Perplexity AI and the use of Bing

Perplexity AI is a conversational search engine that combines an AI language model with the power of the Bing search engine to provide accurate, up-to-date, and sourced answers to users' questions​. In other words, it leverages Bing's web index to retrieve relevant information in real time, then integrates these results into the AI-generated response while citing the sources.

Querying Bing via Perplexity

As soon as a user asks a question, Perplexity uses it to perform a web search via Bing. Specifically, the user's query (formulated in natural language) is sent to the Bing search engine, which serves as an information retrieval system for Perplexity. Thanks to the intelligence of the language model, the question can be semantically understood and transformed into appropriate search terms, even if it is formulated informally or complexly. Perplexity thus benefits from Bing’s ability to interpret natural queries and find relevant results beyond mere keyword matching.

Perplexity acts as an intermediary: it understands the intent of the user's question through its language model, then triggers a Bing search in the background corresponding to this intent. This allows it to access Bing’s vast web index without the user needing to leave the Perplexity interface.

Retrieving and Processing Bing Results

Once the query is submitted to Bing, Perplexity follows a series of steps to retrieve and process the results:

1. Retrieving relevant results: Bing returns a list of results (web links) associated with the query. Perplexity analyzes these results and selects the most relevant pages, usually the very first links returned or snippets highlighted by the search engine. By leveraging Bing’s index, the system quickly obtains web pages that are likely to contain the answer to the question.
2. Extracting content from pages: For each selected page, Perplexity retrieves the textual content. This involves crawling the relevant pages via the Bing API to extract the text. Perplexity has also developed its own crawler, PerplexityBot, capable of browsing links and fetching content while respecting websites' robots.txt directives. The raw text of the web pages is then cleaned (to remove HTML code, menus, etc.) and prepared for analysis.
3. Analyzing and filtering information: The collected content is then scrutinized by the AI system. The language model scans the pages and identifies passages that precisely answer the posed question. Perplexity can also classify or reorder the obtained information based on its contextual relevance (re-ranking) to ensure that the most useful and reliable data is highlighted. The goal is to isolate key facts and answer elements present in the documents while discarding off-topic or less reliable sections. At this stage, Perplexity benefits from Bing’s intrinsic ranking (which prioritizes authoritative sources), facilitating the identification of quality content. By combining this filtering stage with the contextual understanding of the LLM, the system synthesizes a set of knowledge ready to be formulated into a response.

Integrating Results into Perplexity’s Responses

After identifying relevant information from web pages, Perplexity integrates these results into the final response provided to the user. This process is carried out via the language model (LLM), which generates a natural language response based exclusively on the previously collected content. During generation, each key piece of information is accompanied by a citation linking to its original source.

Concretely, the response text includes numbered references corresponding to the consulted web pages, allowing the user to verify where the AI found specific information with a single click.

This integration of Bing’s results into the response aims to provide not only a concise answer but also a justifiable and traceable one. Perplexity places a strong emphasis on transparency: the model has been trained to avoid stating facts that are not supported by the retrieved sources. Thus, if information is not found in Bing’s results or within the accessible corpus, the AI avoids making it up. This methodology, inspired by academic citation practices, helps minimize AI hallucinations by strictly linking generated content to verified data.

The result is a conversational-style response that is immediately usable while maintaining the reliability of a traditional search engine, thanks to cited sources.

As the user asks follow-up questions in the conversation thread, Perplexity can conduct new Bing searches while considering the already established context. The system maintains a coherent conversation while updating information if necessary, incorporating newly obtained results into its responses at each iteration—always with supporting citations.

Technical Interaction Between Perplexity AI and Bing

From a technical perspective, the integration between Perplexity and Bing is achieved through the Bing Search API, which allows real-time queries to Bing. When Perplexity receives a question, its backend sends a request to Bing via a REST query to the Bing Web Search API with the keywords from the question. Bing then returns structured JSON results containing titles, snippets, and URLs of the pages. Perplexity can use this data directly: the snippets provided by Bing give an overview of the content, and the URLs allow full text retrieval if necessary. This machine-to-machine communication between Perplexity and Bing is transparent to the user, leveraging Bing’s proven infrastructure for web search.

It is known that in Perplexity’s early days (2022), the platform relied almost entirely on Bing to power its search module, in combination with OpenAI’s models for generating the textual response.

This strategic choice enabled Perplexity to quickly launch its service by benefiting from Bing’s vast index and ranking algorithm rather than immediately building its own web index. This interaction was likely facilitated by Microsoft’s available APIs (Bing Search API) and possibly by a technical collaboration between Perplexity and Microsoft Bing, given the volume of queries involved.

Over time, Perplexity has evolved its architecture to gain greater autonomy. The company has developed its own PerplexityBot crawler and an internal indexing/ranking system, enabling it to enrich or cache certain web data without relying exclusively on Bing.

Nevertheless, Bing remains an important component of its pipeline, particularly for providing comprehensive web coverage and quickly obtaining up-to-date results. In practice, Perplexity likely uses a combination of sources: its in-house index for frequently accessed content and Bing (or other third-party search engines) to supplement missing or very recent information.

The deep integration between Perplexity’s LLM and the search engine is a key technical aspect. The founders have stated that they designed a joint training infrastructure that tightly links the web search module and the language model. This means that the AI model has been trained end-to-end with the search loop included so that it learns how to use Bing’s results to formulate its responses.

This approach improves response relevance and reduces hallucinations since the model learns to rely on actual retrieved data rather than just its internal knowledge. In other words, Bing and Perplexity’s model work hand in hand: Bing provides raw, up-to-date facts, and the AI transforms them into a coherent response.

Every user query on Perplexity triggers a technical synergy between the Bing search engine and Perplexity’s conversational model. Bing provides Perplexity with the richness of its real-time web index, and Perplexity delivers a synthesized natural language response to the user, ensuring that Bing’s results are always transparently (via citations) and reliably integrated into the answer. This fusion of search and language generation allows Perplexity AI to offer an innovative search experience that combines the factual accuracy of a search engine like Bing with the convenience of an intelligent conversational assistant.

At HyperLinker, we ensure that all our exclusive sites are correctly indexed by the major search engines (Bing, Google ...), as well as by the crawlers of the main LLMs, to guarantee you fast and complete indexing!

Sources: allthings.how, ethanlazuk.com, hyscaler.com