The New Era of AI-Powered Search

By Amos Ductan, Ryan Jones, and Sean Stahlman • Posted

Microsoft unveiled a major upgrade to its search engine Bing and a new version of its Edge browser. Now powered by artificial intelligence, Bing promises a more personalized and conversational search experience.

Microsoft and other search engines have long been using AI models to understand queries and boost performance, but this first-ever, consumer-facing AI integration is a transformative moment for our industry.

While many expected Microsoft to simply add ChatGPT to Bing, it went a step further by integrating a different large-language model (also developed by OpenAI) that’s customized for search and said to be even more powerful. If the hype is true, searchers should expect to get more complete answers to complex questions and spend less time curating search results.

Google has also announced the integration of Bard (a simplified version of their AI, LaMDA) into their search engine as a similar, AI-powered chat and answers feature. LaMDA may ring a bell for some of you. Remember the Google engineer who claimed the machine is sentient? (Don’t worry, it’s not.) This stripped down version is also optimized for search.

At the time of this writing, few people have access to these features. Interestingly, smaller search engines like have offered these features for quite a while now.

Why we search

At Razorfish, we believe search is a behavior, not a media channel. As far back as 2009, Bill Gates famously proclaimed that “the future of search is verbs” and we see this daily in our data. People search to complete tasks like planning a trip, researching potential purchases, or learning more about an experience through ratings and reviews.

Integrating AI into search is a natural evolution for meeting these needs, making the process more effective and efficient. Ideally, we’ll free up countless hours spent scanning blue links. Like any good technology, it will give us more time for the things we enjoy and minimize mundane tasks.

For most queries, a website was never what people wanted. It was simply the best that technology could offer at the time. But Google has always aimed to “organize the world’s information.” Its founders hoped the platform would evolve into something like a Star Trek computer.

As of now, we expect AI to have the biggest impact on trivial searches, those easily answered or summarized. When asking “how do I sort a list in python?” users don’t want a website, they want a snippet of code. Websites that don’t evolve may see a decline in clicks similar to when search engines rolled out featured snippets or other search result features leading to “zero click searches.”

For navigational and transactional queries, we don’t see AI making a big dent in total search traffic. As advanced as it is, AI still can’t book a hotel, open a checking account, or sign you up for a service. Searchers will still need and want websites for task-based searches—and this is where focus will be important.

Where things stand

Before we get into tactical implications, it’s important to understand a few basics of just how AI works and differs from traditional search engines. Both Google and Bing use language learning models (LLMs). Unlike a search engine that stores an index of a website’s content, LLMs are “trained” on a large corpus of text and (using some very technical stuff like embeddings and word vectors) store a statistical model of what words are likely preceded and followed by other words across the entire dataset. Essentially, they’re just predicting what words may or may not come next.

Given that training can take months, Bing and Google have been feeding in results they’d normally show and “asking” pre-trained models to “summarize” the results. Instead of content from one website, AI delivers a new and unique answer based on all websites discussing the topic.

With this increased convenience comes big implications for brands.

For brands that rely on SEO, not much changes. Since AI answers are based on summaries of top-ranking results, our SEO goals and tactics remain largely unchanged. The same factors are used to rank websites and our best-in-class strategies continue driving results. Our steadfast aim is to be included in top results.

For companies focused on a more generic content strategy, that is, “common knowledge,” some monitoring and changes may be necessary based on specific data related to those sites.

But for brands and advertisers that rely on paid search as a key, lower-funnel, performance-driving channel, here are three critical implications to consider:

1. Upper funnel ads will be more important

If chatbot features work as intended, we see a future where search categories change dramatically, AI does more heavy lifting, and people interact with fewer links and webpages (and by extension see far fewer ads.)

Imagine, for instance, if a chatbot could plan your big trip to Hawaii. This would eliminate dozens of searches that would normally be part of the process. This makes upper-funnel advertising that much more important. Brands will need to build awareness, so when the chatbot serves up options, customers will choose what’s most familiar.

As a brand leader, you want searchers to know what your products and services stand for, so they’re easy to distinguish when it comes time to buy.

2. Bing broad match will get a boost

Microsoft is applying an AI model that will power chatbot features to its core Bing search ranking engine and is already citing a large jump in relevance. This model will likely improve the performance of broad match for paid search. Google has also made significant improvements to broad match, which have led to substantially better performance outcomes than earlier versions.

This AI model could make Bing’s version of broad match even more effective and, in turn, open up incremental paid search traffic to advertisers. This could help Bing increase its share of paid search budgets relative to Google, which would be a net positive for advertisers.

3. Brand safety is of paramount importance

Microsoft’s press announcement talked at length about “innovating responsibly,” saying it was implementing safeguards to defend against harmful content.

As we know, the internet is full of discrimination, misinformation, and negative content. Because large-language models are merely combing the internet to generate responses, there’s always a risk that harmful content will find its way in. But the fact that Microsoft, Google, and other big players are paying close attention to safety is a strong signal they’ll incorporate robust measures to prevent brands from showing up alongside offensive content.

No system is perfect, of course, but it’s good to know that safety is baked into the process. Below are answers to some common questions our team has been fielding.


Can we block AI from crawling our content?

Not currently. AI isn’t specifically crawling websites. It’s trained on common crawl data or Google’s index or other “bulk” features. The only way to fully block AI from being trained on a website’s content is to block all search engines and bots, which would be devastating to most brands.

Are there copyright concerns?

Not that we’ve seen yet. In our testing, AI answers have been unique content that appears generated like a summary of all relevant websites.

What is the impact on paid search ads?

While it’s still early days, Google and Microsoft have made clear that they see this technology as a complement to the existing experience and have shared their thoughts about the ad experience specifically. This leads us to believe there will be ample paid advertising opportunities although advertisers will need to adapt and account for different types of queries.

Navigating these new frontiers is both exciting and challenging for everyone. If you’re wondering how to make the most of these emerging technologies, our experts can help.

Share this article

You Might Also Like

Ready to learn more?