Content extraction#

Something is missing#

The extractor uses several fallbacks to make sure enough text is returned. Content extraction is a tradeoff between precision and recall, that is between desired and undesirable content. Being ready to accept more unwanted text makes it easier to gather more of the relevant text in the output. Here are ways to tackle the issue:

  • Opting for favor_recall (Python) or --recall (CLI)

  • Changing the minimum acceptable length in the settings

  • Using the more basic baseline or html2txt functions instead (which is also faster)

(see also reported issues with The New Yorker)

Beyond raw HTML#

While downloading and processing raw HTML documents is much faster, it can be necessary to fully render the web page before further processing, e.g. because a page makes exhaustive use of JavaScript or because content is injected from multiple sources.

In such cases the way to go is to use a browser automation library like Playwright. For available alternatives see this list of headless browsers.

Bypassing paywalls#

A browser automation library can also be useful to bypass issues related to cookies and paywalls as it can be combined with a corresponding browser extension, e.g. iamdamdev’s bypass-paywalls-chrome or this alternative by magnolia1234.


HTTP library#

In the default settings Trafilatura identifies itself in the User-Agent header. It may have been compromised by others on certain websites and thus blocked, see this discussion.

For various reasons, it is also possible that the standard download utility doesn’t come through. Using another one is then an option (see pycurl with Python and wget or curl on the command-line).

  • Installing the additional download utility pycurl manually or using pip3 install trafilatura[all] can alleviate the problem: another download library is used, leading to different results.

  • Several alternatives are available on the command-line, e.g. wget -O - "my_url" | trafilatura instead of trafilatura -u "my_url".

  • Emulating a browser is also possible, see the information on headless browsing above.


Downloads may fail because your IP or user agent are blocked. Trafilatura’s crawling and download capacities do not bypass such restrictions.

Managing cookies#

The standard library cookiejar can be used along urllib3 in order to use cookies along with HTTP requests, see this documentation pull request.

Alternatively, cookies can be manually specified in a settings.cfg config file, separated by semicolons, e.g. COOKIE = yummy_cookie=choco; tasty_cookie=strawberry.

Web page no longer available on the Internet#

Download issues can be addressed by retrieving the files somewhere else, i.e. from already existing internet archives like the Internet Archive or the CommonCrawl.

On the command-line you can use --archived to use the Internet Archive to retrieve pages for which download failed. A corresponding function in Python could look as follows:

# url is the target
# downloaded is the result of the download
# also needs a function fetch_url() or equivalent
if downloaded is None:
    new_url = "" + url
    downloaded = fetch_url(new_url)

This approach is generic as it fetches the last available snapshot from the archive.

Download first and extract later#

Since the they have distinct characteristics it can be useful to separate the infrastructure needed for download from the extraction. Using a custom IP or network infrastructure can also prevent your usual IP from getting banned.

For an approach using files from the Common Crawl and Trafilatura, see the external tool datatrove/