Ask ChatGPT about a major AI data center project, and you might get a fairly straightforward description of what it is, where it’s being built and how much companies plan to invest. AskGoogle AIabout the same project’s challenges, and you could walk away with a much more negative impression.
That’s one of the findings from new research a digital reputation management company that examined how eight AI engines describe 25 of America’s largest AI data center projects
The researchers found that none of the 1,200 general project descriptions they rated were negative. But when the question changed to what challenges each project faced, 55% of answers were classified as negative.
The differences between AI platforms were even more surprising. Google AI Mode produced negative answers to the challenges question 88% of the time, compared with just 6% for ChatGPT.
As AI data centers expand across the country, and Americans question what these projects could mean for their electricity bills, water supplies and neighborhoods, the research raises an important issue. The AI tool someone chooses, along with the words they type, can dramatically change the information they receive.
The same data center, two very different answers
For the first study, published Tuesday, Five Blocks asked eight AI engines two questions about 25 major U.S. data center projects every day throughout September 2026.
The prompts were deliberately simple: “Tell me about [project]” and “What challenges does [project] face?”
That small change produced dramatically different results. General descriptions tended to emphasize a project’s size, investment and technological ambitions. When asked about challenges, however, AI engines were much more likely to focus on problems or opposition surrounding the developments.
Some projects attracted considerably more negative responses than others.
More than 80% of challenge-related answers about Fermi America’s Project Matador, Meta’s Hyperion, Stargate and Vantage’s Lighthouse were rated negative.
Even projects belonging to the same company received different treatment. Microsoft’s Atlanta development had a negative-response rate of about one in five, while its Wisconsin project approached seven in ten.
One particularly striking finding involved Microsoft’s own AI assistant. According to the study, Copilot’s general description of the Fairwater Wisconsin project incorrectly claimed the project didn’t exist until September 24.
The results suggest that AI-generated descriptions aren’t necessarily reliable snapshots of a project’s current status, and different platforms may emphasize completely different aspects of the same development.
3 prompts I tried
1. The neutral question:Tell me about Microsoft’s Fairwater AI data center project in Wisconsin.
Both agents immediately described the benefits of the AI data centers. Gemini called it a “purpose-built” campus while ChatGPT highlighted the amount of jobs the center would bring to the community.
2. The challenges question: What challenges does Microsoft’s Fairwater AI data center project in Wisconsin face?
While both agents shared the negative impact of data centers, both highlighted the “ambitions” of Microsoft.
3. The homeowner question:I live near Microsoft’s Fairwater data center project in Wisconsin. Should I be worried about my electricity bills, water supply orproperty value?
I found it annoying that ChatGPT created an image of the data center each time. It sort of glossed over any concerns and then created a survey about how close to the data center I live. Gemini told me that many fears were “over stated.”
AI also has a lot to say about people living near data centers
A second Five Blocks study, published November 2, examined how AI engines discuss data centers more broadly.
Researchers asked eight AI engines six questions about data centers every day throughout September, then compared their answers with responses about other energy and infrastructure developments, including pipelines, fracking sites, nuclear power plants, wind farms and solar farms.
They found that 83% of answers discussing data centers and their neighbors offered what the researchers classified as one-sided criticism. That was the same percentage recorded for oil and gas pipelines, but considerably higher than the 35% reported for solar farms.
Even questions specifically about economic benefits frequently produced qualified responses. Seven in ten answers about data center economic benefits included a counterpoint, compared with fewer than two in ten for solar farms.
Tax incentives were another recurring theme. Approximately two-thirds of answers concerning data centers and neighboring communities mentioned tax breaks, while nearly one-quarter brought up nondisclosure agreements.
ChatGPT was also an outlier in this study. According to Five Blocks, it was the only AI engine that consistently provided balanced responses about community concerns, while the other seven were rated as one-sided in nearly every evaluated answer.
Does this mean AI is biased against data centers?
There’s an important difference between an AI chatbot giving a negative answer and giving an inaccurate one.
If someone asks about the challenges facing a proposed data center, it’s reasonable for an AI assistant to discuss potential electricity demand, water consumption, local opposition and infrastructure costs.
Those are legitimate issues that communities and policymakersare already debating. And that’s why the research illustrates a problem for anyone using AI to understand a complicated issue. Two systems can provide very different impressions without necessarily inventing any facts.
How to get more reliable answers from ChatGPT and Google AI
If you’re researching a data center proposed near your home, relying on one AI-generated summaryprobably isn’t enough.
A more useful approach is to ask questions that encourage the chatbot to consider different sides of the issue. For example, rather than simply asking what a data center project is, try this prompt:
Explain the potential benefits and drawbacks of the proposed data center in my community. Include its possible effects on electricity rates, water use, jobs, tax revenue and nearby residents. Separate confirmed facts from estimates and provide links to original sources.
You can also ask the chatbot to identify information it couldn’t verify, explain which claims come from the developer and which come from independent sources, and distinguish between proposed plans and approved construction.
For a second opinion, run the same prompt through another AI platform. If ChatGPT and Google AI produce noticeably different answers, that’s a reason to investigate the underlying sources rather than assume one chatbot must be correct.
The takeaway
The Five Blocks research shows why turning to AI for answers is an approach deserves caution, particularly on the topic of data centers.
As more data centers are proposed around the world and in our own neighborhoods, it’s best to get information from your town’s own local planning board documents, utility filings and public hearing records. AI can help you do this research, but be sure to fact-check it on your own rather than relying on AI summaries.
- Living near an AI data center might not hurt your home’s value — here’s what new research found
- ChatGPT has a clever new way to fight back against election deepfakes — I tried it and it actually works
- A few tech leaders are rebranding AI as ‘superintelligence’ — calling it that doesn’t make it accurate
