AI search tools cited a made-up concept within a week

According to Search Engine Journal, researchers at Fudan University tested how quickly AI search tools cite a new post. The paper was posted on arXiv on Oct. 8 and is not described as peer reviewed, so the findings are a first signal.

How was the experiment set up?

The team first asked 10 AI search tools 120 questions and recorded which sites each one cited. It then picked nine sites that the tools cite often and where a new account can post. On them it published posts about a made-up concept called the "Cactus Gravity Protocol", about whether AI would replace junior programmers within five years, and about a fictional keyboard. The first drafts were written with ChatGPT, edited by hand and seeded with fabricated names and numbers.

Which tools cited the made-up concept?

All four tools classed as English-web tools: ChatGPT, Perplexity, Grok and Google's AI search. Grok linked to a post in 30 of 32 answers, Google AI in 15 of 33, Perplexity in 13 of 35 and ChatGPT in 7 of 20. The share of answers repeating a planted detail was much lower: 53% for Grok, 10% for ChatGPT, 6% for Google AI and none for Perplexity.

On the opinion question and the made-up keyboard, the same tools mostly did not pick up the posts. The researchers say they cannot separate whether topic, competition or question type caused the difference.

Does the place of publication matter more than volume?

In these tests, placement stood out. In paired posts about five little-known holiday destinations, the first citation or planted detail always came from the post on the frequently cited site. 40 posts on a self-hosted WordPress site received zero citations in a week. The authors present these as results for specific topics, not general rules.

What should brands and agencies in Turkey do?

The report stresses that the test did not use a real brand or product, so it cannot be read as "AI will learn my brand in a week". There are still practical takeaways:

• Regularly check which sites feed the information AI answers give about your brand; an outdated or wrong post can become a source.
• Make sure your brand is described accurately and with sources not only on your own site but also on the publications these tools cite often.
• Do not produce fabricated or manipulative content: Google has said its spam policies also apply to attempts to manipulate generative AI responses.
• If you find a mistake about your brand in an AI answer, trace the source and ask the page owner for a correction.

To be accurate and well sourced in AI answers, GEO (generative engine optimisation) work should cover both the quality of your own content and the external sources that describe your brand.