What AI app do TikTokers use?

The Short Answer: CapCut Dominates — But Creators Are Reconsidering

Ask a room full of TikTok creators which AI editing app they reach for, and one name comes up more than any other: CapCut. Owned by ByteDance — TikTok's parent company — CapCut pairs a deep AI toolkit (script-to-video, AI avatars, auto-captions, long-video-to-shorts conversion) with a constantly refreshed catalog of trend-driven templates that map directly to what is going viral on TikTok and Reels. Its huge user base also makes it the default learning environment for new creators, with tutorials and template packs everywhere. In hands-on testing, Buffer's reviewer called it a "jack-of-all-trades" and their tool of choice "because it's just that easy to start with."

That dominance, however, is no longer unchallenged. Multiple editorial roundups note that since 2025, many creators and brands have started actively looking at alternatives, driven by three distinct pressures.

• Licensing concerns: A June 2025 terms-of-service update granted CapCut a perpetual, irrevocable, royalty-free, worldwide license to use, modify, and distribute content uploaded to the app — a red flag for brands and agencies whose footage carries commercial value.
• Feature paywalling: Watermark-free exports, Pro templates, and longer auto-captions have been moved behind paid tiers, with Pro running $19.99/month.
• Ban worries: CapCut briefly went dark in the US in January 2025 alongside TikTok before returning in February. It is not currently banned, but the temporary pull from US app stores has left some users wary.

The beneficiary of this unease is largely Edits by Instagram. Meta launched the app in April 2025 as a direct answer to CapCut: fully free, no subscriptions or in-app purchases, no watermark, 4K export, and a respectable AI feature set including auto-captions, AI cutout, and AI image animation. For creators burned by CapCut's shifting terms, the pitch is straightforward.

A caveat worth keeping in mind: no published survey data actually measures what percentage of TikTokers use each app. "Most popular" claims rest on editorial reviews and feature comparisons rather than adoption statistics. That distinction matters — the same skepticism applies to any tool recommendation, whether it's an editing app or an AI agent skill. It's exactly why SkillzBench is building independent verification and benchmarking for agent skills, testing each one in a sandbox to publish measurable lift rather than vendor assertions.

For now, the practical picture is this: CapCut remains the most-cited AI app among TikTok creators, but its grip is loosening. Creators who value free, watermark-free editing are trialing Edits by Instagram; those prioritizing price stability lean toward InShot's lifetime plan. The default answer is still CapCut — it's just no longer the only answer.

The Full Toolkit: AI Editors, Repurposing Tools, and TikTok's Built-In AI

Ask ten TikTokers what app they use and you'll get ten answers — but the real story is that the toolkit now splits into distinct categories depending on what the creator is trying to accomplish. Understanding those categories matters more than memorizing app names.

AI-enabled editors are traditional editors with AI bolted on. Buffer's hands-on testing places CapCut, Canva, and Adobe Express here — you stay in control, and AI handles captions, background removal, and silence trimming. CapCut remains the default: its ByteDance ownership ties it to the TikTok ecosystem, and its trend-driven templates map directly to what's going viral, per comparative testing. Its auto-captions benchmark at roughly 94% accuracy across 100+ languages, though that figure is vendor-cited.

AI-led editors flip the model: you describe the edit in plain language and AI does the work. Buffer's practitioner tests found Vyra produced a finished edit in about 20 minutes versus the author's usual 30, while OpusClip and Stanley Studio take similar instruction-first approaches. The catch, per that same testing: "The tools are no longer the bottleneck. Knowing what to ask for is."

Repurposing tools attack the dominant use case — turning long-form video into shorts. Quso AI ($10–$25/mo), Spike Studio (free tier, Pro+ at $15/mo), and Podsqueeze convert podcasts and YouTube uploads into short clips with smart cropping and multi-language subtitles, according to a Podsqueeze roundup. CapCut claims its own long-to-short tool produces 10+ viral clips from one video.

TikTok's native AI deserves its own column:

• AR filters and generative effects via Effect House
• Text-to-video generation inside the app
• AI script and caption suggestions
• AI music creation tools

Per analysis of TikTok's AI strategy, the platform positions these as major growth drivers by lowering production barriers.

The biggest 2025 shake-up is Edits by Instagram, launched April 2025 as "a direct answer to CapCut" — fully free, no watermark, 4K export, AI cutout, and AI image animation, per Sendible's editor comparison. It's drawing creators spooked by CapCut's June 2025 licensing terms, which grant the app a perpetual, irrevocable license to uploaded content.

That licensing dust-up mirrors a broader lesson: before adopting any AI tool, verify what it actually does with your work. It's the same principle SkillzBench is applying to AI agent skills — security-scanning and benchmarking each one rather than trusting the marketing. Whether it's an editor or a skill package, the tool's claims and its terms both deserve scrutiny before it touches your content.

What the Tools Actually Cost — and Which Claims to Trust

Money answers the "which app" question faster than any feature list — but in this market, the pricing itself is contested, and the boldest claims come from the vendors selling the tools.

CapCut's paid tier runs $19.99/month ($179.99/year), with a Standard tier at roughly $9.99/month, according to Sendible's pricing breakdown. That buys watermark-free exports, Pro templates, and longer auto-captions — features that were previously free, which is part of why creators started hunting for alternatives in 2025. Edits by Instagram undercuts everyone: it's fully free, with no subscriptions or in-app purchases, 4K export, and no watermark.

InShot's lifetime pricing is where sources openly disagree. Sendible cites $49.99 lifetime, while pexo.ai's comparison lists $39.99 — a $10 gap on the same product, which tells you how volatile this market's pricing documentation is. Treat every number here as approximate.

The wider AI-tooling spectrum spans a wide range:

• Descript: $35/month, best for speech-heavy video rather than crisp TikTok edits
• Entry-level AI video tools: $9/month (Snapy) up to $70/month (VMake Pro)
• Repurposing tools like Quso AI: $10–$25/month
• Vyra: $24/month with your own AI, $54/month with Vyra's

Now the claims. CapCut's much-quoted ~94% auto-caption accuracy across 100+ languages is vendor-cited, flagged by the source itself as needing independent verification. CapCut also calls itself "widely regarded as one of the best AI video generators for TikTok" — on its own product page. Canva's FAQ quietly concedes the "best" generator "may vary depending on the creator's needs."

The bigger honesty gap: no source provides actual adoption data. Every "most popular" claim in this space — including CapCut's dominance — is an editorial judgment, not a measurement. That's the same problem SkillzBench is built to solve in the AI-skills ecosystem: separating measured performance from vendor assertion, and publishing benchmark Lift separately from ratings so money never touches scores.

Until someone benchmarks these editors under identical conditions, the practical move is simple: start free with Edits or CapCut's free tier, verify caption accuracy on your own footage, and treat every "best" claim as marketing until proven otherwise.

How to Get Better Output: Prompting AI Editors Like a Pro

The best AI editor won't save a vague prompt. After hands-on testing of more than seven AI video tools, Buffer's conclusion was blunt: the tools are no longer the bottleneck — knowing what to ask for is (hands-on testing by Buffer). Creators who learn basic editing vocabulary like J-cuts, safe zones, and hook text consistently get better AI output than those who type "make it viral."

Buffer's testing produced a simple four-part brief you can adapt to any AI-led editor, from Vyra to OpusClip. Each prompt should combine:

• What to cut — silences, filler words, dead air, tangents
• What to add — captions, b-roll suggestions, transitions, hook text
• Format and platform — vertical 9:16 for TikTok, clip length, safe zones
• Style reference — a creator or edit style you want the AI to emulate

This structure matters because AI-led editors work from plain-language instructions, unlike AI-enabled editors where you drive the timeline yourself (Buffer's testing). The more specific the brief, the less manual cleanup afterward.

Sendible's guidance is equally direct: your hook in the first three seconds decides whether people keep watching (editorial guidance from Sendible). When prompting an AI editor, explicitly instruct it to front-load your strongest moment — the punchline, the surprising stat, the visual payoff — rather than letting it default to chronological cuts.

In Buffer's tests, this prompting gap showed up in real numbers. Descript got roughly "40% of the way" to a finished video, while Vyra produced a complete edit in about 20 minutes versus the author's usual 30 — results that varied more with prompt quality than with the tool itself (Buffer's comparative tests).

Prompting can't fix everything. Podsqueeze's assessment is that AI editors do well on simple projects but lag on complex ones (Podsqueeze's analysis), and Buffer found Descript best suited to speech-heavy content rather than the crisp, trend-driven edits typical of TikTok. For multi-clip storytelling, precise beat-synced transitions, or brand-specific visual language, an AI-enabled editor with manual control beats an AI-led one.

The practical takeaway: match the tool to the job, then invest your time in the brief. It's the same principle SkillzBench applies to agent skills — measurable, like-for-like testing reveals what actually performs, so you stop chasing hype and start picking tools that fit your workflow.

The Bigger Lesson: Don't Trust AI Tools at Face Value — Verify Them

The most useful skill you can develop around AI tools isn't knowing which app is trending — it's knowing how to distrust marketing claims gracefully. CapCut calls itself "widely regarded as one of the best AI video generators for TikTok," a self-reported claim no independent adoption data actually confirms. Meanwhile, a June 2025 terms-of-service update quietly granted CapCut "a perpetual, irrevocable, royalty-free, worldwide licence" over uploaded content, as Sendible's analysis documented — a detail buried far below the feature headlines.

This pattern — impressive demo up front, fine print underneath — is exactly what creators should watch for everywhere AI tools circulate. It applies doubly to a newer category: AI agent skills, the instruction packages that extend coding and chat agents like Claude Code, Cursor, and Gemini CLI. These skills are spreading rapidly across GitHub, and published research found roughly 26% of them carry at least one vulnerability. An editing app with a bad license term risks your footage; a compromised agent skill risks your code, your credentials, and your workflow.

That gap between what a tool claims and what it actually does is what SkillzBench exists to close. The lab is building a verification pipeline that treats every skill the way a careful creator treats a new editing app:

• Version pinning — every skill is pinned to a specific commit SHA, so what you test is what you run
• Security scanning, license checks, and quality validation before anything is recommended
• Sandboxed benchmarking that runs the same model with and without the skill, then publishes the Lift in percentage points
• A 0–100 trust score built from security, quality, license, maintenance, and community signals — with Lift always shown separately from ratings, because money never touches scores

The benchmarking approach mirrors what Buffer's hands-on testing of AI video editors did well: testing multiple tools against the same footage rather than trusting vendor demos. When Buffer's author found Vyra produced a finished edit in roughly 20 minutes versus their usual 30, that was a measured comparison, not a marketing claim. That's the standard AI tools should be held to — and it's the standard SkillzBench is applying to agent skills as it builds toward launch.

The broader takeaway for any creator or developer: verify before you adopt. Read the terms, question the "widely regarded" claims, and prefer tools with transparent, reproducible evidence. If you work with AI agents and want skills that are commit-pinned, security-scanned, and benchmarked for actual Lift, you can join the SkillzBench waitlist at skillzbench.com — verified, benchmarked skills are the goal, and the skeptics get there first.