Ask AMO: Are Readers Actually Using AI Chatbots?

Since the emergence of OpenAI’s ChatGPT in late 2022, a flurry of media publishers has gone on to launch their own AI chatbots and answer engines, but are readers actually using them?

Well, sort of.

Travel publication Skift, which was one of the earliest adopters of an answers engine at the beginning of 2023, answers around 10,000 queries a month, which is more than triple the amount of questions the tool was answering prior to a redesign in September, said Jason Clampet, chief product officer at Skift, in an interview with A Media Operator.

Most of the time Skift readers will start with a recommended prompt, and increasingly they are asking questions in pairs and triplets, Clampet said.

“It might be like five Airbnb questions in a row, and that’s the same user building and building and building, which is what we really want to see,” said Clampet, noting that most questions are topical such as the impact of the Trump administration and tariffs on the travel industry. Skift recently rolled out Ask Skift across all its sites such as Airline Weekly, Daily Lodging Report and Skift Research, enabling the chatbot to provide more detailed responses.

Nursing Times, a title from media publisher Emap which rolled out an answers engine in February 2024, is answering around 10,000 to 15,000 questions a month and recently passed a total of 200,000 answers since launch, Robin Booth, managing director at Emap, told AMO.

“They are definitely really using it,” Booth said. “What we haven’t seen, which I was nervous of at the beginning, was something I’d call a curiosity curve… your audiences start playing around with it and then they just stop using it and they go back to their old habits—we’ve seen very consistent usage over time.”

Around 80% to 90% of the usage comes from the recommended questions inserted into articles and the remainder is user generated, Booth said. This behavior is consistent across seven of Emap’s sites, which already have the engine including Cyclist and Architects’ Journal, he said. He’s looking to redesign the user experience to encourage users to ask their own questions.

“Even the best [questions] are still quite like a traditional Google search,” Booth said. “You’re not yet seeing the audience really understand how detailed they could make that question for the LLM to respond.”

The Financial Times, which launched its answers engine to FT Professional readers using a staggered rollout with a beta launch in March 2024 and to all professional readers in April 2025, doesn’t provide readers with sample prompts but does provide hints at what the tool is good at. A total of 56,000 questions have been answered since beta launch.

“People do tend to sort of take some notice of [the pointers] and use it for those use cases, but we do push them to ask their own questions,” Zoe Coutinho, product director for FT Professional, told AMO. “If we were looking at rolling it out in other parts of the FT, [prompt questions] may be one thing we think about … but it’s not something we do currently.”

Building With a Business Need in Mind 

FT Professional readers are asking the site to tailor the day’s news toward their job, summarize events and provide predictions such as where interest rates are headed, Coutinho said. Ask FT was built because many FT Professional readers were already using the publication for research. Around 83% of FT Professional readers agreed with the statement that the FT is a tool that helps them do their job, according to an internal survey. 

“For Ask FT, really what we were trying to do was to give people a much better starting point for that [research] process, and allowing and giving them that summarization of all of our archive of content back to 2004, so that they can much more easily do that research,” Coutinho said.

To respond accurately to real-time questions about today’s news, Ask FT works slightly differently from a tool like ChatGPT, which relies on knowledge it has built up. Ask FT’s model is updated with what’s been happening at a high level, Coutinho said, then uses source finding, which is a type of advanced search that will gather those sources and the LLM will work on them to provide the answer. The tool was built in-house by the FT and leverages Anthropic’s Claude, though it is designed to be model agnostic. 

“We’ve actually very recently reviewed whether we should change the model,” Coutinho said. “The new model was great and it was slightly better in terms of the quality of the answers but it was significantly slower and so we decided that it wasn’t a trade-off we were willing to make.”

They decided for now to continue with the older model because of the need to balance of speed and quality.

Nursing Times opted to work with a partner, Miso Technologies, to build its answers engine. Discovery was at the heart of this move with Booth recognizing the site has a vast archive of content, which often gets buried by the latest news being published.

“This was really about people being able to have their own very fluid dynamic personal experience and for them to be able to delve deeply into the content,” Booth said. The challenge both Booth and Coutinho are facing is how to make the tool more ingrained in day-to-day behavior.

Provable ROI? 

When users visit Ask Nursing Times, they see a strong click-through rate of between 5% to 6%.

“We do see this as a net positive driver of page views of the FT journalism, which is amazing, right?” Nick Fallon, managing director of FT Professional, said. “Most publishers are pretty alarmed by AI summaries being something that will cannibalize and substitute for actually reading the core articles. We’re seeing, not every time, but we’re seeing a higher degree of click through from Ask FT interactions than we are from perhaps some other AI sources out there.”

The most measurable return on investment so far is conversions, Booth said. Around 200 subscribers have come from links positioned within Ask Nursing Times. To continue driving conversions and engagement, publishers will need to deliver premium content that cannot be found anywhere else, as well as a premium experience, Booth said. 

But the FT and Skift have looked at the success of the tools more through the lens of retention rather than conversions.

“We know when we talk to companies that have enterprise subscriptions that they really like the Ask Skift element and so that’s where we see it adding value now,” Clampet said. Fallon echoed those sentiments noting that Ask FT has become a vital tool in conversations for new business deals.

“Ask FT is definitely having an impact and that is driving engagement, it’s driving more page views, which therefore triggers more chargeable core readers, which is good,” Fallon said. 

FT Professional offers three subscription packages—team, group and enterprise. Two of those packages have an engagement-based pricing plan that charges based on core readers, which is anyone who reads nine or more articles in a 30-day period, rather than a set number of readers. Companies pay for their core readers, with a volume discount applied, while access for a wider group of “occasional” readers is not charged.

The FT is also exploring monetization opportunities for the tool via sponsorship, following in the footsteps of media players like financial media publication Citywire, which launched a chatbot with sponsorship from asset manager Baillie Gifford.

A New Era of Search?

The success in the engagement with the tool has publishers thinking about whether it becomes an alternative to the traditional site search used by readers to find articles on publishers’ sites.

The FT has had a hybrid search engine, which is an AI-driven approach combining traditional text-based search and vector search, in place since September 2024. Skift took a similar approach late last year when Google sunset their on-site search product and they needed an alternative. They paired Ask Skift with a more traditional keyword search. Clampet believes the change improved reader engagement with Ask Skift and that a better era of site search could be on the horizon. Booth is now in the process of testing out hybrid search.

“On-site search for us, and most other publishers, simply isn’t good enough,” Booth said. “It delivers very poor quality results and a very poor quality user experience. Thankfully, AI is changing all of that and we’ll move next to hybrid search.”