AI-powered search capabilities are continuing to work their way into toolkits as developers seek to speed up, optimize, and enhance their work.

Fast-growing startup Perplexity has been a key player behind that push, and is now further bolstering its capabilities with a new search application programming interface (API). This gives devs access to the enormous web index, comprising billions of web pages, that powers the company’s answer engine.

“Perplexity opening up its search API is a big deal because it gives developers access to a real-time index of hundreds of billions of pages, something historically locked up by Google and Microsoft,” said Wyatt Mayham of Northwest AI Consulting. “This move turns Perplexity from a consumer‑facing answer engine into a platform, making it part of the search infrastructure itself.”

Ranking snippets for more accurate retrieval

Perplexity’s Search API is built for the unique demands of AI workloads, serving up what the company describes as “rich structured responses” immediately ready for use in AI apps. Its indexing and retrieval infrastructure divides documents into sub-units that are scored against the parameters of the original query. This allows the API to return relevant, ranked snippets.

This component is a stand-out for Mayham. “Instead of returning whole pages, they rank individual passages and blend keyword and semantic signals,” he said. “That means better context for large language models and less messy preprocessing.”

Users can filter data by region or date, and bundle multiple queries under a “straightforward” pricing scheme ($5 per 1,000 requests), he noted. “For devs, it means you can build retrieval‑augmented generation, agentic workflows or search products without scraping or cobbling together third‑party APIs.”

For AI builders, fresh data is everything, he said, and legacy search agent results page (SERP) scraper APIs are either “stale, restrictive, or shut down.” Enterprises spend thousands of dollars a month on web scrapers to get around this.

There are “compelling reasons” why developers would be interested in Perplexity’s Search API, agreed Thomas Randall, research director at Info-Tech Research Group. Eliminating crawling, deduplicating, ranking, and staying compliant with robots.txt (which instructs bots what they can and can’t access) can provide a “huge lift” for devs.

“Perplexity’s solution promises to abstract that away,” he said, describing a “plausible niche” where the Search API could become the default large language model (LLM) retrieval layer for startups and internal tools, especially if it integrates with popular models’ APIs.

Along with the new offering, Perplexity has released a software development kit (SDK), open-source evaluation framework, and a “deep dive” into how they designed and evaluated the Search API.

The company is urging researchers and devs to use that framework, searchevals, to test any publicly available search API. It claims to lead the competition in output quality and latency on single-step search and deep research agentic workflows.

The SDK makes it easy to get started with the new search API, Mayham noted, and structured responses with citations help to avoid hallucinations. He pointed out that Perplexity has promised not to train its models on customer data — which is particularly important for enterprises — and called its open evaluation toolkit “refreshing compared to the black‑box nature of most search APIs.”

Perplexity takes aim at information staleness

Perplexity has prioritized accuracy from the start, the company says, focusing its R&D investments on corroborating answers and sources.

“Our own experience reveals that information staleness is one of the biggest failure modes for AI agents, and we’ve optimized our indexing workflows to make Perplexity a truly real-time assistant,” the Perplexity team wrote in a blog post.

The company says its AI-powered system processes tens of thousands of index update requests a second to provide fresh results. Its content understanding module generates parsing logic to handle the natural messiness of the open web and optimizes itself through iterative self-improvement.

Further, Perplexity has designed its products to be used by both humans and AI agents, and says that its own engineers have employed the search SDK alongside AI coding tools to develop “impressive” product prototypes in less than an hour.

Perplexity further differentiating itself in the AI-powered search race

Since its founding in 2022, Perplexity has continuously rolled out new updates and capabilities, firmly establishing it in the nascent AI-powered search space. The company claims it now processes millions of user queries each hour.

It launched its proprietary search engine, Perplexity Search, in December 2022, not long after the game-changing emergence of ChatGPT. Subsequent rollouts have included Deep Research, which scans the web and creates comprehensive reports for users; Perplexity Labs, which can work autonomously for at least 10 minutes to draft materials; the Comet browser with integrated AI features; and an AI assistant that automates email management.

The timing of the new Search API comes as Google faces antitrust scrutiny, and users are becoming frustrated by the simplistic AI overviews offered up by established search players. Perplexity seemed to make its market intentions perfectly plain when it recently offered to buy Google’s Chrome browser for $34.5 billion.

But will it actually compete with Google (at least at this point)? Probably not, says Mayham.

“I’d frame it more as putting pressure on Google’s grip on search data than a head‑to‑head race,” he said, pointing out that Google still processes billions of queries a day and has a 20‑year head start. “If nothing else, it gives developers a credible alternative and might spur more openness in the market.”

Info-Tech’s Randall agreed that “there are presently no signs” that Perplexity could compete with Google, Microsoft, or others with this new API offering. He pointed out that there isn’t yet any independent benchmarking indicating that it matches Google or Bing’s breadth, latency, or reliability at scale.

“Without published coverage stats or third-party audits, developers are taking Perplexity’s claims on faith,” he said. “While the Search API is not a marketing stunt, it is an untested index.”

Perplexity must articulate what in its Search API is “structurally difficult” for Google or Bing to replicate, he added. Is it a willingness to crawl long-tail sources? Better alignment with developer needs?

Randall also cautioned that, from a company standpoint, Search API may ultimately not be sustainable for Perplexity. “Crawling and indexing at web scale is expensive,” he said. “Even with efficient architecture, bandwidth, storage, and deduplication, the solution would still cost hundreds of millions annually. “