NYT Cooking keeps AI recipe changes separate from its tested originals

NYT Cooking’s new AI customisation feature lets readers alter ingredients, servings or equipment while keeping the tested recipe alongside the generated variation. Changes appear in orange, and readers can switch back to the original while cooking.[1]

The beta was available to Cooking subscribers on iOS at the time of the launch account by Cooking editor Genevieve Ko, quoted in Nieman Lab’s 28 October report. Ko described a gradual rollout, with the team monitoring performance and the kinds of changes readers requested. Nieman Lab also reported a substantial negative reaction among commenters, alongside readers who welcomed practical help adapting dinner.[2]

The source recipe remains visible

The official help page gives a source hierarchy: the recipe and author’s notes, the Times’s substitution guide, community comments, then AI applying editorial principles. Readers can inspect sources and toggle between versions, with changes highlighted in orange.[1]

The help page says custom versions have not been cooked, tested or reviewed by the Cooking team, nor endorsed by the original creator. They appear in the same trusted app, with a different editorial history.[1]

That is a design issue with consequences beyond recipes. A publisher can own excellent source material and still produce a new answer whose specific combination of changes has never been tried. Showing the source helps a reader understand the basis of an answer. It cannot, by itself, establish that the resulting instructions will work together.

Testing the feature and testing the dish

According to Ko’s account as reported by Nieman Lab, editors evaluated many examples against how they would respond as culinary experts. That is useful product evaluation, but it is a different exercise from cooking every customised recipe. Maintaining that distinction lets the newsroom explain its quality work without lending each generated dish a test-kitchen history it does not have.[2]

Keeping both versions visible gives readers a way to inspect the exact changes instead of relying on a general assurance about AI. It also preserves the distinction the product needs to communicate: the original has been tested in a kitchen; a particular generated variation has not.

Report an error or suggest a correction