Hi, I’m media innovation journalist Ulrike Langer and you’re reading my weekly AI in media newsletter. Thank you for being here! If someone has forwarded it to you please subscribe and never miss any future post.

Sometime in the last year, the media industry changed what it was optimizing for. There was no announcement and no debate. The goal used to be clicks. Now it is citations — showing up inside the answer when someone asks ChatGPT or Gemini a question. SEO teams have been renamed GEO and AEO teams, dashboards measuring "share of voice in AI answers" are selling briskly, and Digiday now calls AI visibility publishers' newest currency.
But before you restructure a team around it, ask what a citation actually delivers. The answer turns on a decision most publishers have never made deliberately. This is the fifth in a series on the choices AI forces on publishers, after whether to license at all, what to expose to agents, how they might pay, and how you would know what they took.
AI sends unusually valuable readers
The conversion numbers are the strongest part of the case. A visitor arriving from an AI assistant is worth more than almost any other visitor you get.
Similarweb's clickstream panel puts ChatGPT referral traffic at a 7.1% conversion rate, second only to paid search and ahead of direct, organic, social, email and display. Microsoft's analysis of 1,277 publisher and news domains puts AI traffic at roughly three times the conversion rate of other channels, with the multiple varying widely depending on which assistant sent the visitor.
Someone clicking a blue link on Google is often still deciding whether they want what is behind that link. Someone clicking through from an AI answer has already done that comparison in the chat. The assistant summarized the options, weighed them, and handed over a short list. If you still click, you want more than the summary gave you. The low-intent visits that always padded organic traffic without ever converting now stay inside the chatbot.
One caveat about all these numbers: They come either from vendors selling AI-visibility services or from platform owners whose own assistants are being measured. Treat any single figure as a sales artifact. What holds up is the pattern every source agrees on: AI referrals are a small stream of unusually high-converting visitors.
Another caveat applies to news specifically. Chartbeat finds that news and media sites receive the highest total pageviews from AI platforms and the lowest engagement per article. Readers arrive for a fact, take it, and leave. Home and garden sites benefit more from AI traffic than news sites do.
Sponsored
Save 10+ Hours a Week With 37 Claude Prompts
Every manager faces the same situations before lunch: a message to land, a meeting to run, a hiring call, a report due. The AI Report built 37 Claude prompts for exactly those moments, organised by the situations every manager faces.
Copy the prompt, fill the brackets, run it in Claude, and get back 10+ hours a week. Oh, and it's free.
All you have to do is subscribe to The AI Report, a 5-minute daily AI brief read by 400,000+ business leaders at IBM, AWS and Microsoft, and the full prompt pack lands in your welcome email. The newsletter and the prompts, both free. Subscribe and grab both
High value traffic, but low volume
AI traffic may convert well. But it arrives in very small quantities.
Chartbeat, which measures pageviews inside news sites rather than across the open web, found ChatGPT pageviews up more than 200 percent year over year — with all chatbots together still accounting for less than 1 percent of publishers' pageviews. That figure is probably low, for reasons discussed in the last section of this newsletter.
Set it against what publishers expect to lose. In the Reuters Institute's 2026 trends survey, 280 news executives across 51 countries expect an average decline in their search traffic of 43 percent over the next three years, with a fifth expecting to lose more than three quarters of it. They are extrapolating from a decline already underway. The same report draws on Chartbeat data showing pageviews from Google Search down 34 percent between December 2024 and December 2025.
So, the conversion rate says this channel matters and the volume says it does not pay the bills yet. A publisher who dismisses it because it is small is ignoring the best-converting traffic they have. A publisher who reorganizes around it is committing real resources to a trickle.
Getting cited and getting read are two different things
In news, open publishers get cited more. Similarweb's 2026 data puts Reuters and The Guardian at the top of AI visibility in the news category and finds open-access publishers outperforming paywalled publishers across the board.
An AI system can only cite what it can reach. Reuters and The Guardian keep most of their journalism in front of the reader, which means in front of the machine. Journalism kept behind a paywall gives the assistant nothing to quote and gets cited less for it.
Google worked that way too, up to a point. But Google delivered the click. The reader landed on your page, hit your paywall, and the wall could do its job. An AI answer that cites you without sending anyone removes even that.
Then the citations stop tracking with the traffic.
Similarweb's own referral numbers, reported by Digiday in March, put The New York Times first in AI referral traffic and eighth in the visibility index. Reuters runs the other way: First in visibility, ninth in referrals. The Times is paywalled. It is also blocking OpenAI over its lawsuit. It leads the traffic table anyway.
Reuters knows this. Its AI citations are growing fast, the company told Digiday, but from a very low base.
Numbers from outside the visibility business point the same way. DataDome, a bot-detection firm with no dashboard to sell, logged 17.7 billion AI agent requests across the sites it monitors between April and June, up 45 percent from the previous quarter. Meta's crawlers alone accounted for 9.1 billion requests and sent back almost no one. Claude referrals more than doubled in the same period and reached 876,000 — a seemingly large-sounding number against billions of requests. Billions of requests go in and hundreds of thousands of visits come out.
The industry has been asking how to get more AI visibility. The question was whether the citations are reaching anyone. But now, the publishers highest in visibility are not the ones receiving the most traffic from it.
One publisher has found a way through. TIME, the most open of the three companies I wrote about earlier in this series, does not wait for citations to become clicks. It exposes almost everything to agents, tracks obsessively where it appears in AI answers, and sells that measurement to brands as a product. The visibility itself is the revenue. No click required.
Every other approach here depends on traffic showing up. The Economist made the opposite choice, keeping its journalism behind the paywall and forgoing the visibility to protect what readers pay for. Both decided on purpose. Most publishers have not decided at all. They are optimizing for a citation count and assuming the readers will follow.
Publishers are being asked to trust a measurement blindly
AI assistants routinely strip the referrer, so a large share of AI-driven traffic lands in analytics mislabeled as "direct." How large a share, nobody can say with confidence. Published estimates run from roughly a third to over ninety per cent, the studies behind them measure different things, and their samples range from a few dozen visits to a few hundred sites. Almost all come from firms selling analytics or visibility services.
The tooling is only starting to catch up, and it is not transparent either. Google Analytics added a native AI Assistant channel in May 2026. Google names ChatGPT, Gemini and Claude as examples but has never published the full list of assistants that the channel recognizes. Publishers are being asked to trust a measurement whose definition is not public.
So, publishers are undercounting the one channel that converts and making the expose-or-gate decision on numbers that hide it. You cannot decide whether to chase something you cannot see. Before any of the strategic questions there is a measurement question, and most newsrooms have not answered it.
AI visibility arrived with new metrics, new job titles and new software to buy. Underneath all of it sits the question publishers have always had to answer: What is your work worth? And what do you put behind a paywall? A citation is worth whatever your answer makes it worth. The publishers who get something out of this currency will be the ones who decided on purpose what sits behind the citation. The same archive that makes you reachable by an AI assistant is the archive that makes you able to charge one. Everyone else is collecting impressions they cannot use for a meaningful purpose.
Five takeaways for newsrooms
Measure AI referrals before you judge them. Most of this traffic is hiding in your "direct" bucket. Without AI-channel tracking you are deciding blind, and even with it on you are seeing a floor.
Separate the rate from the volume. AI traffic converts several times better than organic traffic and still amounts to under one percent of publisher pageviews. Plan accordingly on both counts.
Don't assume citations become readers. Open content gets cited more, but the publishers highest in AI visibility are not the ones receiving the most AI referral traffic. Track citations and arrivals as two separate numbers. A vendor reporting only citations is showing you only half of the transaction.
Ask any visibility vendor what the citation converts to. Share of voice in AI answers describes how often you appear and stops there. If they cannot connect it to a subscription or a sale, you are buying a report.
Tie discovery to your gate decision. Being cited is the front door to the exposure question every publisher now faces. Optimizing to be found only pays once you have decided what happens when the reader arrives.


