Do AI Tools Harken the Death of the Academic Subscription?

The late show with Steven Colbert is cancelled and, like me you may be spluttering with anger over another CBS sell out, but the reason why this show was cancelled had more to do with the combination of money, demographics and highlight reels and less the thin skinned dear leader.

When Johnny Carson was at the height of popularity he regularly pulled in 11million viewers per night and if you missed the show, you were out of luck at the water cooler the next day. In 1975, Carson’s show made about $300MM in today’s dollars. Today Colbert brings in just about 2MM viewers and according to CBS losses $40MM a year. As if that’s not enough, Colbert also has demographics and the highlight reel working against him.

Today’s younger audiences are not predisposed to appointment TV and do not watch Colbert at 11:30 each night. They may, however, see the highlight reels in any number of places but the joke is on the content provider. These highlights aren’t monetized, and it is now much too late to fix what seemed like a really good idea at the time: to break the show into funny bits to drive traffic to watch the whole thing. That model, now proven broken, only benefits Google, Yahoo, Ticktock and other ‘channels.

In academic publishing we may be on track to make a similar catastrophic mistake in allowing our content to be programmatically snipped into ‘key takeaways’ and ‘two paragraph summaries’.  As AI tools fast become standard in search, traffic to the full text of research articles is showing the early signs of a significant dislocation. Traffic counted as ‘inquiries’ – generated by AI – is increasing. Pretty soon, just like Colbert all the best jokes are going to be free before the show is even over and these jokes will not have driven more traffic to the full show. Replace ‘joke’ with ‘research’ and publishers will soon have some explaining to do to the librarian about the collapse of full text retrievals.

It may be impossible to change the dynamic, but publishers must consider collecting transaction fees from these search partners for each of the AI ‘inquiries’ they perform. Otherwise, I think this may mark the beginning of the end for the subscription model. Contributing to this is the march of consumer preference, which as in the Colbert example above, young researchers are not methodically collecting 200 research articles and working through each: No, they use AI on 1000 instead and generate summaries, cross references and a draft paper within a fraction of the time. Publishers’ value add becomes frayed in this scenario.

It may seem strange to draw comparisons between a television show and academic research, but the changing dynamics are similar. Consider these and how they might apply to academic publishing,

• Pressure on legacy media: Colbert ad revenue dropped $20MM between 2024 and 2025 while viewership in the key 18-49 age bracket fell by 20%. Academic researchers 18 – 49 are our key demographic and they be different in their behavior

• Market fragmentation and platform shifts: Colbert’s monologues and segments go viral online but these don’t generate revenue like traditional TV ads. Without ads (read subscription) there is no economic model.
• Decline of linear viewing: User behavior has radically changed in the past 20 years and media consumption is far more driven by on-demand and at-convenience such that the consumer of media (read “researcher”) is in charge.
• Strategic realignment: Just like CBS and other media companies are realigning so are academic publishers and technology companies forming alliances which may redefine how academic publishers create value. There is likely to be more consolidation rather than less in academic publishing as strategic realignment takes shape.

Publishers need to be very wary of how they work with AI companies and search companies with embedded AI tools. On the surface, this functionality benefits the researcher and at least in the short term makes publishers seem tech savvy and accommodating of researcher needs. However, if researchers can retrieve data, information and research with significantly less effort the value proposition – regardless how cynical that sounds – for the publisher begins to erode. 

With some of the strongest media properties such as Netflix, Apple + and HBO now protected behind subscription walls (in contrast to legacy television), it would be ironic if academic publishers which started from this place of significant strength gave it all away by letting in the AI Trojan horse. Many publishers may also believe they have 'channel' strength when in fact they may have more 'brand' strength where researchers form stronger bonds with specific journal titles. Behind the paywall, perhaps more can be done to foster this brand strength and thus make the 'channel' even stronger. With respect to working with AI and LLMs , one of my colleagues said the other day, “I don’t know if I am shooting myself in the foot or the face”. 

If you have your own subscription platform, keep pace with AI functionality, engage in a broader manner with users but don’t let the horse in.