AI chatbots and search engines are sometimes negative about brands, and the end result—while arguably good for the end consumer—is a wake-up call for companies.

A study of hundreds of millions of prompts across three industries (apparel, electronics, and education) conducted by search engine optimization company BrightEdge found Google’s AI Overviews was 44% more likely to display negative information about a brand than OpenAI’s ChatGPT. Still, when consumers prompted ChatGPT to decide between the two products, the roles flipped, with ChatGPT being more negative.

While the overwhelming majority of responses analyzed in the study were either positive or neutral, a small percentage of responses were negative for both Google AI Overviews and ChatGPT, 2.3% and 1.6%, respectively.

BrightEdge CEO Jim Yu told Fortune while these percentages may seem small, multiplied across hundreds of millions of results, they can still equate to loads of negative queries, which can affect a company’s image in the eyes of potential consumers. For every million queries, an estimated 23,000 would yield a negative response by AI Overviews, based on the data from the study.

Google, in particular, Yu said, is pulling out negative information associated with products that can sometimes be years old because of the way it pulls information from the internet. These searches, though, depend heavily on what people search for and what is publicly available about a company.

“Instead of it being on the back pages that’s way further down, now it’s pulling it into the front page, as people are looking for things about your brand,” he said. “That’s a huge change for businesses.”

A spokesperson for Google told Fortune the report used a flawed methodology to make sensational claims and found a negligible difference of 1% between AI Overviews and ChatGPT in terms of negative responses.

“It also misunderstands how AI Overviews work: They’re based on what sources on the web say about a topic and change depending on what someone is searching for,” said the spokesperson in a statement.

OpenAI did not immediately respond to a request for comment.

To mitigate the negative information being brought to the forefront of AI, companies need to make it a priority to respond to nearly every negative review published by people online, Matt Blumberg, the CEO of Markup AI, a tech company that uses AI to review marketing content, told Fortune.

“I do think it’s more important than ever, because those things are getting picked up more, and they’re getting picked up in different ways by different AI applications,” Blumberg said.

The study shows a clear shift in how AI presents information to people. Consumers are using AI to become better researchers and to get a clearer and arguably more objective picture of the positives and negatives of any product, Yu said.

For companies, this new reality of search means companies need to be pushing out fresh content to cater to AI’s preference for newer content, while also being strategic about where they place it.

“It’s a new dynamic that they do have to really think about,” Yu added.