• by Ray Schultz , Columnist, July 14, 2025

Publishers that use AI to answer readers queries could be serving up the wrong information, judging by a new study from WordStream. 

Almost all chatbots powered by AI or a large language model have a disclaimer noting that they can make mistakes, the study notes. WordStream decided to test this by asking questions about pay per click (PPC) advertising.  

“Our hypothesis was that LLM AI tools are not always 100% accurate when providing PPC data, strategy, and insights,” Susie Marino writes. 

That was the correct guess.

Of 225 answers delivered during this test, roughly 20% were wrong. And Google’s AI Overviews had the highest level of inaccuracy: 

• AI Overviews—26%
• ChatGPT—22%
• Meta AI—20%
• Perplexity—13%
• Google Gemini—6%

Those numbers are not going to help publishers or advertisers. 

Google’s AI Overviews had the worst performance, with “12 total wrong answers to the 45 queries I asked,” Marino writes..  

Marino adds, “This is concerning since our recent roundup of AIO statistics found that AIOs are rapidly becoming more prevalent on the search engine results page, showing on 55% of searches."

Yet Google’s other AI tool, Gemini, had the best score, giving the wrong answers to only three out of 45 questions asked.

Remember that this experiment concerns only one subject. The results may not be the same for all topics. Be careful where you look. 

“Our study proved that no AI tool is 100% accurate when it comes to PPC advice, strategy, and information,” Marino concludes. “If you ask five different tools the same question, each one will give you a different answer. So, be sure to have a real, reliable source of truth to refer to whenever you’re asking AI about PPC. Industry publications, third-party marketing partners, and more can all help you uncover the real answers to your burning PPC questions.” 

Want the answers on PPC? Click here.