ChatGPT's 'labrador' news engine is really a partner-content pipeline
David Konitzny mapped 43 domains behind the field OpenAI labels as news, and most of them publish recipes, reviews and lifestyle content.
David Konitzny posted on LinkedIn on 27 October 2026 that ChatGPT runs three news engine families: labrador, serpapi_auto_news and mai_grounding_news. Only labrador fills a provider field in the code. Konitzny identified 12 providers covering 43 domains behind it.
The named providers include Dotdash Meredith, Future, Hearst, Condé Nast, Axel Springer, News Corp, Reuters and AP. The domains run from Allrecipes and People to Tom's Guide, WIRED and WELT. Much of it is recipes, product reviews and buying guides rather than news reporting.
How partner pages reach the model
Partner pages are not crawled live at query time. Their URLs arrive through the labrador engine carrying a snippet::summary field. Konitzny describes this as a finished summary generated in the background before ChatGPT uses it.
Regular web pages reach the model as content chunks. Partner pages reach it only as that pre-written summary. In the prompt_string ChatGPT writes from, labrador entries sit at the top, ahead of every web result.
What the summaries actually name
Konitzny analysed 1,000 of these summaries from partner pages containing ranked lists. He checked which entries each summary named by rank position.
- Ranks 1–3: 70%
- Ranks 4–6: 41%
- Ranks 7–10: 23%
- Rank 11 and below: 12%
A product ranked #8 in a partner listicle will usually not appear in the summary ChatGPT sees. The full article still exists on the publisher's site. The model works from the summary.
We think
The label news sits on a pipeline that is mostly commercial editorial content from licensed partners. For practitioners tracking visibility in ChatGPT, this reframes what a citation from one of these 43 domains means. It is placement inside a priority slot that other publishers cannot reach by crawling.
Before acting on this, verify it against your own category. Pull the ChatGPT citations you already track and check how many sit on Konitzny's list of partner domains. Check whether the pages citing you are ranked lists, and where you fall in them.
If partner listicles drive a meaningful share of your citations, the editorial relationship with that publisher matters more than any on-page change you make. Ranking inside the top three of a partner list is a different objective from ranking on the open web, and it is reached through the publisher's editorial process. Treat this as one input. Retrieval from the open web, direct answers and other engines all still run in parallel.