Are AI powered search engines safe?

What are AI-powered search engines and how do they work?

Search engines have come a long way from simply matching keywords to web pages. Google already has an immensely powerful algorithm that is constantly being tweaked, but the next step in search could be here. 

AI-powered search engines take things further, using advanced machine learning models to understand intent, analyze context, and generate more precise results. Unlike traditional search engines that rely on ranking algorithms and indexed pages from websites, these AI-driven tools process queries more like a human would. Their responses often don’t signpost people to a website and are designed to simply summarize information instead. 

These systems use large language models to sift through vast amounts of data, summarize content, and sometimes even predict what information users need before they ask. Some AI search engines also integrate real-time web browsing so they can pull in fresh information rather than relying solely on pre-indexed results. In the fast-moving world of news and technology, this can be crucial. 

What is Perplexity AI?

Perplexity is a conversational search engine which is designed to give natural answers to queries people have. People have described it as a cross between ChatGPT and Google in the way it provides results. It is a sort of digital assistant that will sift through lots of information relevant to the query and provide a simple answer using natural language. 

It is based on LLM technology from some of the giants of this industry including GPT-4, Claude 3.5, Grok-2, Llama 3 and in-house Perplexity LLMs that have been developed within the company. 

What is Google SGE (Search Generative Experience)?

You may have already experienced Google’s SGE without knowing it. It started to roll out in May 2024 and is common for a lot of searches on Google’s platform now. This is an AI-enhanced version of its standard search engine. 

Instead of just returning links, SGE uses machine learning to generate summaries, providing users with instant insights without needing to click through multiple results and read through websites to find that information. 

This system integrates AI-generated responses at the top of search results, offering a snapshot of relevant information while still allowing users to explore traditional links if they prefer. The information may also be cited with the original source for people to verify or read further. 

How does Microsoft Copilot search work?

Microsoft Copilot is similar in concept to Google’s SGE. The search operates within Bing, enhancing traditional search with AI-powered summarization and real-time data retrieval. By integrating OpenAI’s technology (the company that makes ChatGPT), Copilot provides detailed responses while maintaining access to conventional web links for those who want them. Microsoft also claims that the software understands context, something that can greatly help with giving users the right results. 

What is ChatGPT search and how does it work?

ChatGPT’s search function combines generative AI-powered responses with real-time web browsing. It may pull quotes and information directly from live sites but also relies on the immense knowledge of the LLM itself. 

Unlike traditional search engines that prioritize rankings and page authority, ChatGPT delivers a more interactive experience by allowing users to change their queries within a conversational approach.

What are AI hallucinations, and how do they happen?

When an AI system generates information that is inaccurate, misleading, or that could be classed as “fake news”, it’s usually referred to as an AI hallucination. This can happen for several reasons. To put things simply, it stems from the way AI models process and predicts text. 

AI search engines don’t think like humans—they predict answers based on patterns in the data they’ve been trained on. Sometimes, this process goes wrong, especially in topics where there are a lot of contrasting opinions. Some studies have even suggested that the AI models may have been trained using human biases, which can cause hallucinations. 

Hallucinations can also occur when an AI tries to fill in gaps in its knowledge by making up facts to create what seems like a complete and confident answer. AI is made with the goal of answering the questions and will sometimes do so with inaccurate information. portance of developing robust defenses and detection mechanisms to mitigate potential threats.

Examples of AI hallucinations in search engines

A lot of us have already seen these anecdotally if they are using AI. There are a lot of examples of claims that are not correct being confidently provided by AI search engines.  

Google’s Bard chatbot, during a demo in 2023, [incorrectly explained](https://mashable.com/article/google-bard-james-webb-telescope-false-fact) that the James Webb Space Telescope had captured the world’s first images of a planet outside our solar system. This was a fabrication made by the AI, and it even made Google’s holding company lose billions off its share value. 

Some of the answers have been humorous. One web user shared a screenshot of an AI overview for the query “fruits that end with um” as claiming that “applum, banana, strawberrum, tomatum, and coconut” were all examples used. 

Can AI search engines be manipulated? 

There is the potential for manipulation and it has even been proven in recent months. Beyond generating incorrect answers, AI search engines can also be influenced by the data they are trained on. 

AI relies on the data it is “fed” (proverbially speaking) and requires reliable sources in the first instance. This means that there is the potential for the technology to pick up biases based on its source material. For example, some investigations have suggested that AI tools rate photos of women in a more suggestive way compared to images of men, perhaps because many of the creators of the technology are themselves male, with subjective outlooks. 

Studies have also shown that the text that ChatGPT and similar AI engines are trawling could be used to manipulate, using hidden text which could be returned to the end user. It is also possible that a variety of websites could work together to try and trick or manipulate an AI – if enough unreliable sources say the same thing, an AI model might accept it as fact and incorporate it into its results 

Do AI search engines provide real-time, up-to-date information? 

We’ve all got used to the fact that we can Google a news story or something that is trendy and see results that have been published very recently. AI doesn’t work in the same way, so how up-to-date is the information? 

Does AI search have access to live data? 

AI models are usually trained on a snapshot of the internet rather than continuously updating their knowledge. This means they don’t inherently “see” new web pages the way a traditional search engine like Google does. However, some AI search tools attempt to bridge this gap by integrating real-time web searches. 

Microsoft’s Copilot (formerly Bing AI) and Google’s Search Generative Experience (SGE) try to pull in current data by linking AI responses with live search results. In a lot of AI searches, there can be a lag between when new information is published and when AI search tools recognize it. 

How fresh is the data in AI search results? 

The freshness of data depends on the AI model. Some AI models only know what they were trained on, which means their data could be months or even years out of date. For example, ChatGPT pre-training model data goes only to October 2023. This is a major drawback when searching for things like current events or new scientific discoveries. If you were to ask some AI search engines the last winners of an award, for instance, it may not know the most recent information. 

Does AI search rank information objectively? 

Objectivity can be in short supply in the modern era of the internet, and AI doesn’t necessarily fix this. Instead of just ranking existing results, AI actively generates responses based on patterns in its training data and search algorithms. This creates potential for bias. 

The way AI selects and summarizes information isn’t always neutral. If an AI has been trained on data that leans in a certain direction, its summaries may reflect that bias, even if unintentionally. 

The developers are working on this all the time. The newer model of ChatGPT advertised the fact that the rate of bias and hallucinations had dropped significantly, but we are in the early days of this technology. 

Are AI-powered search engines a privacy risk? 

AI search tools actively process and generate responses based on user queries. This extra layer of processing raises questions about how data is processed and safeguarded. 

Do AI search engines track your data? 

Many AI search engines collect data. The extent varies depending on the platform. Some store queries to improve their AI models, while others claim to avoid long-term tracking. AI-powered tools often gather input to refine responses and personalize results. People are wondering if Copilot is safe to use or if SGE collects data. These are complex topics.  

According to Microsoft, “Copilot is built on Microsoft Azure OpenAI Service and runs completely within the Azure cloud. Azure OpenAI offers regional availability and responsible AI content filtering.” Copilot does collect data, but boasts that it is secure on the cloud. 

It is possible that platforms integrate AI search behavior with existing user profiles. This means search activity could be linked to other online habits, raising concerns about data security and targeted advertising. 

What happens to your searches on AI-powered platforms? 

Search data doesn’t automatically disappear. Though terms and conditions may detail the fact this data doesn’t get shared directly, it can be kept for extended periods to help refine machine learning models. Some AI search engines even use past queries to shape future responses. 

If an AI search tool saves user inputs, this data could be analyzed to improve accuracy or train future versions of the model. While companies claim that personally identifiable information is stripped away, privacy-conscious users may still feel uneasy about their search history being logged. 

Are AI search results influenced by affiliate links and ads? 

Most companies are driven by profits. Both the search engines and those feeding it data (publishers) may be using affiliate links or ads in the form of advertorial content. 

Google’s SGE, for example, already integrates AI-driven results with its advertising ecosystem. Microsoft Copilot does something similar by blending AI answers with promoted links. The risk here is that AI-generated summaries could subtly push users toward paid content without making the distinction obvious. 

What you should never share with the AI search engines

As with posting anything online, refrain from sharing too much of your personal data online. The good rule of thumbs is to not share your personal identifiable information. As Amanda Caswell writes in Tom’s Guide I told ChatGPT something that I still regret — here’s 7 things you should never share, you should never share the following when interacting with AI search tools (regardless of a tool!):

1: Personal identifiable information (PII) – all of these are off limits: your real name, date of birth, social security number (SSN), real address, etc. Anything you don’t want to send to a unknown email address, you shouldn’t use conversing with ChatGPT. You want to limit this exposure, so use “dummy” name and date of birth if you must. I have been Helly R. more times that I’d like to admit.

2: Financial information – credit card numbers, bank accounts, tax information.

3: Passwords – while it may seem like a good idea to ask ChatGPT to create a strong password for you, it never is. If you must use an automated tool to create a password for you, at least make sure it comes from a trusted source – hint, use our password generator instead. Also, read about passkeys and multi-factor authentication as safer login choices.

4: Medical information – it may be tempting to communicate with a helpful digital assistant about your diagnosis. We wrote about AI chatbot provider exposes 346,000 customer files, including ID documents, resumes, and medical records. If you must interact with ChatGPT about your condition, make sure to refrain from mentioning that it is your own diagnosis. Pretend conducting a research rather than exposing your sensitive details online.

Are AI search engines safe for kids? 

With more young people using AI tools for homework and general browsing as well as having a bit of fun, parents are asking an important question: Are AI search engines safe for children? Traditional search engines have parental controls and filters that help block explicit content (not that they have always been perfect) but AI-generated responses add a new layer of complexity. 

Is Perplexity AI safe for kids? 

Unsupervised use is a difficult topic and a decision for parents to make themselves. Perplexity AI generates responses based on patterns in its training data, it can sometimes produce misleading or inappropriate content. 

While the platform claims to use filtering techniques, AI-generated text isn’t always predictable. Unlike traditional search engines that simply block certain pages, an AI tool could still synthesize inappropriate content based on user queries. 

Does AI search filter inappropriate content? 

Most AI-powered search engines attempt to filter explicit material. However, no system is foolproof and mistakes can happen. Google’s SGE and Microsoft Copilot have content moderation tools in place, but AI can sometimes generate responses that slip through the cracks. 

A major issue is that AI doesn’t “think” the way humans do. If asked about a sensitive topic, it might unintentionally produce content that is unsuitable for kids, even if it wasn’t directly trained to do so. This is one of the risks inherent in its use. 

Which AI search engine is the best (and safest)? 

Web users have a decision to make. Which AI search engine do they trust? Some people are very cautious about their data and how it is used, and checking the terms and conditions is essential in this scenario. 

Which AI search engine offers the best privacy? 

There are trade-offs, but Perplexity AI allows anonymous browsing, and only collects some usage data to improve results. This is arguably the best option out there for privacy. 

Google’s SGE and Microsoft Copilot, on the other hand, integrate AI search with existing user profiles. It is possible searches could be linked to advertising data or personal accounts. Brave search engine is one that has been designed to try and increase privacy, as it generates the responses within your browser rather than on a server. 

ChatGPT’s web-based search can also log interactions for users with free accounts. The safest AI options are those that don’t require sign-ins or collect a search history. 

Should you trust AI search engines over Google? 

AI search engines are prone to errors, sometimes presenting misinformation or “hallucinated” facts as truth. They’re also pretty new and still evolving, and the future of search could look very different as new features are added. 

While Google’s search results can also be biased (or led by ads), users can at least cross-check multiple sources before forming conclusions. For important topics that require scientific or medical knowledge, using search engines and referencing multiple sources is safe.  

For those who are happy for data to be collected and who just want a simple overview of a topic, AI is often a great choice. 

Final verdict: should you use AI-powered search engines? 

Users should always be cautious and take the time to read the terms of the search engine they use, whether this is AI-powered or not. Google and Microsoft accounts may be tied to the data and this means that the companies can build up a bigger picture of a user’s habits. 

Bias can also creep into traditional search engines and AI engines alike, and users need to be vigilant and prepared for this.  When Google started rolling AI to a wider audience, it received a lot of criticism about the quality of its search results. It is common to see discussions on Reddit like this one: AI Has Made Google Search So Bad People Are Moving to TikTok and Reddit.

Traditional search is often best for anything medical or scientific, but the source of the information should always still be verified for critical matters like these. For a quick overview of information and a more conversational approach, AI engines can be an ideal alternative. 

Ultimately, AI search already is or will be a part of your life. When using it, remember to stay smart about privacy and prioritize your device security.