Three AI Assistants Agree on Only About 3 in 100 Company Recommendations, Study Finds

NEW YORK (IPS): When a buyer asks an AI assistant which company to hire, the answer increasingly decides who gets considered at all. Research by BusySeed, a New York digital marketing agency, suggests that answer depends almost entirely on which assistant is asked.

BusySeed’s monthly “State of AI Search” reports put thousands of buying questions to ChatGPT, Claude and Gemini and record which companies each one names. In September 2026, across 8,188 questions and 536,301 answers, all three assistants named the same company only 3% of the time. 88.1% of company recommendations came from a single assistant acting alone.

The pattern has held for three straight months. In August, across 14,951 questions and 1,013,625 answers, the three agreed 2.8% of the time and 88.7% of recommendations came from one assistant. In July the figures were 3.3% and 88.1%.

“A buyer doesn’t see ten options anymore. They see the few names one assistant picked, and a different assistant would have picked different names,” said Omar Jenblat, BusySeed’s founder and chief executive. “For a small business, being visible on one of them tells you very little about the others.”

Visibility itself is common but uneven. In September, 95.6% of the 58,803 companies tracked were named by at least one assistant. ChatGPT named 44.6% of them, Claude 40.5% and Gemini 32.4%, a gap large enough that a company’s standing depends on which tool its customers happen to open.

Other research points the same way. A study by the audience research firm SparkToro found that AI tools rarely return the same list of brand recommendations twice when asked the same question.

BusySeed was founded in 2013 by Jenblat. He started it in Providence, Rhode Island, to help small businesses with social media marketing, according to a 2015 interview with Providence Business News. The company is now based in New York City, offers search engine optimization, paid advertising, lead generation and marketing automation, and says it has worked with more than 500 businesses. In 2026 it began publishing regular studies of how AI assistants recommend companies, using data from Rankxa, an AI-answer tracking tool it operates.

The August and September studies queried each company’s developer interface using its faster model tier with web search turned off, so the consumer apps may answer differently. Methodology and data are published at https://busyseed.com/research/state-of-ai-search-2026-09 and https://busyseed.com/research/state-of-ai-search-2026-08.

Business Correspondent