Does Perplexity fetch_url always extract a complete PDF?

Definition: No. Perplexity documents fetch_url as a best-effort tool, and says that PDFs may return limited or unrelated extracted content. Longer fetched content may also be truncated, while scanned PDF pages may contain images without searchable text.

Key facts

• Perplexity describes fetch_url as a tool for fetching and extracting a known page, article, document, or report during an Agent API request. The Fetch URL Content documentation says the returned item contains the requested URL, page title, and extracted snippet.
• The same documentation says the tool is best effort. A completed response can explain which content was available when a URL cannot be fully fetched or extracted.
• For non-HTML content, Perplexity says the tool extracts whatever text the server returns. Its PDF row warns that PDFs may produce limited or unrelated extracted content.
• Fetched content is placed in snippets for model context and may be truncated for longer pages.
• W3C’s PDF7 technique says a scanned PDF can consist of images of text rather than searchable text. OCR can convert those images into actual text, with recognition depending on resolution and text clarity.

What does fetch_url return?

fetch_url is intended for a URL that the application already knows. The response can contain a fetch_url_results item, whose content record exposes the URL, title, and extracted snippet. That snippet is the text available to the model, so the existence of a result record does not show that every page of a document was included.

{
  "type": "fetch_url_results",
  "contents": [
    {
      "url": "https://example.com/report.pdf",
      "title": "Example Report",
      "snippet": "Extracted content from the fetched page."
    }
  ]
}

The tool also reports failure markers in the snippet field for failed items. Perplexity names HTTP client errors, robots-policy blocks, rate limiting, and no_result_returned among the reasons that can appear. These markers are unavailable content rather than document text.

Why can a PDF be incomplete?

Perplexity’s limitation is about extraction, not the PDF file’s page count. A server may return a PDF successfully while the extracted result contains only part of its text or text unrelated to the document. The documentation does not provide a PDF-specific page, byte, or token cutoff. It does state that longer fetched content may be truncated.

PDF structure adds another limit. W3C applies PDF7 to scanned PDF documents and describes a scanned document as images of text that are not searchable text. Its example says OCR converts words and characters to actual text, but text that is not recognized is reported as an OCR suspect. Visual text, tables, reading order, and OCR errors therefore need separate checking when the source matters.

How can you check the extracted content?

Compare the returned snippet with the original PDF. Check the beginning, middle, and end of the document, headings, tables, footnotes, and text embedded in page images. For a scanned file, first confirm that selectable text exists or that OCR has been applied. W3C’s procedure recommends saving the converted document as text and checking that the result is complete and in the correct reading order.

For repeated retrievals, record the requested URL, the returned snippet, and any failure marker. If the document is longer than the available snippet, treat the fetch as partial and use a separate extraction workflow for the sections that the answer needs. Perplexity also allows max_urls from 1 to 10 per tool call, which limits the number of URLs in one fetch operation rather than making a single PDF complete.

Related terms

• How can a Perplexity Agent read a known webpage directly?
• How are inaccessible pages marked in Perplexity fetch results?
• How can URLs fetched per Perplexity tool call be bounded?
• How do I apply noindex to a PDF?

FAQ

Does a successful fetch prove that the whole PDF was read?

No. Perplexity says a PDF can return limited or unrelated extracted content, and longer content may be truncated. A result item proves that an extraction result was returned, not that it covers every page.

Are scanned PDFs searchable by default?

No. W3C describes scanned PDF text as image content rather than searchable text. OCR creates actual text, but recognition can vary with resolution and clarity.

Can fetch_url bypass a PDF login wall?

No. Perplexity documents that fetch_url does not bypass access controls. The content item contains a failure marker when the origin rejects the request or requires credentials.

Sources

1. Perplexity, "Fetch URL Content", 25 September 2026
2. W3C, "Technique PDF7: Performing OCR on a scanned PDF document to provide actual text", 15 July 2025

---

Check how a page reads to answer engines. Open the GEO score analyzer