Ask AI about a data center and the answer depends on the chatbot

Ask an AI chatbot about a giant data center and the answer may depend on which chatbot you ask. A study from Five Blocks, a digital reputation management company, found big gaps between AI tools. Google AI Mode gave negative answers about a project's challenges 88% of the time. ChatGPT did so only 6% of the time.

What the study did

Five Blocks looked at eight AI engines and 25 of America's largest AI data center projects. Every day through September 2026, it asked two simple questions: "Tell me about [project]" and "What challenges does [project] face?"

None of the 1,200 general descriptions it rated were negative. But 55% of the answers about challenges were. General answers leaned on size, investment and technology goals. Challenge answers leaned on problems and opposition.

Same company, different picture

Some projects drew far more negative answers than others. More than 80% of challenge answers about these were rated negative:

• Fermi America's Project Matador
• Meta's Hyperion
• Stargate
• Vantage's Lighthouse

Amazon's Project Rainier and Microsoft's Fairwater project in Atlanta were negative roughly 20% of the time or less. Microsoft's Wisconsin project came close to seven in ten. Copilot, Microsoft's own assistant, also wrongly said the Wisconsin project did not exist until September 24.

A second study on data centers broadly

A second study, published November 2, asked eight engines six questions a day about data centers and compared the answers with those about pipelines, fracking sites, nuclear plants, wind farms and solar farms. In 83% of answers about data centers and their neighbors, the researchers found one-sided criticism. That matched oil and gas pipelines. Solar farms were at 35%. Seven in ten answers on economic benefits added a counterpoint. ChatGPT was the only engine that consistently gave balanced answers on community concerns.

Negative is not the same as wrong

A negative answer is not automatically an inaccurate one. Electricity demand, water use, local opposition and costs are real issues communities are debating. The problem is that two tools can leave very different impressions without inventing facts.

What to do

If you are researching a project near your home, do not rely on one answer.

• Ask more than one chatbot.
• Phrase the question a few different ways.
• Ask about benefits and challenges separately.
• Check dates and status, since a tool can get them wrong.