Four publishers, four verdicts on AI

Two statements, made a few weeks apart to the same audience, by publishers facing broadly the same market pressure and using broadly comparable technology:

“Our users love the AI-generated summaries.” — a regional publisher

“We need to be very careful if we give [our audience] AI summaries.” — a premium national title

Both are almost certainly correct. Each is a conclusion drawn from what their own readers told them.

Over the past year, Twipe has hosted a series of what we call Knowledge Breakfasts: a customer joins us early in the morning, presents a project or topic they are currently working on, and the Twipe team gets to ask questions. We did not set out to survey anybody about artificial intelligence, but came up every single time so here are some insights from across four different European publishers.

Four positions, side by side

Across the four different publishers we’ve hosted this year, here’s how each shared they are handling AI:

Ship it at scale. One regional business has AI summaries live in its articles, has launched short local audio briefings generated daily, and is extending into conversational and interactive formats. Their reasoning is partly defensive (“as our customers experience these features in other digital products, they will increasingly expect them from us as well”), and partly evidenced: listeners to the automated briefings went on to read articles afterwards rather than treating the audio as a substitute.

Handle it with care, or not at all. A premium national title has reached the opposite conclusion from its own reader feedback, quoted above. Its audience is old, loyal, and explicit about what it is buying: “they trust our company, that they trust our editors and what they think is important.” Automated summarisation competes with that promise.

Slow by design. At another premium national, the product team would like summaries, but editorial standards are the binding constraint. The objection, as relayed to us, is related to quality: “We really need to rely one hundred percent on the truth of the information.” As their product side observed, when your brand’s entire value rests on journalistic reliability, that is not an unreasonable line for a desk to hold.

Use it on the work, not the product. A regional daily’s subscription team doesn’t use AI in the reader-facing editorial product, but in the areas around it (e.g. generating alternative wordings for paywall and campaign copy).

What actually decided it

Two variables explain the spread in how AI is used in different publishing groups.

What the audience thinks it is buying sets the ceiling. Where readers have paid for a curated, finished, editor-selected product, they can experience automated editorial judgement as a downgrade. It becomes a substitution of exactly the thing they are paying a human institution to provide. Where readers have paid for access to a broad regional information service, an automated briefing that would not otherwise exist reads as pure addition. The same feature is a retention risk in one product and a genuine reader benefit in the other, because the two audiences bought different promises.

Who holds the veto sets the speed. Where AI work moved at all, the unlock was organisational. One publisher described senior sponsorship as what made the editorial conversation possible in the first place: “It helps to have that position directly at management, and to go into the discussions with the editorial teams.” The same organisation relies on a willing internal test bed, proving features in one part of the business before attempting to introduce them across the whole.

Where that leaves us

If there is one thing we would offer any publisher weighing up an AI feature, it is this: test it against the promise you have made to your readers, not against a competitor’s feature list. The publishers with the firmest positions on AI were the ones who could say most precisely what their audience was paying for — and that clarity licensed confident adoption in one case and deliberate caution in another. Both were right about their own readers.

As we at Twipe work with publishers on both sides of this question our job is to make sure the platform can carry an AI-generated audio briefing for one customer and an untouched, editor-curated edition for another without either feeling like the compromise.

Which is why we keep asking our customers to come and talk to us at eight in the morning. The work gets better when the people building it have heard, first-hand, why four publishers looking at the same technology arrived at four different answers.