NL API Summarization

by Kris Negulescu

The NL API now has the ability to summarize text. Summarization in the NL API uses the Diffbot LLM for fast inference. There is no truncation. Long documents are broken into chunks and summarized individually. A final summary is generated from these chunk summaries.

See the NL API summarization demo here.

Enhance will now generate Coverage Reports in the background rather than during the execution of the bulkjob to speed up completion of bulkjobs. Previously, users could download the report immediately following completion of the bulkjob. Users should now poll the bulkjob status to ensure that the coverage report is generated. Status is available under content.reports[].status and should have the value of COMPLETE to indicate that the report is generated.

The changes are described here:
*Please note that this change DOES NOT affect the behavior of bulkjob results - only the coverage report.

We have trained a new LLM model for the generation of Organization.summary. We can generate about ~5 million summaries per month using this new approach. We have prioritized private corporation entities to update first. These new summaries began appearing in the KG on July 7th. Please share your feedback and coverage requests with [email protected].

We've updated Diffbot's Privacy Policy.
Effective Date: April 10, 2023

We care deeply about your privacy. Recently, we discontinued all advertising, cookie, and web beacon-based tracking on Instead, we use Plausible Analytics for Visitor analytics and website reporting. Please note: Trial Users & Subscribers who create an account with Diffbot are still required to accept cookies to use the Diffbot Dashboard or Leadgraph web applications.

While we've discontinued advertising, we do plan to continue to send the occasional marketing newsletter to update you on recent product launches and programs. Want to change how you receive these emails? You can update your preferences or unsubscribe from the list by contacting us at [email protected] or by clicking the 'unsubscribe' link in the footer of the newsletter.

Major additions to the Diffbot Privacy Policy include:

  • a supplemental Privacy Policy for CA Residents that can be found here.
  • a supplemental Privacy Policy for EEU Residents that can be found here. Please Note: a section of this policy is dedicated to individuals residing in the European Economic Area (EEA), the UK, and Switzerland.

These additions expand the protections and rights of Users under the California Privacy Rights Act, Cal. Civ. Code §§ 1798.100 et seq. (“CPRA”), EU Data Protection Laws, other laws of the European Union or any Member State thereof, the UK GDPR, and the laws of any other country or state to which you may be subject.

Major edits made to and removals from the policy and additional terms include:

  • an updated list of subprocessors [most notably we added Plausible Analytics and removed all third parties supporting advertising services],
  • an updated Cookie Policy,
  • an updated DPA [contact [email protected] to execute the April 2023 DPA or to amend a DPA we already have on file with your organization], and
  • clarification of data collected from Visitors vs. Trial Users vs. Subscribers.

We pushed several breaking changes to Article data, Job Posts, Discussion Threads, and Product Offers in the KG.


The 'types' array was sorted with the most specific type first. We reversed the order to make the array consistent with 'types' in Organization, Person, and Place entities in the graph.

The 'types' array currently always contains GlobalIndexDiffbotEntity for Articles/Job Posts/Discussions/Product Offers. We have removed that unnecessary 'types' designation, i.e.

"types": [


"types": [

similar to

"types": [

Some 'types' arrays previously contained GlobalIndexComplexTypeWrapper as well. We removed that unnecessary 'types' designation, i.e.

"types": [


"types": [

In the month of March, we rolled to production two key service layer optimizations to the Diffbot Knowledge Graph architecture. Both radically reduce response time to queries and enrichment requests. The first is called the KG Engine, a look-up service that enables us to improve the performance and scalability of both Enhance and DQL and to update the graph between builds for Organization and Person data. The second is a new search indexing layer that aggregates all Articles in the KG that were crawled &/or published in the last six months which reduces response times when querying recent news.

The Diffbot Knowledge Graph now supports NAICS 2022 Classifications (labels and codes) in addition to NAICS 2017 Classifications. See NAICS docs for more details.

The following is a list of arguments you can add to a Diffbot API call in order to reduce processing time when calling the Article or Analyze APIs:

  • Turn off natural language processing to provide sentiment analysis (natural language processing adds the most time to an on-demand extraction call)
  • Turn off entity tagging and linking in text
  • Many publishers of articles provide text without having to render the page. If you need full HTML outputs for each page, you will not want to use this argument.