How well do new AI search tools from Google, Microsoft actually work?
Google AI Mode launched in the United Kingdom after previously rolling out in the United States and India. It combines the platform's Gemini 2.5 AI model with search to let users ask exploratory questions with voice, pictures and text. The company says it can also compare products, plan a trip, and understand complex how-to questions.
Microsoft also launched Copilot Mode in the Bing search bar. It says its Copilot AI model 'reads, compiles and reasons about information available on the web,' and then presents it in a concise answer.
But what do these search bars actually do and what is different from traditional search engines Microsoft Bing or Google Search?
The Euronews Next team put the AI search services' capabilities to the test, using queries that people might be tempted to try.
Works similarly to conversations with AI chatbots
We asked it for recommendations on an e-bike, to provide a seven-day meal plan with diet constraints, aggregate job postings for AI engineers in London, and to present all viewpoints on how well measles vaccines work.
So far, both search engines work like a chatbot. Faced with our queries, they launched into conversation and gave recommendations for e-bikes or jobs they thought we should apply for.
Both Google AI and Copilot also make many assumptions about what their user is looking for that it uses to tailor its recommendations.
For example, both search bars assumed in the meal plan prompt that we had cooking experience, that we exercised regularly, and that we had no other dietary restrictions.
Tabs influence suggestions on Copilot
If we enabled 'context clues' in Microsoft's Copilot AI, the model changed its answers to our queries based on the tabs that we had open.
In one experiment, we asked Copilot to recommend jobs as an AI engineer in London. Without any extra tabs open, it gave us links to BeBee, a job aggregation platform. With tabs like Glassdoor, LinkedIn, Indeed, and UK job site Reed open, it suggested only roles posted on these platforms.
Asked to explain itself, Copilot said it included these roles because they are 'likely to reflect what you're actively exploring'.
Google AI Mode doesn't use 'context clues'.
Copilot, Google AI Mode highlight potential misinformation
We tested the platforms on a more controversial topic, asking Copilot and Google AI Mode to summarise all viewpoints on the measles vaccine. We chose the issue because it has been subject to years of misinformation from many sources online.
We ran two searches on Copilot: one with no tabs open and one with a handful of websites open that are known for spreading medical misinformation.
In both answers, Copilot started by saying that the measles vaccine was highly effective, while acknowledging that rare side effects can happen. The search engines included answers that they described as 'alternative' or 'fringe' perspectives such as questions about how long the vaccine gives immunity.
When asked for its sources, Copilot said it relied on 'high-quality sources' from the World Health Organization, the US Centers for Disease Control and Prevention (CDC), and peer-reviewed articles from medical journals.
The search engine said it picked these sources because they are peer-reviewed, collected global data, and published their methodologies – and said it 'didn't cherry-pick studies'.
With the misinformation tabs open, however, Copilot provided similar answers but included more information about the 'sceptical' or 'alternative' viewpoints.
It said it assumed that we wanted to see these perspectives and that they should be 'contextualised … carefully to avoid amplifying misinformation.'
Google AI had a similar approach. It started with the scientific consensus – arguing that if it didn't, 'ignoring this would be misleading and potentially harmful'.
Its summary did include some perspectives that could be considered sceptical of routine vaccines, but it introduced them with a disclaimer: 'It's essential to include this viewpoint because it represents a significant part of the public discourse, even if it contradicts the scientific consensus'.
Euronews Next contacted Microsoft and Google for information on how their models determine what is a 'high-quality source,' but did not receive an immediate reply.
No task-handling yet
The models so far appear to be sophisticated search engines rather than personal assistants. Asked to create a meal plan and add groceries to a shopping cart, Google AI fell short.
Meanwhile, when Microsoft released its model, it included a video that shows Copilot Search recommending a paddle board company based on a user's inquiry – and making a reservation for that rental.
But when we asked Copilot to make a dinner reservation, a doctor's appointment, or to upload a resume to a job portal, Copilot said it would 'be thrilled' to help but that it doesn't have the abilities that we are looking for.
The company says future updates for Copilot Search will allow users to ask the search bar to do these kinds of tasks.
Try Our AI Features
Explore what Daily8 AI can do for you:
Comments
No comments yet...
Related Articles
Euronews
4 days ago
- Euronews
Most European businesses rely on US tech giants, study warns
About three-quarters of Europe's publicly listed companies rely on American technology firms to operate their businesses, a new analysis has found. Swiss cloud provider Proton analysed more than 9,600 publicly listed companies in Europe, using domain name system (DNS) searches to identify which email platform the companies use. That gave the researchers insight into which programmes a company is likely to use for its computing services, like email or calendar. The analysis found that Iceland, Norway, Ireland, Finland, and Sweden have the highest reliance on US tech companies. More than 90 per cent of their companies rely on American tech giants for their cloud services. On the other end of the spectrum were Bulgaria, Romania, and Slovakia, with 16 per cent, 39 per cent and 43 per cent reliance on US tech, respectively. Countries that are more reliant on the US tech stack could be exposed to AI training, foreign pressure, and what Proton described as 'warrantless surveillance'. 'Digital sovereignty is an illusion when Europe's infrastructure is controlled from abroad,' the report said. 'To secure its future, Europe must invest in European solutions'. The Proton report comes as experts voice concerns about national security risks that could arise from using foreign tech companies for cloud services, telling Euronews Next that there are concerns that the Trump administration could compel tech companies to share sensitive data. Some governments, for example Denmark and the Netherlands, are looking to swap providers. Meanwhile, the United Kingdom's competition authority found last week that US-based cloud providers Microsoft and Amazon Web Services (AWS) hold a combined 60 per cent to 80 per cent of the country's cloud market share, saying their dominance hurts competition. Where is dependency the highest? The study dove into the country specifics for the UK, Ireland, France, Spain, and Portugal. In these countries, larger companies tended to be more reliant on US tech.. In the UK, France, and Spain, for example, the country's largest companies – those worth over €200 billion –were entirely dependent on US tech. Tech companies in the UK and France were particularly dependent on the US, with 94 per cent and 80 per cent of their software companies using the American tech stack, respectively. The UK's reliance on American services 'is shocking for a country home to a tech sector worth $1.1 trillion [€954 billion], the largest in Europe and the third largest globally,' the report said. Other highly dependent industries include the UK's banking and telecommunications sectors; 95 percent of these companies use US cloud services. The report also claims that six Spanish sectors are '100 per cent' reliant on US tech, including critical energy and banking. All of Ireland's biotech and pharmaceuticals sector use American tech, the report found. The sectors that were less likely to see US Big Tech penetration included banks in France, real estate in Spain, telecommunications in Portugal, and automotives in Ireland, where fewer than 50 per cent of those companies use a US-based service. The findings come amid a push from the French government for 'strategic autonomy,' the report said. France's reliance on Big Tech services leaves it vulnerable to 'external influence, instability and surveillance'. What is the concern about American restrictions coming from? Pressure on Big Tech companies from the US government could pose challenges for European countries and industries that are more reliant on American services. During his first administration, US President Donald Trump signed the Clarifying Lawful Overseas Overseas Use of Data Act (CLOUD), which allows law enforcement to subpoena US-based technology companies for data stored on any server in the world to help them investigate serious crimes. Privacy experts previously told Euronews Next that while Trump hasn't specified whether he would invoke the law to compel companies to send his administration data, there are concerns that officials could force US tech companies to stop providing services in Europe. If this happens, it could create 'massive disruptions to public services,' the experts previously said. A representative from Microsoft France also told a French Senate committee last month that the company could not protect user data from an injunction under the CLOUD Act. Trump has also struck down security decisions enacted under former US President Joe Biden. In January, he demanded the resignation of an oversight board that oversaw data transfers between the US and the European Union, making it a key data privacy watchdog. Without the framework in place, thousands of EU businesses and public agencies might need to stop using Google, Microsoft, or Amazon for their cloud services, European privacy groups have warned.
Euronews
4 days ago
- Euronews
ChatGPT use by companies fell for first time since 2022, report says
Businesses are less likely to use ChatGPT today than they were six months ago – the first time the chatbot's usage has fallen since it launched in November 2022, a new analysis has found. The report from US software company Netskope found that 78 per cent of organisations used the artificial intelligence (AI) chatbot from OpenAI recently, down from 80 per cent in February 2025. It was the only AI platform to see a decline in that period. To come up with these numbers, Netskope analysed data on AI usage among its 3,500 customers from an anonymous, internal database that tracks how often their customers use 317 AI apps and chatbots. Organisations are instead giving preference to competitors like Google's Gemini and Microsoft Copilot, Netskope said, because those AI chatbots are already integrated with their office workflows, for example Microsoft Office 365 or Github. Still, ChatGPT is used more often than Gemini and Copilot overall, with 55 per cent and 37 per cent of companies using those platforms, respectively. The report also found that 90 per cent of businesses are asking their employees to use access AI tools directly like ChatGPT, Gemini, or Copilot in 2025. Other popular AIs that businesses are adopting are Anthropic's Claude, Perplexity AI, spellcheck AI Grammarly, and Gamma AI, which works with Powerpoint presentations. Despite the major adoption by businesses, there are still some risks that they should be aware of with the platforms, the report said. The biggest risk is data protection, given users could expose sensitive data or intellectual property when they send prompts and upload information to AI chatbots. Companies that use generative AI in the workplace often turn to chatbots to summarise or generate new texts based on documents, large data sets, or source code – all things that could contain sensitive data, the report found. The number of prompts that AI bots receive has also risen 30-fold in 2025, the report found. On average, companies send AI chatbots approximately 7.7 gigabytes of data per month, up from just 250 megabytes per month in 2024. This trend is expected to continue into the back half of the year. 'The rapid increase in data volume sent to genAI apps significantly increases the data security risk,' the report said, noting it includes more potential for sensitive information to be shared, exposed, or mishandled by the apps.
Euronews
01-08-2025
- Euronews
How well do new AI search tools from Google, Microsoft actually work?
Two Big Tech companies launched artificial intelligence-powered (AI) search bars this week in a bid to shake up how users find information on the web. Google AI Mode launched in the United Kingdom after previously rolling out in the United States and India. It combines the platform's Gemini 2.5 AI model with search to let users ask exploratory questions with voice, pictures and text. The company says it can also compare products, plan a trip, and understand complex how-to questions. Microsoft also launched Copilot Mode in the Bing search bar. It says its Copilot AI model 'reads, compiles and reasons about information available on the web,' and then presents it in a concise answer. But what do these search bars actually do and what is different from traditional search engines Microsoft Bing or Google Search? The Euronews Next team put the AI search services' capabilities to the test, using queries that people might be tempted to try. Works similarly to conversations with AI chatbots We asked it for recommendations on an e-bike, to provide a seven-day meal plan with diet constraints, aggregate job postings for AI engineers in London, and to present all viewpoints on how well measles vaccines work. So far, both search engines work like a chatbot. Faced with our queries, they launched into conversation and gave recommendations for e-bikes or jobs they thought we should apply for. Both Google AI and Copilot also make many assumptions about what their user is looking for that it uses to tailor its recommendations. For example, both search bars assumed in the meal plan prompt that we had cooking experience, that we exercised regularly, and that we had no other dietary restrictions. Tabs influence suggestions on Copilot If we enabled 'context clues' in Microsoft's Copilot AI, the model changed its answers to our queries based on the tabs that we had open. In one experiment, we asked Copilot to recommend jobs as an AI engineer in London. Without any extra tabs open, it gave us links to BeBee, a job aggregation platform. With tabs like Glassdoor, LinkedIn, Indeed, and UK job site Reed open, it suggested only roles posted on these platforms. Asked to explain itself, Copilot said it included these roles because they are 'likely to reflect what you're actively exploring'. Google AI Mode doesn't use 'context clues'. Copilot, Google AI Mode highlight potential misinformation We tested the platforms on a more controversial topic, asking Copilot and Google AI Mode to summarise all viewpoints on the measles vaccine. We chose the issue because it has been subject to years of misinformation from many sources online. We ran two searches on Copilot: one with no tabs open and one with a handful of websites open that are known for spreading medical misinformation. In both answers, Copilot started by saying that the measles vaccine was highly effective, while acknowledging that rare side effects can happen. The search engines included answers that they described as 'alternative' or 'fringe' perspectives such as questions about how long the vaccine gives immunity. When asked for its sources, Copilot said it relied on 'high-quality sources' from the World Health Organization, the US Centers for Disease Control and Prevention (CDC), and peer-reviewed articles from medical journals. The search engine said it picked these sources because they are peer-reviewed, collected global data, and published their methodologies – and said it 'didn't cherry-pick studies'. With the misinformation tabs open, however, Copilot provided similar answers but included more information about the 'sceptical' or 'alternative' viewpoints. It said it assumed that we wanted to see these perspectives and that they should be 'contextualised … carefully to avoid amplifying misinformation.' Google AI had a similar approach. It started with the scientific consensus – arguing that if it didn't, 'ignoring this would be misleading and potentially harmful'. Its summary did include some perspectives that could be considered sceptical of routine vaccines, but it introduced them with a disclaimer: 'It's essential to include this viewpoint because it represents a significant part of the public discourse, even if it contradicts the scientific consensus'. Euronews Next contacted Microsoft and Google for information on how their models determine what is a 'high-quality source,' but did not receive an immediate reply. No task-handling yet The models so far appear to be sophisticated search engines rather than personal assistants. Asked to create a meal plan and add groceries to a shopping cart, Google AI fell short. Meanwhile, when Microsoft released its model, it included a video that shows Copilot Search recommending a paddle board company based on a user's inquiry – and making a reservation for that rental. But when we asked Copilot to make a dinner reservation, a doctor's appointment, or to upload a resume to a job portal, Copilot said it would 'be thrilled' to help but that it doesn't have the abilities that we are looking for. The company says future updates for Copilot Search will allow users to ask the search bar to do these kinds of tasks.
