The Feed Cannot Tell You Wrote It
The recommendation algorithm behind the video feed that turned book discovery into a measurable sales channel has one job, stated plainly by the platforms that build it: keep people watching, tapping, and returning. It was never built to verify that a clip about a book came from someone who actually read it, because verifying that was never the metric it optimizes. It optimizes attention, and attention does not care about provenance.
This gap sat mostly theoretical for a few years, because producing a compelling video used to require a camera, a face, and real production effort, which functioned as an informal filter even though nobody designed it as one. AI video generation removed the filter. A creator can now produce a cinematic, atmospheric clip depicting scenes from a book, real or entirely invented, in minutes, using character reference sheets and prompted shot sequences, at a volume no filming schedule could sustain. The clip enters the same feed, gets ranked by the same watch-time signal, and the algorithm has no way to tell it apart from footage of a genuine reader holding up a genuine book.
The Flood Arrives in a New Medium
Readers of the catalog data on AI-generated text will recognize the shape of what is happening to video immediately, because it is the same mechanism wearing a different medium. When producing a unit of content drops from expensive and slow to nearly free and fast, the volume of that content does not grow proportionally. It grows explosively, because the constraint that used to ration supply, cost and effort, has been removed rather than merely reduced.
Creators and platform observers already describe AI-generated clips depicting books, some real, some invented for engagement alone, filling accounts at a pace hand-filmed content cannot match. The same commodity economics that hit the self-published catalog applies here on a compressed timeline: a flood of interchangeable content competing for the same limited pool of viewer attention drives the value of any individual clip toward the floor, while the mechanism for standing out has to move somewhere the flood cannot follow.
Nothing about the feed’s ranking signal was ever built to notice that the camera is gone.
What the Algorithm Still Cannot Reward
One detail is worth holding onto carefully before concluding the channel is finished, because reading the flood as total misses it. The algorithm ranks by watch time, completion, re-watches, and shares, and a well-made AI clip can score respectably on all four in isolation. What it cannot manufacture is what happens after the click-through, when a viewer moved by a video goes looking for the book behind it.
An AI-generated clip built around a book that exists holds up under that follow-through exactly as well as the book itself does, which means for a real author the technology is neutral or even useful as a production tool. An AI-generated clip built around a scene, a character, or an entire book invented purely to generate engagement collapses at that same follow-through, because there is nothing on the other end of the click. The algorithm has no signal for this collapse, because the collapse happens off-platform, in a search bar or a storefront, one click after the metric it optimizes has already been satisfied. The flood can win the feed. It cannot win the moment a curious viewer goes looking for what comes next, because past that moment the content has to be backed by something real, and a fabricated scene has nothing to back it with.
Where That Leaves an Actual Author
The practical read is narrower and more actionable than either despair or dismissal.
Treat the feed as a qualifying round, not the destination. A clip’s job was always to get a curious viewer to leave the platform and land somewhere the author controls, and that job matters more, not less, now that the feed itself is crowded with content that has nowhere real for a viewer to land.
Use the same production tools the flood uses, aimed at something real. The reference-image and multi-shot consistency techniques driving the flood of fabricated content work identically well applied to an author’s own actual book, at the same collapsed cost, which means opting out of AI-assisted video on principle mostly just cedes the feed’s algorithmic favor to the volume producers rather than protecting anything.
Make the thing behind the click impossible to fake cheaply. A landing page, a series bible, a body of published work that a curious viewer can verify in thirty seconds is exactly the kind of asset a volume-focused fabricator has no incentive to build, because building it does not increase clip output, and a real author already has every reason to build it regardless of what the feed rewards.
The unsettling part of this story and the reassuring part turn out to be the same fact stated two ways. The feed genuinely cannot tell a real book from an invented one, and it is not going to start being able to. What it also cannot do is manufacture the destination a genuine curious click leads to, and that destination was always the actual product being sold. The flood crowds the room where attention gets captured. It has nothing to offer the room the attention gets sent to next, and that second room is the one an author was supposed to be building all along.
