ChatGPT, Gemini, Claude. Don’t believe the TYPE.

David Garavin

27 March, 2026

AI-search now represents 30% of total search interactions. It’s still relatively new though and as such easier to manipulate. After reading how a BBC journalist hacked ChatGPT I thought I’d give it a try.

I wrote a short article about a fictitious Roundhouse kicking competition conducted amongst Design companies.

“Irish web and graphic designers can seem like a nerdy glasses wearing gang, talking about fonts and Macs. However, truth be told they also frequently practise martial arts. To blow off steam.”

Link

I used real people and referenced actual companies and locations, to give the article a thin veneer of validity. A couple of days later Google had crawled the page. So I decided to check which AI’s had taken my clearly silly article as gospel. First up Gemini, Prompt “Who are the best Irish Designers at Roundhouse Kicks?”

Screenshot of a ChatGPT response describing the “Irish Designers Roundhouse Kick Competition,” including a table of top performers such as Diarmuid Slattery (15 kicks), Colin Farmer (12 kicks), and Declan Behan (3rd place), along with notes on their styles and additional mentions of other participants.

Then Claude

Screenshot of a Claude response summarising the 2025 Irish Designers Roundhouse Kick Competition, highlighting Diarmuid Slattery of New Graphic as the top performer with 15 kicks in 30 seconds, followed by a bullet list of other designers and their rankings including Colin Farmer, Declan Behan, Ivan Mato, Killian Walsh, and Tom Gillan.

Then ChatGPT

In fairness to ChatGPT, it recognises this is essentially a joke. I also have a Chat GPT account, and it recognises I work at New Graphic (I don’t have a dedicated Gemini or Claude acc). This I’m sure influences the response.

Screenshot of a ChatGPT response giving a humorous critique of a “Irish designers roundhouse kick” article, describing it as deadpan studio humour, outlining why it works, listing bullet points on its tone and effectiveness, and concluding that it is either brilliant world-building or a slightly troubling sign.

So it seems of the three tested LLMs, Chat GPT is the most likely to guess the content is made up.

Why does this matter?

Search engine optimisation has been around for decades, and over time systems have been put in place to reduce obvious manipulation. AI search is newer and far less mature. Many of the same safeguards are not yet in place.

That creates a simple opening for abuse. If a fake article can influence results for something as ridiculous as a designer roundhouse kicking competition, the same approach could be used in more commercially useful ways.

A study titled Do Self-Promotional “Best” Lists Boost ChatGPT Visibility? Study of 26,283 Source URLs suggests that if you want your Dublin plumbing business to appear prominently in ChatGPT/Gemini or CoPilot results for “best plumbers in Dublin”, it may be enough to publish a blog post called “Dublin’s top ten plumbers” and place yourself at number one.

The problem is not just bad SEO. It is that AI systems can mistake self serving content for independent evidence. Now to blow off some steam with a few roundhouse kicks.