AI Dramas vs Live-Action Microdramas: The Honest Comparison

Updated August 19, 2026 · Written by the AI Drama Reviews editorial team · 30 min read

DramaBoxEditor's pick — biggest short-drama library, lowest price to go unlimited 9.8Excellent

• Thousands of series, roughly 200 new titles every month
• Free episodes daily, plus ad-unlocks — no card needed to start
• Unlimited plans from about $5.99 a week, the lowest entry price we found
• Runs on Android, iOS and in a desktop browser

Go to DramaBoxAffiliate link · free to download · in-app purchases available

Affiliate link. We may earn a commission if you install through it, at no extra cost to you — it never changes our scores or placements. How we test · Disclaimer

The short answer, with the evidence attached

Live-action vertical drama costs about $150,000 to $200,000 per screen hour, or roughly $2,500–$3,300 per finished minute. AI production costs $20,000 to $50,000 per screen hour, and a heavily optimised pipeline has been reported at about $30 per finished minute. That single comparison explains most of what has happened to this category since 2025, and it explains almost nothing about whether you will enjoy watching the result.

The reason is that cost and quality are not the same axis. FlexTV vice-president Tang Tang told MIT Technology Review on May 15, 2026 that moving North American production to AI cut costs 80 to 90 percent and compressed schedules from three or four months to under one, with teams of about ten people. That is a vendor describing its own results, and it is the most specific public figure anyone has published. It says nothing about whether the resulting episodes hold a face steady across a cut.

The honest version of this comparison has three findings. AI production has already displaced live action in the volume tier, decisively and measurably, at least in the one market where output is counted. It has not displaced live action where a scene depends on a human face doing something small and continuous. And the cost saving went to producers rather than to viewers, who still pay $0.20 to $0.50 an episode either way.

Everything below tests those three findings against the numbers, including the places where the numbers do not exist. This site is an independent tracker and we earn affiliate commission from DramaBox, which sells a mixed catalogue and does not label which of its titles are AI-produced. We say so on this page as we do on every other.

What is the difference between AI and live-action drama?

A live-action microdrama and an AI drama share a format and differ only in production method, which is precisely why the comparison is so often botched. Both are 9:16 vertical, both run 60 to 120 seconds an episode, both run 40 to 100 or more episodes a season, and both put a coin wall after the first three to ten episodes. The container is identical. Only the contents were made differently.

What a live-action microdrama is

A live-action microdrama is a conventionally filmed vertical serial: cast, crew, locations, a shooting schedule and an edit. ReelShort's flagship slate is the clearest US example, shot in English with US casts. The production logic is compressed television, not compressed cinema, and the budgets reflect that.

What an AI drama is

An AI drama is the same serial with generative systems producing or materially assisting the visuals, voices, music or editing. A 90-second episode is assembled from a dozen or more separately generated clips, because shots rarely exceed 10 to 12 seconds. That is a model constraint rather than a directorial choice, and it shapes everything downstream, including the editing rhythm you notice without being able to name.

The hybrid middle that breaks the comparison

Most of what US viewers actually watch is neither pole. A series filmed conventionally in Chinese and dubbed into English by a voice model is AI-touched, not AI-made, and a large share of what DramaBox and ShortMax show American audiences falls into exactly that bucket. No app labels it. No dataset measures it. Any comparison that treats the catalogue as a clean binary is comparing two things that only partly exist.

Our definition pillar covers the five production models in more detail: what are AI dramas? The rest of this page compares the two ends of that spectrum, and flags the middle whenever it changes an answer.

How much cheaper is an AI drama than live action?

Per finished minute, live-action vertical drama costs about $2,000 and AI production has been delivered for as little as about $30, with realistic independent work landing near $280. That is a spread of roughly 70 to one at the extremes, and it is the reason more than 95% of new Chinese microdramas in Q1 2026 were AI-generated according to the China Netcasting Services Association.

The per-screen-hour framing is less dramatic and more honest. Live action sits at $150,000 to $200,000 per screen hour; AI production sits at $20,000 to $50,000. That is a four-to-ten times saving rather than a seventy times saving, and it is the range a producer planning a real slate would use. The $30 per finished minute figure describes a heavily optimised Chinese pipeline and is not a typical US result.

What the headline price leaves out

The cost lines that do not appear in any of the figures above are failed generations, retakes, upscaling, voice correction, human editing, music and sound effects, storage, translation and QC, distribution fees, legal review and refund handling. Above all it leaves out paid user acquisition, which is now often the largest single line. A producer who cut production by 90% and then spent the saving on ad buying has not built a cheaper business.

The metric that survives contact with a real pipeline is cost per accepted shot, not cost per generation. A 90-second episode may need many minutes of raw generated material before enough shots pass review. Our full breakdown of the creator side sits in what it costs to make an AI drama.

The gap that does not exist in public

No US benchmark exists for the cost of a completed, edited, legally cleared AI-drama episode. Every figure on this page is either Chinese-market reporting, a vendor statement or an individual creator's accounting. We would rather say that plainly than average three incompatible numbers into a false precision.

How much faster is AI production than live action?

FlexTV reported that AI production cut its North American schedule from three or four months to under one month, with teams of about ten people (Tang Tang, MIT Technology Review, May 15, 2026). That is the only dated schedule comparison anyone has published for this category, and it comes from the company that made the change.

The compression is structural rather than magical. A live-action shoot is a scheduling problem before it is a creative one: cast availability, location permissions, crew calls, daylight, weather, and a reshoot day that has to be found in three diaries. Generation removes every one of those dependencies and replaces them with queue time and review time.

What speed buys is not just earlier release. It buys iteration. A live-action microdrama commits to its script at the shoot; an AI production can regenerate a scene after seeing the retention curve on the first three episodes. That is the mechanism behind the "variation" half of the parallel-category claim we test later on this page.

How does AI video quality fall short?

Generated video fails in specific, repeatable, nameable ways, and knowing the eight of them turns "something looks off" into a diagnosis. None of these failures occurs in live action, because a camera cannot forget what a character was wearing four seconds ago.

The research literature treats this as the central open problem rather than a temporary bug. An ACM survey, A Survey: Spatiotemporal Consistency in Video Generation, published May 18, 2026, frames spatiotemporal consistency as the field's core difficulty; work on OpenReview (March 26, 2026) and arXiv:2510.04999 (October 6, 2025) reaches the same conclusion about narrative and temporal coherence across long sequences.

There is a ninth effect that is not a per-shot failure at all. Models trained on similar data produce similar faces, so different series start to feel cast from the same small pool of people. Live action has a version of this problem in typecasting, but not at the level of the face itself.

Mitigations reduce all eight and eliminate none. Character reference images, a locked character bible, fixed seeds, first-and-last-frame controls, shorter shots, shot-level human review and continuity editing are the working set. Each of them costs human hours, which is why the gap between the $30 floor and the $280 realistic figure is mostly labour.

Our field guide to spotting these in the wild is how to tell if a drama is AI-generated, and the consistency problem gets its own explainer in why AI drama characters change.

What can live-action actors do that AI cannot?

Live action owns micro-expression and sustained performance, and that is not a small remainder. A generative model can deliver a face that is lit correctly, attractive and approximately on-emotion. What it cannot yet deliver is the continuous, involuntary, small-scale behaviour a camera records when an actor holds a difficult beat for eight seconds without cutting.

This is the mirror image of the failure-mode list. The eight modes above describe what breaks between shots; performance describes what fails within one. A generated close-up tends to read as a series of correct expressions rather than one changing state, and the shorter the shot, the less exposed that is. Shots rarely exceeding 10 to 12 seconds is a technical constraint that happens to hide the weakness.

Two consequences follow. First, AI-heavy series edit faster than they otherwise would, because cutting away is the cheapest fix for a performance that will not hold. Second, any scene whose entire payload is a face reacting is the wrong scene to generate, which is a genre statement disguised as a technical one.

Live action also carries the whole apparatus of casting: a recognisable performer, a specific voice, a physical presence audiences follow between titles. Nothing in the AI catalogue has produced that yet, and the same-face effect described above works directly against it.

Which genres suit AI, and which suit live action?

The AI-versus-live-action question is settled genre by genre, not in general, and the split is consistent enough to predict. Stylisation hides generation artefacts; realism exposes them. Effects that would destroy a microdrama budget in live action are nearly free in generation.

Reviewer Jenny Cooper, writing at Vertical Drama Love on June 24, 2026, reached the same split from the other direction: after watching 25 AI titles and finishing three, she concluded that story remained decisive and that fantasy was the strongest category. That is one reviewer's experience rather than a study, and it agrees with the technical logic, which is the most we can claim for it. More on the genre landscape in the AI drama genre guide.

How many AI dramas are there versus live action?

On Douyin in April 2026 there were about 44,200 AI titles against 3,248 live-action ones, a ratio of roughly fourteen to one. DataEye counted about 221,900 AI dramas published on the platform across the first half of 2026, an average of roughly 1,200 a day, up from about 470 a day in January 2026.

That is substitution, and it is the strongest evidence against the pure parallel-category reading. In the market where output is actually measured, AI did not open a new shelf next to live action; it took over the shelf. The China Netcasting Services Association figure of more than 95% of new Chinese microdramas being AI-generated in Q1 2026 says the same thing from the supply side.

The US picture is unmeasured. No authoritative dataset shows what percentage of any American-facing app's catalogue is AI-generated, and no app publishes it. What we can say is that FlexTV has publicly moved its North American production to AI, StoReel raised $34 million against a stated target of 100 AI dramas a month, and DramaBox adds around 200 new titles a month across a mixed catalogue without saying how they were made.

What actually differs for the viewer: less than you think

The production method changes almost nothing about the experience of paying for and watching a season. Coins cost $0.20 to $0.50 an episode whether the footage was filmed or generated. Weekly unlimited plans run $5.99 to $29.99 across the category. The first three to ten episodes are free, then the wall arrives mid-cliffhanger, exactly as it did before AI existed.

The 80 to 90 percent production saving did not reach viewers. There is no discount for a generated title, no app charges less for one, and no app has announced that it intends to. That is worth stating plainly because the assumption that cheaper production means cheaper viewing is the single most common misreading of this category.

Three structural annoyances apply identically to both. Catalogues never overlap, because every platform finances its own titles and none licenses to another. There is no cross-app search. And no app labels which titles were AI-produced, so a viewer who cares about the distinction has no way to act on it beyond choosing a whole library.

What does differ is the texture. AI-heavy series cut faster, favour spectacle over conversation, and occasionally break in one of the eight ways listed above. Live-action series look more ordinary and hold together. Neither difference appears on a pricing page.

Do viewers actually dislike AI dramas?

Four pieces of public evidence exist on how viewers respond to AI dramas, and every one of them is an opinion or a single observation rather than data. We present all four because presenting only the flattering two, or only the damning two, is how this argument is usually conducted.

Read together, the four are less contradictory than they look. Gao and Nesvit describe the casual majority who watch on a phone in a queue. Cooper and the Reddit threads describe the attentive minority who notice artefacts and object. Both audiences exist. Neither has been measured in the United States, and no representative US dataset on this question has been published by anyone.

The one thing all four agree on implicitly is that reaction is driven by story and genre rather than by production method as such. Cooper says it outright. Gao's claim only holds where the material is good enough not to prompt inspection. We take the whole argument apart in AI slop: do viewers actually care?

Labour: a real cost, borne by real people

Reporting from the Chinese market describes microdrama pay falling by half or more, with some studios shrinking by around 70%. That is the clearest displacement figure available in this category, and it belongs in any honest comparison rather than in a footnote.

We want to be direct about it. A production method that cuts cost 80 to 90 percent and compresses a ten-person team into a schedule that used to need three or four months does not achieve that by finding efficiencies in the abstract. It achieves it by not hiring people who were previously hired. Crews, background performers, location staff, small-part actors and the technicians around them are the line item that moved.

Two caveats belong with it, and neither softens it. The figures are Chinese-market reporting, and there is no equivalent measured US number. And displacement in the volume tier is not the same as displacement in the premium tier, where the performance requirement described above has not moved at all.

On the rights side, the machinery that exists is contractual rather than statutory. SAG-AFTRA defines a digital replica as a program using a performer's voice, image or performance to generate new performances, and requires consent, disclosure, compensation and control, with materials dated November 17, 2025 and July 2025 interactive-media protections covering digital voice replicas. The WGA's 2023 MBA limits requirements to use AI and addresses the status of AI-written material. Neither reaches a Chinese microdrama studio or an independent creator working alone. The longer argument is in will AI replace actors?

Which carries more legal risk, AI or live action?

A live-action microdrama arrives with ordinary copyright in its footage. An AI drama does not, and that is the largest structural difference between the two that has nothing to do with how they look.

The US Copyright Office stated in Copyright and Artificial Intelligence, Part 2: Copyrightability, published January 29, 2025, that purely AI-generated material is not protected, while human selection, arrangement and editing may be. Applicants must disclose more-than-de-minimis AI material and exclude it from the human-authored claim. Part 3, covering training data, remains pre-publication and is not a final rule. There is no US federal law specific to AI dramas as of August 2026.

The practical asymmetry is this: an AI producer carries copyright, publicity, trademark, contract, music licensing, voice rights, labour agreement, false endorsement, privacy and platform disclosure exposure, and state law differs across all of it. A live-action producer carries most of the same list minus the copyrightability problem and minus the digital-replica question. Full US treatment in are AI dramas legal?

The economics of a hit: cheap production did not make the business easy

DataEye counted about 221,900 AI dramas published on Douyin in the first half of 2026 and found roughly 1,055 of them, or 0.47%, passed 100 million views. That is a hit rate of about one in two hundred, in a market producing 1,200 new titles a day.

The payoff side is real. The revenue split between platform and creator is commonly 50/50, and a title reaching a billion views can return $100,000 to $500,000 to its creator. Against a production cost of $3,000 to $14,000 for a 60 to 80 episode series, that is a genuine lottery with a genuine prize.

It is the odds that decide the business. Two hundred attempts at $3,000 to $14,000 each is $600,000 to $2.8 million spent to buy one title that clears 100 million views, and 100 million views is not a billion. Nobody publishes the distribution of returns below the hit threshold, which is exactly the number a producer would need.

The line that grew while production shrank

Paid user acquisition has become the largest cost line in this category even as production costs fell. That single sentence undoes most of the intuition around the 80 to 90 percent saving. If the constraint on a business moves from making the show to buying the audience, then cutting production cost by an order of magnitude changes who can enter but not who can win.

It also explains the volume. When a title costs almost nothing to make and everything to distribute, the rational strategy is to make an enormous number of cheap titles and let the market pick, which is precisely the pattern DataEye measured. Cheap production has not made hits common. It has made misses cheap.

How do AI and live action compare, criterion by criterion?

This is the whole comparison in one place. Read the right-hand column as a verdict on that row only, because the rows genuinely disagree with each other.

Count the verdicts and the score is close, which is the point. AI takes cost, speed, iteration, spectacle and volume. Live action takes performance, consistency, continuity, copyright and labour. The tie is the only row that concerns your wallet.

Which one should you actually watch?

Choose by genre and by tolerance for artefacts, not by production method as an ideology. Because no app labels individual titles, choosing an AI drama in practice means choosing an AI-first app.

We test every app on a US account, watch until the free tier runs out and pay through at least one paywall before scoring. The method, including what we do not test, is on how we test, and every app we have been through is listed in the reviews hub.

Testing the parallel-category claim

The claim under test, from a former Meta executive in April 2026, is that AI dramas are "not a substitute, but a parallel category, one defined by speed, variation, and the ability to cater to almost any audience preference imaginable." It is the most quoted framing in this space. It is also a claim with four testable parts, and it does not survive all of them intact.

Test 1: speed. Supported.

Three or four months to under one month, teams of about ten, and the ability to regenerate a scene after seeing the retention data. The only published schedule figure in the category comes from a vendor describing itself, which is a weakness, but nothing contradicts it and the structural argument for it is obvious. This part of the claim holds.

Test 2: variation. Supported.

Roughly 1,200 new AI titles a day on Douyin in the first half of 2026, up from about 470 a day in January. Variation is exactly what a cost structure like that produces, and DataEye's numbers are the strongest quantitative support for any part of this framing. This part holds too.

Test 3: catering to almost any audience preference. Unproven.

Volume is not the same as coverage. The genre evidence says AI production is strong in fantasy and weak in contemporary drama and comedy, and the same-face effect works against variety at the level viewers actually perceive. A catalogue of 221,900 titles that are disproportionately fantasy is not obviously catering to any preference imaginable. Nobody has published a genre distribution, so this part is unproven rather than false.

Test 4: not a substitute. Contradicted in the volume tier.

This is where the framing breaks. In April 2026 Douyin carried about 44,200 AI titles against 3,248 live-action ones, and more than 95% of new Chinese microdramas in Q1 2026 were AI-generated. Substitution is not a risk in that market; it already happened. Add the labour reporting, pay down by half or more and some studios down around 70%, and "not a substitute" reads as a claim made from the premium tier about a tier that has already changed hands.

Where this goes next, and what grows regardless

The category is growing fast enough that the AI-versus-live-action question does not determine its size. US microdrama revenue was about $0.82 billion in 2024 and $1.3 billion in 2025 (Fieldfisher, June 1, 2026), with Omdia forecasting roughly $1.5 billion for 2026 and about $2 billion for 2027 (via Reuters, August 2026). Those are whole-format figures covering filmed and generated titles alike.

The category is growing whichever way titles are made

US microdrama monthly active users went from 26 million in 2024 to 66 million in 2025 (Omdia, 2026). Globally the format was worth about $11 billion in 2025 and is forecast above $14 billion for 2026 (Omdia). Omdia also put microdramas at nearly twice the size of the global FAST market in March 2026. None of those figures separates AI-produced titles from filmed ones, and none can be added to Deloitte's in-app-purchase figures of $3.8 billion for 2025 and $7.8 billion for 2026, which measure a different scope.

The demand curve, in other words, is being driven by the format rather than by the production method. Viewers came for 90-second cliffhangers on a phone. Whether the footage behind them was filmed in Los Angeles or generated in a queue is a supply-side question that the audience has not been asked.

What would change the answer

Three things would move this comparison materially. A generation model that holds a single continuous performance well beyond 10 to 12 seconds would take the performance row away from live action, and Seedance 2.0's February 2026 release showed the step changes are not finished. A US labelling requirement would let the attentive minority act on its preference, which currently it cannot. And a published genre distribution of AI catalogues would settle test 3 above one way or the other.

What will not change it is price to the viewer. Nothing in the structure of this market creates pressure to pass production savings to the audience, because the audience is not choosing between a filmed and a generated version of the same title. Our forward look is in the 2027 forecast, and every number we track sits in the statistics page.

Frequently asked questions

The bottom line

AI production won cost, speed, iteration, spectacle and volume. Live action kept micro-expression, sustained performance, continuity, copyright and a crew. Those are not two positions on one scale, and the eighteen-row table above is the honest form of the answer: the verdicts genuinely split, and any summary that does not split is selling something.

For a viewer the practical guidance is short. Watch AI titles in fantasy, xianxia, werewolf romance and superpower stories, where stylisation hides the eight failure modes and the spectacle is real. Watch live action for contemporary drama, comedy and anything that lives on two faces in a room. Expect to pay the same either way, because you will.

For anyone assessing the business, the number that matters is not the 80 to 90 percent cost cut. It is the 0.47% hit rate and the fact that paid user acquisition is now often the largest cost line. Production got cheap; distribution did not. That combination produces enormous volume, a thin hit layer and a category that grew to about $1.3 billion in the US in 2025 (Fieldfisher) without ever telling anyone which titles were made by machines.

And the framing. "Not a substitute, but a parallel category" is true where the work is hardest and false where the work is cheapest. We would keep the phrase and attach the Douyin numbers to it every time it is used.

DramaBoxEditor's pick — biggest short-drama library, lowest price to go unlimited 9.8Excellent

• Thousands of series, roughly 200 new titles every month
• Free episodes daily, plus ad-unlocks — no card needed to start
• Unlimited plans from about $5.99 a week, the lowest entry price we found
• Runs on Android, iOS and in a desktop browser

Go to DramaBoxAffiliate link · free to download · in-app purchases available

Affiliate link. We may earn a commission if you install through it, at no extra cost to you — it never changes our scores or placements. How we test · Disclaimer

Read next

Sources

• MIT Technology Review (May 15, 2026) — FlexTV vice-president Tang Tang on an 80–90% production cost reduction, schedules falling from three or four months to under one, and teams of about ten people. A vendor describing its own results. Same report: ~470 AI titles a day on Douyin in January 2026, via DataEye.
• Category production economics, 2026 — live-action vertical drama at $150,000–$200,000 per screen hour (~$2,500–$3,300 per finished minute); AI production at $20,000–$50,000 per screen hour; a heavily optimised Chinese pipeline at ~$30 per finished minute, which is a floor rather than an average.
• TVBS (August 7, 2026) — independent creator Jiang Lan's own accounting of ~$280 per finished minute and ~$28,000 for twelve eight-minute episodes. NOW News (December 15, 2025) — 60–80 episode AI microdramas produced for $3,000–$14,000.
• DataEye (2026) — ~221,900 AI dramas published on Douyin in H1 2026, roughly 1,200 a day; ~1,055 of them (0.47%) passed 100 million views; April 2026 comparison of ~44,200 AI titles against 3,248 live-action titles.
• China Netcasting Services Association, via Global Times (2026) — more than 95% of new Chinese microdramas were AI-generated in Q1 2026.
• ACM, A Survey: Spatiotemporal Consistency in Video Generation (May 18, 2026); OpenReview on narrative and temporal consistency in long video (March 26, 2026); arXiv:2510.04999, Bridging Text and Video Generation: A Survey (October 6, 2025) — the technical basis for the eight failure modes and for treating consistency as an open research problem.
• US Copyright Office, Copyright and Artificial Intelligence, Part 2: Copyrightability (January 29, 2025) — purely AI-generated material is not protected; human selection, arrangement and editing may be; disclosure of more-than-de-minimis AI material required. Part 3, on training data, remains pre-publication.
• SAG-AFTRA materials (November 17, 2025) and July 2025 interactive-media protections — digital replica definition and the consent, disclosure, compensation and control requirements. WGA 2023 MBA — limits on requiring writers to use AI. Tennessee ELVIS Act, in force July 1, 2024. California AB 1836 and AB 2602 (2024). FTC impersonation rule, in force April 1, 2024 for government and business impersonation, with extension to private individuals not finalised.
• Reception evidence, all qualitative and none of it a survey: director Qingge Gao via PetaPixel/Today (August 2026); reviewer Jenny Cooper, Vertical Drama Love (June 24, 2026), 25 AI titles watched and 3 finished; Reddit comments on an AI mythological serial; Bogdan Nesvit of Holywater via TheWrap (2026) on comparable engagement metrics. The "not a substitute, but a parallel category" framing is from a former Meta executive, April 2026.
• Labour: reporting from the Chinese market describing microdrama pay falling by half or more and some studios shrinking by around 70%. No equivalent measured US figure has been published.
• Market size: Fieldfisher (June 1, 2026) — US microdrama revenue of $1.3 billion for 2025. Omdia via Reuters (August 2026) — ~$1.5 billion forecast for 2026 and ~$2 billion for 2027; US 2024 at ~$0.82 billion. Omdia (October 21, 2025) — $11 billion global for 2025; above $14 billion forecast for 2026 via Broadband TV News (May 29, 2026); microdramas at nearly twice the size of the global FAST market (March 2026); US monthly active users of 66 million in 2025, up from 26 million in 2024. Deloitte (November 18, 2025) — in-app micro-series revenue of $3.8bn for 2025 rising to $7.8bn for 2026, a different scope that must not be combined with Omdia's.
• Platforms: 36Kr (March 26, 2026) — StoReel's $34 million raise and 100-titles-a-month target. Investing.com (July 20, 2026) — DramaBox at 90m+ registered users across 200+ countries with ~200 new titles a month. Character.AI — (c.ai) Series launch, July 2026. ReelShort vendor pricing page, indexed August 2026, plus our own coin-unlock testing at $30–$50 for a full series.
• iiMedia Research via CGTN (February 25, 2026) — Chinese AI manju market of about 18.98 billion yuan for 2025. C2PA — provenance standard, not a rights registration and not a detector. Google — SynthID watermarking on Veo 3.1 output. ByteDance — Seedance 2.0, released February 2026.

Affiliate disclosure. Some outbound links on this page, including the DramaBox links, are affiliate links. If you install through one we may receive a commission at no additional cost to you, and it does not change what we write — this page states plainly that DramaBox does not label which of its titles are AI-produced, and that no app in the category passes its production savings to viewers. Accuracy notice. Several figures here come from single sources, from Chinese-market reporting or from vendors describing their own results, and we label those rather than presenting them as verified. Reception evidence in this category is entirely qualitative and is presented as opinion, not data. Market estimates from different research firms measure different scopes and cannot be added together. No authoritative dataset shows what share of any US app's catalogue is AI-generated. Verified August 19, 2026.