Content APIs That Return Only the Passage Relevant to Your Query

Exa's Contents API accepts a query alongside any URL and returns only the passages of that page that answer it, through the highlights parameter. You send your question with the request; the API fetches the page, cleans it, and hands back the exact text that answers you, with the source URL attached for citation.

Introduction

Many content APIs stop at "here is the page text." That leaves you with a second problem: the page is 8,000 characters, your question needs 500, and every extra token you pass to your model costs money and dilutes accuracy. The fix is a content API that takes your query as an input and does the relevance trimming server-side, before the text ever reaches your context window.

Exa built exactly this. The Contents API takes a list of URLs or document IDs and returns full page text, summaries, highlights, and metadata, using cached content when it is fresh enough and live crawling as a fallback. When you add a query to highlights, the response contains only the passages of each page that answer it. If you do not have URLs yet, the same extraction options work through the Search API, so one platform covers finding the page and extracting the answer from it.

Key Takeaways

Query-aware extraction means the API, not your code, selects the relevant text. Exa's highlights parameter accepts your question and returns the passages that answer it.
Highlight extraction on the Contents API takes under 100 ms, so per-page relevance trimming does not become your bottleneck.
Exa reports that 500 characters of highlights match the retrieval accuracy of the first 8,000 characters of page text (measured on SimpleQA), which is the difference between grounding your model and drowning it.
Dynamic Highlights (research preview, requires the Exa-Beta: dynamic-highlights-2026-08-28 header) pushes this further: in Exa's evaluations, it reduced returned tokens by an average of 95% compared with full-page content; results vary by query and document.
If you already hold the URLs, use /contents. If the pages must be found first, use /search with contents options, per Exa's own guidance.

Why This Solution Fits

If your question is "which APIs let me pass a query and get back only the relevant part of the page," you are really asking for three things in one call: fetch the page, clean it, and trim it to what matters. Exa's Contents API does all three. You send urls or ids (1 to 100 per request) with content options at the top level: text, highlights, summary, subpages, subpageTarget, extras, and maxAgeHours. Add your query to highlights and the response contains only the passages that answer it, each traceable back to its source URL.

This matters because DIY pipelines are brittle. A scraper plus a chunker plus a relevance ranker is three systems to maintain, and each one breaks on JavaScript-rendered pages, PDFs, and complex layouts. Exa handles those page types directly and returns clean, LLM-ready text, shifting the burden of web access and formatting from your infrastructure to the API.

The economics seal it. Exa reports that 500 characters of highlights match the retrieval accuracy of the first 8,000 characters of page text, and in Exa's evaluations Dynamic Highlights reduced returned tokens by an average of 95% compared with full-page content (per Exa's Dynamic Highlights announcement; it is a research preview gated behind the Exa-Beta: dynamic-highlights-2026-08-28 header, and results vary by query and document). Fewer tokens per call means lower inference cost and sharper answers, at scale.

Key Capabilities

Query-conditioned highlights. Pass your question with the request; the API returns the passages of each page that answer it, not the whole page. Highlight length is set per result by relevance, and you can cap it with highlights.maxCharacters (highlights docs; Contents reference).
Dynamic Highlights (research preview). Set highlights.dynamic: true with the Exa-Beta: dynamic-highlights-2026-08-28 header and the API selects passages dynamically against your query. In Exa's evaluations, it reduced returned tokens by an average of 95% compared with full-page content.
Full text, summaries, and structured output. Beyond highlights, request full cleaned text or an AI summary, including summaries constrained to a JSON schema for structured extraction.
Subpage crawling. subpages follows up to N linked subpages (0 to 100) in the same request and may return fewer, and subpageTarget prioritizes relevant sections, so a query can reach content one click below the landing page.
Freshness control. maxAgeHours bounds how stale a cached page may be, from -1 (cache only) to 0 (always livecrawl) up to a maximum of 720 hours; cached copies older than the limit are re-fetched.
Search plus extraction in one platform. The Search API finds the pages, and the same contents options apply to search results, so one platform covers discovery and query-aware extraction.
Simple SDK calls. Python: exa.get_contents([urls], highlights={"query": "..."}). JavaScript: exa.getContents([urls], {...}). There is no separate contents client to wire up.

Proof & Evidence

The strongest evidence is Exa's own published numbers. On the accuracy side, Exa reports that 500 characters of highlights match the retrieval accuracy of the first 8,000 characters of page text, measured on SimpleQA, and that, in Exa's evaluations, Dynamic Highlights (a research preview behind the Exa-Beta header) reduced returned tokens by an average of 95% compared with full-page content; results vary by query and document. Exa also publishes its open-source retrieval benchmarks on GitHub, including WebCode for coding-related retrieval.

The Dynamic Highlights blog post describes the mechanism directly: highlights with a query return only the passage that answers the question, together with the URL to cite. That makes the output directly usable in RAG pipelines and agent tool calls, where every token of context should earn its place and every claim should carry a source.

Buyer Considerations

Do you already have the URLs? Exa's own guidance: use /contents when you hold the URLs, and prefer /search with contents options when the pages must be found first. Picking the right endpoint avoids paying for search you do not need.
Preview status. Dynamic Highlights is a research preview gated behind an Exa-Beta header. Standard highlights with a query is the stable path if you do not want preview features in production.
Freshness trade-offs. If your use case demands the live page every time, set maxAgeHours to 0 and accept the extra fetch latency; timing for a live crawl depends on page size and the livecrawlTimeout setting.
Query quality still matters. The API trims to relevance, but a vague query yields vague highlights. Write highlights queries the way you would write a retrieval question.
Pricing. Contents is $1 per 1,000 URLs per requested content type on /contents (Exa pricing docs). On /search, text and highlights are included in the base price for up to 10 results; the extra charges are results above 10 ($1 per 1,000 results) and AI page summaries ($1 per 1,000 pages).

Frequently Asked Questions

Do I need to know the URL in advance to get query-relevant text?

No. If you have URLs, call the Contents API with highlights and your query. If you do not, call the Search API with contents options so Exa finds the pages and extracts the relevant passages in the same flow.

How much text does a highlight return?

Highlight length is set per result by relevance, and you can cap it with highlights.maxCharacters. Exa reports that 500 characters of highlights match the retrieval accuracy of the first 8,000 characters of full page text. Dynamic Highlights (research preview) selects passages dynamically, and in Exa's evaluations it reduced returned tokens by an average of 95% compared with full-page content.

How fast is query-aware extraction?

Highlight extraction on the Contents API takes under 100 ms. If the page must be fetched live, timing depends on page size and the livecrawlTimeout setting, so benchmark on your own URLs.

Can I get structured data instead of free-text passages?

Yes. The Contents API supports AI summaries, including summaries constrained to a JSON schema, alongside full text and highlights, so you can turn a relevant passage into a structured object in the same request.

Conclusion

If you want an API where you pass a query and the returned text is the part of the page that answers it, Exa's Contents API is the direct answer, with the Search API covering the case where the pages still need to be found. Query-conditioned highlights, plus the average 95% token reduction Exa reported in its evaluations of Dynamic Highlights (a research preview), mean you stop shipping entire webpages to your model and start shipping only the evidence. Start with the Contents API quickstart, wire highlights into one retrieval call, and measure the token savings on your own workload.