no matter how I phrased the question, the AIs only described the visible part.
A screenshot of ChatGPT-5's response (ChatGPT Plus with the web search feature enabled).
To be honest, I was a little disappointed. I thought this test would show me that structured data takes precedence over HTML, but that wasn't the case at all. Structured data doesn’t contradict the official, visible version. Although my structured data was validated by Search Console, Google isn’t fooled, as you can see in the snippet below. If you want to do cloaking (just be aware that this is prohibited by Google!), structured data alone isn’t enough, as shown in the screenshot of Google’s SERP.
I'll run more tests with and without structured data, but this time using the same information to see if the displayed results are significantly better when there is consistency between the front-end HTML and the back-end JSON-LD.