What started as a weekend experiment is now a fully-fledged AI voice assistant on Hearst’s recipe site Delish, helping home cooks follow recipes hands-free.
Hearst’s senior director of AI initiatives, Alexandria Redmon wanted a personal chef assistant — something she could talk to while cooking, without smearing her phone with batter. Using OpenAI’s GPT large language model (Hearst has a content licensing deal with OpenAI), Redmon built the prototype in a few days. It’s aimed at helping to solve everyday kitchen problems: running multiple cooking timers, managing ingredient swaps, and navigating recipes hands-free.
Redmon teamed up with Delish editorial director Joanna Saltz and director of product Ashley Szwec. They spent five months developing the “Cooking Coach” voice assistant, which rolled out across the Delish site this month. It’s trained only on Delish’s content of over 30,000 recipes on the 10-year-old site.
It’s the first AI product for Delish (Hearst’s Good Housekeeping publication developed an AI-powered gift guide last year).
A pop-up on the site invites users to enable their microphone, unlocking access to the Cooking Coach assistant. Once activated, the voice tool helps users choose and follow Delish recipes (“I’m looking for chicken tacos that take less than an hour”). It can also suggest ingredient swaps, adjust serving sizes, set multiple timers and answer questions about cooking techniques and terminology — all in multiple languages. (It took a while to finesse — the original Cooking Coach voice had a “Scottish brogue,” Saltz noted.)
Publishers are ramping up generative AI products — from chatbots to search assistants — to keep readers on their websites for longer, and surface more of their own content on the platforms they own and operate. The push comes as AI products and search engines increasingly siphon off users’ attention, by serving up quick, simplified answers that bypass publishers altogether.
“User expectations are shifting from reading content to relying on AI tools that help them complete tasks step by step. Publishers need to meet that shift by building interactive, utility-driven experiences. Tools like this are one way to do that,” said Josh Jaffe, AI and media consultant and former president of media at the publisher Ingenio.
As Google adds more generative AI summarizations to its search engine, some lifestyle publishers are finding it a uniquely challenging time to keep traffic stable amid declining search referral traffic. Google’s AI-generated summaries can generate full recipes for users, putting recipe sites particularly at risk. Delish has seen “a little bit of an impact” on its traffic from these Google AI features, Saltz said. But she’s not worried.
“Coming out with innovations that are different and interesting and fun and make cooking easier? I think it’s going to ensure that we’re relevant for a long time,” she said.
Voice assistants are getting smarter with the help of generative AI technology. Amazon is courting news publishers for potential AI licensing partnerships to feed quality content into a smarter version of Alexa, Amazon’s voice assistant product, Axios reported last December. The New York Times’ announcement of its recent AI licensing deal with Amazon hints at having its content power the new Alexa+.
Redmon declined to comment on whether Hearst was in talks with Amazon about making its content and AI voice initiatives available through Alexa, though she said she can see the value of device-specific integrations.
“Instead of limiting myself to just folks that have Alexa… I at least wanted to be able to prove out what could be done,” said Redmon. “And so making it a mobile-optimized web app was a way to reach basically everybody in this space at launch.”
The Delish team is tracking engagement metrics to gauge the success of the Cooking Coach product. The Delish team can also track chat logs with Cooking Coach to see how users are interacting with recipe discovery, Szwec added. The team plans to expand features that are popular with users down the line, she said.
“It’s not groundbreaking tech, but it’s a smart start,” Jaffe said. “A next-level version would integrate predictive personalization and contextual tips based on user behavior, turning it into a true personal sous chef. That’s the kind of AI-native product publishers need to be building now if they want to stay competitive.”
Meanwhile, the Delish team is exploring using AI technology to improve the search function on its site. For example, using natural language search to give users a way to find recipes that exclude certain ingredients or use terms like “vegetarian,” even if those terms aren’t in the title of the recipe. They are also looking at developing photo-based search, where a user can take a picture of items in their pantry and ask the cooking assistant what to make.
“This is just the beginning,” said Ronak Patel, gm of the lifestyle group at Hearst Magazines.
Jaffe believes that people will only prefer a publisher’s AI experience over a generic ChatGPT result if the UX, tone and content stand out. “That is where publishers still have an edge,” he added.
More in Media
Creators turn to agentic AI to manage fan engagement
Creators are using AI agents to interact with their fans — and saving time and money as a result. However, marketers have some concerns.
WTF is ‘query fan-out’ in Google’s AI mode?
One major way Google’s AI Mode search feature differs from its traditional search engine is a complex technique called “query fan-out.”
Media Briefing: DoubleVerify casts itself as news ally in Cannes — as scrutiny mounts
At Cannes Lions Festival of Creativity this week, the tech vendor has been working to convince advertisers — and skeptics — that it can help fix the brand safety system, which many say it helped break.