As anyone who has a look at the internet in recent weeks has probably noticed it, Google intermediary upgrade to its search engine had a rocky start. A few days following the launch of the company Answers generated by AI to research requests called IA previewThe functionality was largely mocked to produce erroneous and sometimes crazy answers, such as recommendations for eating rocks or making Pizza with glue.
The new data from the search engine optimization company Brightedge suggest that Google has considerably reduced the frequency to which it has shown people’s overviews since the launch of the functionality, and had in fact already reduced the functionality before the effusion of criticism. The company followed the appearance of Google’s AI responses on results for a list of tens of thousands of samples since the functionality was offered for the first time as a beta test last year.
When AI previews were deployed to American users connected in English after E / S Google conference on May 14Brightedge saw the answers generated by AI-AI on a little less than 27% of the requests he followed. But their presence rushedly decreased a few days later, the week before the screenshots of the errors of the IA previews became viral online. At the end of last week, when Google has published a blog article Recognizing the Flubs of its AI function, Brightdge saw the IA previews appearing on only 11% of the search results pages. Their prevalence was essentially the same on Monday.
Jim Yu, founder and executive president of Brightedge, says that the deposit suggests that Google has decided to adopt an increasingly cautious approach to this deployment. “There are obviously risks they are trying to manage closely,” he says. But Yu adds that it is generally optimistic about the way Google addresses AI’s overviews, and considers these first problems as a “Blip” rather than a functionality.
“We continue to refine when and how we show AI glimpses so that they are as useful as possible, including a certain number of technical updates during last week to improve the quality of the response” said Google spokesperson Ned Adriance. Google has refused to share its internal statistics on the frequency to which the IA glimpses appear in research, but Adriance says that Brightdge numbers do not reflect what the company sees internally.
We do not know why Google may have decided to considerably reduce the appearance of IA glimps shortly after its launch, but the company’s blog post recognized last week that the millions of people used the functionality provided new data on its performance and errors. The company’s research manager Liz Reid said Google had “brought” more than a dozen technical improvements “, such as the limitation of the satirical content of its results. His message noted that these modifications would trigger restrictions at the moment when the IA saws were offered but would not detail how these restrictions would modify the frequency to which the AI results appeared.
Brightdge began to follow the AI previews using its list of requests of requests after Google has enabled users to oppose a beta functionality test at the end of last year. The test requests have lasted nine categories, including electronic commerce, insurance and education, and have been designed to expand on current but also rare research. They were tested over and over again, in some cases several times a day.
In December 2023, Brightdge noted that summaries appeared on 84% of his research, but saw this figure drop over time. Adriance de Google said it did not automatically trigger AI glimps on 84% of research but did not clarify its internal measures. After Google opened AI for everyone, Brightedge continued to follow their appearance using a mixture of accounts that had previously registered in the beta test and others who had not seen significant difference between what the two groups have seen.
Google refused to share exactly how much he changed the number of AI sets that he showed to the general public compared to the people registered in the Beta test, but Adriance said that the people who had opted for the test showed IA seen on a larger range of queries.
Brightedge data also highlights the subjects where Google thinks that AI’s overviews can be the most useful. AI responses have appeared on the majority of healthy health care research research, more recently at a frequency of 63%. The requests of requests included in Brightedge data included “foot infection”, “intestinal bleeding” and “urgent care for tele -ante”. In comparison, requests for electronic commerce refer the IA overviews to around 23%, while restaurants or very rarely travel to the responses to the AI preview.