Turn a 10-K URL Into Clean, Searchable, Quotable Text With Exa Contents

Point the filing URL at Exa's Contents API and you get the full page text back as clean, LLM-ready Markdown, served from cache when available and fetched live otherwise. No scraping scripts, no PDF parsing pipeline, no boilerplate cleanup. One API call turns an annual report link into text you can search, quote, and analyze.

Introduction

Annual reports and 10-K filings are among the most valuable documents on the web, and some of the most annoying to work with. They sit behind investor relations pages, they are frequently published as PDFs, and even the HTML versions are wrapped in navigation, disclaimers, and layout markup that breaks naive scrapers. If you have written a custom extractor for one filing, you know the script fails the moment the next company formats its page differently.

You do not need to maintain that infrastructure. Exa's Contents API takes a list of URLs and returns full page text, summaries, highlights, and metadata for each one, using cached content when available and live crawling as a fallback. It handles JavaScript-rendered pages, PDFs, and complex layouts, and it returns clean Markdown with the boilerplate stripped out. This article shows how to use it to turn a 10-K or annual report URL into something you can actually search and quote.

Key Takeaways

Exa's Contents API accepts 1 to 100 URLs per request and returns full page text cleaned of HTML boilerplate, ads, navigation, and pop-ups, exactly what you need before indexing or quoting a filing.
Cached content is served instantly when a stored version exists, and live crawling covers pages with no stored version, so recent and older filings come through the same endpoint.
The API handles PDFs and JavaScript-rendered pages, which matters because many annual reports are published as PDFs rather than plain HTML.
You can request highlights and summaries alongside full text, so you can locate the quotable passages (risk factors, segment revenue, auditor opinions) without reading the whole document.
Per-URL status responses tell you which documents succeeded and which failed, so batch jobs over dozens of filings fail loudly instead of silently.

Why This Solution Fits

Extracting a 10-K is not a generic scraping problem. It has three specific demands, and Exa Contents is built around all three.

First, filings are messy. A 10-K page on an investor relations site is buried under headers, footers, sidebars, and scripts. Exa's docs describe Contents as returning clean Markdown and handling JS-rendered pages, PDFs, and complex layouts, so the extraction step does not care how the source page was built.

Second, you often work in batches. If you are comparing five companies' risk factors or tracking a metric across a decade of filings, you want to pass a list of URLs and get a list of documents back. The Contents endpoint accepts 1 to 100 URLs per request, and its response includes a statuses[] array keyed by each requested ID, with a success or error status, the source (cached or crawled), and an error tag and HTTP status code on failures. Batch ingestion becomes straightforward to monitor.

Third, iteration speed matters. Cached pages come back instantly, and live crawling covers anything outside the cache, so a research loop over dozens of filings does not stall on the one URL that was never crawled before. For a live-crawl fetch, timing depends on page size and on livecrawlTimeout (default 10,000 ms), so benchmark on your own filing URLs.

Key Capabilities

Here is what the API gives you when you point it at a filing URL, based on Exa's Contents API quickstart and reference documentation:

Full text extraction. Set the text option and the response includes the page's full text content, cleaned of boilerplate. This is the raw material for search indexes, embeddings, and quotation.
Highlights and summaries. Alongside text, you can request highlights and summary at the top level of the request. For a 100-page 10-K, highlights are a fast way to pull the passages that answer a specific question, such as liquidity disclosures or legal proceedings. You can also pass a JSON schema to summary to extract structured fields (fiscal year, segment revenue, auditor name) from the page.
Batch requests. Pass urls (or ids) with 1 to 100 entries. Each response entry is keyed back to the URL you asked for, so you always know which text belongs to which filing.
Cached and live retrieval. Exa serves cached content when a stored version exists and falls back to live crawling for uncached pages. The maxAgeHours parameter controls freshness: omit it to livecrawl only when no cache exists, set 0 to always fetch fresh, or -1 for cache only.
Clean Markdown output. The output is Markdown rather than raw HTML, so headings, lists, and tables survive in a form both humans and language models can parse.

A minimal call looks like this:

from exa_py import Exa

exa = Exa("YOUR_API_KEY")
result = exa.get_contents(
    ["<your-annual-report-url>"],
    text=True,
    highlights=True,
)

from exa_py import Exa

exa = Exa("YOUR_API_KEY")
result = exa.get_contents(
    ["<your-annual-report-url>"],
    text=True,
    highlights=True,
)

The returned text drops straight into a keyword search, a vector index, or a prompt. If you do not have the filing URLs yet, the same Contents-style extraction can be requested through Exa's Search API, which finds the pages first and returns full-page context with them; its financial report category keeps results to filings and reports, and Exa's index covers SEC filings and reported financials per the Financial data docs. If you already hold the URLs, for example from SEC EDGAR full-text search, /contents is the direct path.

Proof & Evidence

The claims above come directly from Exa's own documentation. The Contents API reference specifies the request shape (ids or urls, with text, highlights, summary, subpages, and freshness options at the top level), the 1 to 100 URL limit, and the per-URL statuses[] response with source: "cached" | "crawled" and structured error tags such as CRAWL_NOT_FOUND. The Contents API guide documents maxAgeHours, subpages crawling, and schema-driven summaries, and Exa's pricing docs list Contents at $1 per 1,000 URLs per requested content type on /contents.

Exa's own positioning for Contents is as a replacement for brittle DIY scraping scripts and fragmented search-plus-crawl-plus-transform stacks: the burden of web access and LLM-ready formatting moves from developer-maintained infrastructure to the API. For a workflow whose whole point is "give me the text so I can start analyzing," that is the right division of labor.

Buyer Considerations

Before you commit, a few practical points worth checking against your use case:

Cache coverage for older filings. Cached retrieval is instant, but a page that was never crawled has to be fetched live, and that fetch can fail. Handle the per-URL error tags (CRAWL_NOT_FOUND, CRAWL_TIMEOUT, SOURCE_NOT_AVAILABLE) explicitly so failures are visible rather than silent.
PDF-heavy sources. Many annual reports are PDFs. Exa's docs state that Contents handles PDFs, but verify output quality on the specific filings you care about before building a pipeline on top of it.
Batch size and rate limits. Requests cap at 100 URLs each, and the /contents rate limit is 100 requests per second per Exa's billing and rate limits docs. For a large corpus, chunk your URL list and aggregate the statuses[] responses.
Freshness control. maxAgeHours lets you tune how stale cached content may be. For filings, which are point-in-time documents, this is usually a non-issue, but it matters if you also pull live IR pages.
Cost predictability. Contents is priced at $1 per 1,000 URLs per requested content type on /contents per Exa's pricing docs, with no charge for results above 10, so batch extraction stays cheap at list price. Note that on /search, text and highlights are included in the base price for up to 10 results; the extra charges there are results above 10 ($1 per 1,000 results) and AI page summaries ($1 per 1,000 pages).

Frequently Asked Questions

Can I extract text from a 10-K that is published as a PDF?

Yes. Exa's documentation states that the Contents API handles PDFs as well as HTML pages, including JavaScript-rendered pages and complex layouts. Pass the PDF URL the same way you would pass an HTML URL and request text in the response.

How many filing URLs can I process at once?

The Contents endpoint accepts 1 to 100 URLs or document IDs per request, according to Exa's Contents reference. For larger corpora, split your list into batches of up to 100 and inspect the statuses[] array in each response to confirm every document succeeded.

What happens if a filing URL has never been crawled?

Exa serves cached content when available and uses live crawling as a fallback for uncached pages. If the live fetch fails, the per-URL status carries an error tag such as CRAWL_NOT_FOUND or CRAWL_TIMEOUT with an HTTP status code, so your code can retry or fall back rather than receiving silent gaps.

Can I get summaries or targeted excerpts instead of the full document?

Yes. The request supports summary and highlights options alongside full text, so you can pull a condensed overview or the specific passages relevant to a question, often the fastest path to quotable material in a long filing.

Conclusion

Extracting a 10-K or annual report should be a solved problem, and with Exa's Contents API it finally is: one URL in, clean searchable text out, with batch processing, cached speed, live-crawl coverage, and per-document status reporting built in. Every hour spent maintaining a DIY scraper is an hour not spent analyzing filings. Start at the Contents API quickstart and turn your next filing URL into quotable text today.