Release Notes for the 11/11/25 update:
- Branching statements removed for job roles.
- Added web search embeddings via RAG
- Added support for Perplexity and Claude
- Any output from our generative AI tool should be treated as unvetted source material. You should apply your editorial judgment a when considering any information for publication. If you have any doubt at all about the authenticity of source material from our GPT, you should not use it.
- While we’re continuing to test the GPT, we are logging requests and responses and associating them to your originating IP address. We delete this data after 7 days.
Release Notes for the 9/25/25 update:
- The overall context for the GPT contains a branching statement. If your initial reply is “newsroom” the context will be the intersection of AI with journalism and newsroom leadership. If your initial reply is “content marketing” the context will be the intersection of AI with content marketing, etc. There were distinct data training sets for each role.
- The training data consisted primarily of open source research related to the use of AI in newsrooms, and content marketing; a Glossary of AI terms; and a product directory we compiled of AI tools useful for journalists and content marketers plus the short abstracts of 3rd-party content we curate on the site.
- The training data is stored in a Vector data store that is updated daily with the content abstracts we curate for the site.
- Each response includes some recommended search queries. We opted for this rather than citations to specific documents because during our tests of various prompts, hallucinations crept in periodically.
- The “Tools” we provide rely on the same underlying training data, but utilize contexts and directives relevant to the tools’ intended use.
