von Tudora Teis
The success of ChatGPT has led to speculation that it could disrupt the search engine market, with several companies developing their own conversational AI services to compete. In this article we explain how the new Bing is leveraging search results to provide a better user experience. The Bing chat orchestration feature presents a combination of organic and paid links to users within the chat interface, without requiring them to leave it. Additionally, shopping ads are also displayed for relevant queries. Furthermore, Microsoft plans to use AI to create personalized advertisements through generative AI, optimizing them for various platforms and devices. The improved accuracy and relevance of Bing's search results could attract users away from Google, especially if integrated with other Microsoft products.
Als zertifizierter Google Ads Premier Partner helfen wir Ihnen das meiste aus Ihren Kampagnen im Such- und Display-Netzwerk, Google Shopping, in Apps und auf YouTube herauszuholen.
Mehr ErfahrenThis article gives an introduction how AI can provide better user experience in Search and how the biggest search platforms work with it.
Since its release in November 2022, ChatGPT has garnered a lot of attention, with many speculating that it could disrupt the search engine market. The recent launch of Bing, powered by OpenAI’s next-generation language model fine-tuned for search, has strengthened this belief.
There are several reasons why a search engine powered by ChatGPT or any other similar language model could challenge Google’s dominance in search. Firstly, these models can understand and interpret natural language more accurately and effectively, leading to more relevant and personalized search results. Additionally, ChatGPT can learn and adapt to user behaviour, preferences, and intent, making search results more contextually appropriate. However, there are some inherent drawbacks in using language models for search, such as factual errors and outdated data. To overcome these challenges, efforts are being made to integrate more up-to-date search results with the language model. By doing so, users can access the most relevant and accurate information, improving their search experience.
Overall, the release of ChatGPT and other advanced language models is likely to shake up the search engine market, challenging Google’s near-monopoly on search. With ongoing efforts to improve the accuracy and relevance of search results, it will be interesting to see how this new technology shapes the future of search.
ChatGPT is an AI-powered tool designed to enable users to create unique text content. By posing questions, offering creative prompts, and utilizing its capabilities, ChatGPT can generate a wide variety of content types, including but not limited to poems, songs, essays, and short stories. Consider the following example where the chatbot is asked to provide ideas for a 10-year old’s birthday:
Example ChatGPT
ChatGPT is estimated to have reached 100 million monthly active users in January, just two months after launch, making it the fastest-growing consumer application in history. Besides the free version, OpenAI also offers a premium version of the service, called ChatGPT Plus, for $20 per month.
Microsoft recently confirmed it is making a “multibillion dollar” investment in OpenAI, the company behind the viral new AI chatbot tool called ChatGPT. The investment gives Microsoft an advantage over its competitors, including Google, Amazon and Meta in the rapidly growing AI industry.
Google has also launched its AI technology Bard recently with the intention of competing with Open AI’s popular language model ChatGPT. Bard is a conversational AI chatbot that not only delivers high-quality responses but also has the ability to simplify complex information. Below is a screenshot of Bard from Google’s announcement showing the chat-based interface:
Example Bard, Source: Google
Google’s Bard operates similarly to ChatGPT as it allows users to interact with the chatbot through conversational means. This new chatbot is built on Google’s Language Model for Dialogue Application (LaMDA), with the company indicating that it is a lighter version of the original. Currently available for testing by a limited group of users, Bard is anticipated to become publicly accessible in the near future.
The hasty revelation and insufficient details surrounding Bard are clear indicators of the „code red“ situation that was activated at Google after the release of ChatGPT last year. Despite ChatGPT’s underlying technology not being groundbreaking, OpenAI’s choice to provide the system as a free online tool exposed countless individuals to this innovative type of automated text generation. The repercussions have been significant, with debates about the impact of ChatGPT on education, employment, and – of specific interest to Google – the future of online search.
In addition, Google unveiled a collaboration with Anthropic, a startup established in 2021 that has developed an AI chatbot called Claude, similar to ChatGPT. Google has allocated a $300 million investment in Anthropic, which will be responsible for taking on ChatGPT while emphasizing the importance of AI safety.
Chinese search giant Baidu aims to introduce a ChatGPT-like AI service that gives users conversational results, according to reports. This service will be built upon the company’s Ernie system, a large-scale machine-learning model that has undergone years of training to excel in natural language generation and understanding.
Meanwhile, Alibaba, another e-commerce giant in China, is also developing a rival to ChatGPT and has confirmed that it is being internally tested at the company. This effort by Alibaba is just one of many examples of tech firms eager to explore the potential of chatbot technology.
Additionally, Amazon’s cloud unit is expanding its partnership with Hugging Face, an AI startup that is developing a ChatGPT alternative. This move is the latest example of the major tech players forming partnerships with allies in the growing market for generative AI systems.
Microsoft has reimagined how users interact with search, browser and chat by pulling them into a unified experience. The new Chat interface in this new Bing is similar to a messaging app, where instead of just seeing a typical search results page, you have a conversational UI. The new UI unifies Search and Chat in a single interface, where users can easily switch back and forth by clicking on UX elements in the page, or by simply scrolling or swiping up and down. You can ask Bing a question, or ask for help planning a trip, for example, and Bing will return a complete itinerary you might want to follow on that trip. You can then also ask Bing to create an itinerary that you can share with your family, and Bing will return a more readable list of places to visit.
Example Microsoft Bing Chat, Source: Microsoft
In the above example, Bing was asked to propose places to visit which are 3 hours flight from London and the chatbot answers with the various options and links to sources for the information.
Along with the revamped chat interface Microsoft applied AI also to the core search algorithm. This had as a consequence that the Bing search ranking engine now provides more accurate and more relevant results even for basic search queries. Moreover, the new version of Bing chat is using a next-generation OpenAI language model that is much more powerful than ChatGPT, and customized specifically for search. It takes key learnings and advancements from ChatGPT and GPT-3.5 and it is even faster, more accurate and more capable.
Prometheus Model Microsoft Bing, Source: Microsoft
Bing is cleverly integrating the search index with the language model by feeding the results from the search to ChatGPT while at the same time taking advantage of the far more frequent updates of the search index (crawling websites) as opposed to the less frequent and very expensive to train large language model. Microsoft calls this the Prometheus Model, which is a layer that interacts with ChatGPT through the search results of the index. This uses the Bing Chat Orchestrator to ingest your long semantic query and sends out multiple searches towards the index as shown in the above picture.
Prometheus relies on grounding, which involves choosing appropriate internal queries and utilizing corresponding Bing search results. This is crucial because it supplies the model with up-to-date and pertinent information, ultimately reducing inaccuracies. In other words, the model uses Bing data to reason through information, and this is facilitated by the Bing Chat Orchestrator. In the last stage, Prometheus supplements the Chat answer with pertinent Bing search responses, which may encompass information on weather, stocks, sports, news, and more. This approach elevates user experience by making it more immersive and capitalizes on Bing’s capabilities. Additionally, through the Bing grounding technique, Prometheus can embed references into the Chat answer, allowing users to access these sources and confirm the information by clicking on them. This is crucial for fostering a thriving web environment and remains one of Bing’s primary objectives.
You can actually see in the user experience the searches that Bing plus Prometheus is issuing on your behalf.
Source: e-dialog
Next, we’ll explain how the new Bing works by providing an example of a hypothetical user searching for cheap flight tickets. As already mentioned above, the Prometheus Model is a layer that interacts with both the language model as well as the search index. It uses the Bing Chat Orchestrator that works as follows: First it converts the natural language search query of the user into a KW search for the index as can be seen from the following screenshot.
Source: e-dialog
Please note that the input prompt could be a longer text by using natural language such as “Could you please provide some non-expensive flight tickets? I’m interested to fly to Frankfurt, Germany”. The parsing of the input query would result in the same KW search towards the search index by ignoring the rest of the information since it is irrelevant for the index.
In order to understand how the search results are combined in the model’s answer, we executed the same keyword search in the Bing Search tab:
From the above image the three top organic results are from Travelocity, Expedia and Skyscanner. If you compare the organic results with the language model’s answer you could easily see that the three options provided by ChatGPT are using as sources the three top organic results.
The second information that the Bing Chat Orchestrator is utilizing is the prompt to the model in order to generate relevant answers. This is achieved by parsing the website (organic result) in order to find the relevant excerpt to the keywords and converting this to a prompt to be given as input to the GPT model together with other relevant information such as geo-location, conversation, etc. This is repeated as many times as necessary. This way the answer will not only contain the reference of the crawled website, but also the text that is relevant to the KW using particular information from the website.
This is how Microsoft describes the process:
“The orchestrator in parallel actually parses the documents that assigns, and identifies important and relevant pieces of information. This orchestrator is, by nature, curious, and if it comes across something particularly interesting or insightful, it actually starts that same cycle again. This virtuous loop results in a package of information that informs and fuels the responses you are seeing. This magic combination of reasoning capabilities with fresh contextual data sparks the innovative and groundbreaking combination that powers the new Bing.”
Next, the GPT model generates the answer and presents it to the user and at the same time annotating the relevant sentences linked to websites. This is invisible unless hovering, a functionality Bing is using in order not to disturb the user, but instead to serve only for trust.
Also, it provides a “Learn more” section with additional links that are not necessarily contributing to the generated answer, maybe to be used later as a new form of advertising. These links are more highlighted and attract the user’s attention. It is worth mentioning that this feature is similar to Google’s NORA (No One Right Answer) queries with a carousel of “Read more” results as shown in the below image as presented during the Google Bard event:
Source: Google
As we’ve already mentioned before, in addition to searching, the Bing Chat orchestrator also inject additional sources of information to help inform the model by pulling in fresh data, news, answers, contextual signals, such as location information and the context of the conversation, to feed into the model to help ground the information that the model is then using to reason over.
The fact that the new Bing can also apparently cite its sources is an important feature, as the inability of language models like ChatGPT to describe where their information is sourced from makes them less reliable. One additional advantage of showing citations besides the credibility, is to monetize on the conversation as we’ll see in the next section.
As we’ve seen above, the Chat orchestration is converting the user’s prompt to a KW search towards the search engine. This search is similar to the traditional way Bing is functioning, meaning together with the organic results, paid advertisement is also presented as can be seen also from the below image – which is the same search as before that we did just to analyze the way the new Bing works.
By hovering over the chat results, it is easy to cross-reference the advertisement with the search results of the index as it appears in the previous image.
Actually, the Bing chat orchestrator is combining one organic with one paid link as reference to each result – bullet point in the answer – where the advertisement is prominently shown. It is worth mentioning at this point that the user is not leaving the chat interface and sees only the conversation as it unfolds. More particularly, you can see that when you hover your mouse over the citation, it loads the sites it gets the answer from but above it, it also shows an ad above those citations.
Microsoft describes the process as follows:
“Microsoft’s ability to understand complex queries in greater depth and precision, as well as the rich insight we will gain into user intent through deep conversational engagement, will deliver smarter and more actionable opportunities for advertisers to engage with customers. With deeply personalized experiences we expect to be able to deliver even more relevant messages to consumers, with the goal of improved ROI for advertisers. And as usage expands, we expect to see greater volume leading to more opportunities for advertisers to reach their desired customers.
The potential of this new technology is still being explored, but we know that this fundamental change of the search engine experience will give us new opportunities to rethink traditional ad experiences – together. We look forward to innovating with you.”
Bing’s improved accuracy and relevance of search results could attract users away from Google. The shift could be more if users feel Bing provides a better experience. Additionally, the integration of Bing with other Microsoft products may also make it more appealing.
Microsoft’s new AI powered Bing is set to launch a new feature called AI Ads, which will utilize artificial intelligence to create personalized advertisements through generative AI. The new feature will use natural language generation and computer vision to generate ad copy and images. The feature will optimize ads for various platforms and devices, such as desktops, mobile phones, and tablets. In short, AI-powered Bing may offer more relevant and personalized search results for users, wooing away a chunk of them from Google’s flagship platform.
Google has already been offering a similar feature called Responsive Search Ads (RSA) since 2018, which uses machine learning to create and test various combinations of headlines and descriptions for ads. However, Microsoft claims that its AI Ads are more advanced and flexible. AI Ads can create ads in different formats, languages, and images from scratch or utilize existing ones from advertisers’ websites or social media accounts.
The introduction of human-like responses through AI technology has the potential to impact various industries and services, including the media sector. The arrival of ChatGPT has already caused a decline in content-writing positions. Furthermore, the optimization of web searches for business reach by individual internet users is expected to undergo changes. AI-powered search engines employ machine learning algorithms to comprehend the context and significance of users‘ inquiries. This will lead to more relevant and personalized search results. Consequently, conventional SEO strategies like keyword stuffing and link building may become less effective, as search engines become better equipped to understand the content of web pages and the intent behind a search query.
The emergence of conversational AI has led to the development of advanced language models like ChatGPT. The rise of this technology has made many speculate that it could disrupt the search engine market, challenging Google’s near-monopoly on search. ChatGPT and other similar language models can understand and interpret natural language more accurately and effectively, making search results more relevant and personalized. However, there are inherent drawbacks in using language models for search, such as factual errors and outdated data. Efforts are being made to integrate more up-to-date search results with the language model to overcome these challenges. Microsoft has recently launched a revamped Bing search engine powered by OpenAI’s next-generation language model, which has strengthened the belief that language models could challenge Google’s dominance in search. Google has responded by launching its own conversational AI service, Bard, while investing in startups such as Anthropic to take on ChatGPT. Other tech firms such as Baidu, Alibaba, and Amazon are also working on ChatGPT-style technology. The future of search is likely to be shaped by these advanced language models, with ongoing efforts to improve the accuracy and relevance of search results.
Wir freuen uns auf Ihre Anfrage und beraten Sie gerne unverbindlich! Füllen Sie dazu einfach das Kontaktformular aus oder rufen uns direkt an.
Jetzt kontaktierenNewsletter
Holen Sie sich unsere Online Marketing-Insights und Trends direkt in Ihr Postfach!