Every week a robot video melts the internet, and every week the same question follows: is that real? The honest answer is usually “yes, but—” and everything important lives in the but. The robot exists, the motion happened, and yet what you watched may be teleoperated, sped up, the 47th take, or a capability that works precisely once under studio conditions. Since demo videos are how this industry markets machines that cost as much as cars, learning to read them is a core buyer skill. Here’s the 8-point checklist we run on every video before it influences a score on this site.
First, understand the spectrum — “fake” is the wrong word
Robot demos live on a spectrum of honesty, and almost none of them are fabricated. At the trustworthy end: continuous, unedited footage of autonomous operation in a messy environment — rare, and gold when you find it. In the middle: real autonomy filmed selectively (best takes, ideal lighting, pre-mapped rooms) — standard practice, forgivable, but not Tuesday-afternoon reality. Further along: teleoperation, where a human pilots the robot remotely — the motion is real, the intelligence isn’t the robot’s. And at the theatrical end: choreographed set pieces where every object position is rehearsed and any deviation breaks the act.
The industry’s open secret is that teleoperation is also its legitimate learning engine — assisted operation generates the training data that makes autonomy improve, as 1X candidly describes about NEO’s development. The sin is never assistance; it’s undisclosed assistance sold as autonomy.
The 8-point checklist
1. Look for cuts. Every edit is a place where a failure can hide. A task shown in one continuous shot (watch shadows and background clocks for continuity) is dramatically stronger evidence than the same task assembled from six angles. Multi-cut task completion is the single most common inflation technique.
2. Check for speed manipulation. Robots are slower than marketing wants. Watch for unnatural motion in background elements — people walking, screens flickering, fabric settling. Honest videos increasingly label “1x speed” precisely because the practice is rampant; the absence of a speed label on smooth fast footage is itself information. Real-world context: robot laundry folding runs far slower than human speed, which is why our laundry reality check exists.
3. Hunt for the operator. Teleoperation hides off-camera. Signals: suspiciously human-like hesitations and recoveries, a person in VR gear in any behind-the-scenes material, latency-style pauses before complex grasps, and company language like “shared autonomy” or “human-in-the-loop” buried in the fine print. Companies doing genuine autonomy say the word “autonomous” loudly and repeatedly; vagueness is a choice.
4. Audit the environment. Is the demo space suspiciously perfect? Uniform lighting, uncluttered surfaces, objects placed at ideal heights and spacing, no cables, no kids, no chaos. Real capability shows up in messy kitchens and active job sites. This is why commercial deployment footage — Figure handling real packages, Ozmo on an actual Manhattan facade — outweighs any studio reel: reality doesn’t reset between takes.
5. Count the repetitions. One success proves possibility; reliability is a statistics claim. Strong evidence looks like Figure’s deployment numbers — hundreds of thousands of task repetitions with a public failure record — or Jaibot’s five years on job sites. A viral clip is a sample size of one, selected from an unknown denominator. Always ask: how many takes didn’t we see?
6. Separate hardware feats from intelligence feats. Backflips, dances, and parkour showcase actuators and balance control — genuinely hard engineering that says nothing about whether the robot can find your keys. Manipulation of varied objects in novel scenes is the capability that matters for useful work, and it’s precisely what athletic demos don’t demonstrate. Score the two separately, always.
7. Watch the hands and the edge cases. In manipulation demos, the tell is what happens at the boundaries: does the robot recover when a grasp slips, or does the video end? Does it handle the second object differently from the first (adaptation) or identically (choreography)? Fumble-and-recover footage is paradoxically the most credible thing a company can publish — it means they’re confident in the average, not just the highlight.
8. Trace the claim to a deployment. The ultimate check: does anything in the video connect to a machine doing real work for a real customer? Deployment is where demos go to be audited by physics and invoices. Every robot in our index is scored with deployment evidence weighted far above demo footage — it’s the reason a “boring” veteran like Digit outscores flashier machines on trust, and the core of our methodology.
Why this matters more as you get closer to buying
Demo literacy isn’t cynicism training — the machines are genuinely getting remarkable, and some viral clips are exactly what they appear to be. It’s calibration training, because a five-figure purchase decision deserves better inputs than a marketing department’s best sixty seconds. The practical application: before buying, hunt third-party footage (owner channels, developer forums, trade-show floor video shot by attendees), ask vendors directly which shown tasks run autonomously today, and treat the answer’s specificity as data. Our first-robot playbook builds this into the full purchase process.
The bottom line
Robot demos aren’t lies — they’re résumés. Nobody’s résumé mentions the job they got fired from, and no demo reel includes take 46. Run the checklist: cuts, speed, operators, environment, repetitions, hardware-vs-intelligence, edge cases, and deployment receipts. Two minutes of that discipline filters ninety percent of the noise in this industry — and the machines that survive the filter are the ones worth your money and your living room.
Every capability claim on WhichBot is weighted by deployment evidence over demo footage — see how we rate.