How do people perceive and respond to AI videos?

Contents

I. Why AI videos can feel “off” even when they look perfect
   1. Perceiving AI videos: what audiences are concerned about and what they actually mean
   2. Audiences accept AI videos as a tool, but tend to resist them for connection
II. Why trust matters so much in AI videos
   1. When brands get AI video wrong, the backlash is immediate
   2. But when they get it right, the reward is huge
III. YOPRST’s playbook for making AI video land with real humans
IV. Final thoughts

Most viewers react to AI-generated videos with a fleeting mix of fascination and distrust. Yes, such videos can appear cinematic and “expensive,” but AI video production raises new questions: is it real, is it ethical, and why does it occasionally feel emotionally thin or slightly unsettling? AI is slowly but steadily infiltrating music videos, advertisements, and educational content, so people’s feelings matter. We’ll look at the psychology behind viewers’ attitudes, their major concerns, and the creative decisions that make AI videos feel more human. Consider it a field guide for creating synthetic visuals that foster genuine trust.

Why AI videos can feel “off” even when they look perfect

Humans are wired to notice agency, not just visual polish. When we watch videos, we don’t only look at faces. We read micro-timing, intention, and biological motion to predict what should happen next. If an AI clip misses tiny cues like weight shifts, eye focus, almost-right lip sync, or expressions that don’t quite land, the brain registers a mismatch between expectation and perception. That is when the uncanny valley can kick in. The discomfort may feel like personal preference, but it is often your perceptual system flagging a prediction error. Why is that mismatch so sensitive?

According to research on biological-motion perception, the visual system is divided into two streams: one that prioritizes body form and posture, while the other reads optic flow and motion dynamics. When synthetic footage combines a realistic face with subtly robotic timing, the streams no longer agree. This produces a “not-quite-human” signal even when the image is sharp and detailed. According to predictive-processing theories, this behavior occurs because the brain is constantly forecasting movement and flagging minor timing violations before viewers can consciously figure out what is wrong.

That prediction error is not purely theoretical. In a 2025 electroencephalogram experiment, researchers showed participants short videos of joyful and fearful facial expressions in three different formats: real footage, deepfakes, and dynamic morphs. Participants were not informed which clips were synthetic. Even so, deepfakes triggered a larger N400 brain response, a signal frequently associated with expectation violation, than the original videos. According to post-experiment interviews, deepfakes were perceived similarly to real clips. In other words, the brain can tell that something is wrong in an AI video before the viewer can explain it.

Perceiving AI videos: what audiences are concerned about and what they actually mean

Viewers rarely react to AI video in a simple way. They may praise the visuals and still hesitate, scroll, or comment that it feels “off.” The points below translate common reactions into what people usually mean under the surface and what tends to trigger the response. A seemingly shallow complaint about AI’s realism often points to something deeper, such as trust, fairness, or emotional credibility. If you’re thinking of creating an AI video, you should treat these concerns as subtle audience signals, not creative instructions. Please note that these patterns show up across music, ads, corporate videos, and educational content.

^{Source: Nano Banana}

• “Is this emotionally fake or ‘soulless’?” Audiences ask this when the video looks polished but the emotion does not feel earned. Hyper-perfect imagery, generic faces, or surreal spectacle can make it hard to sense a human perspective behind the shot choices, so the feeling does not stick. This reaction is common in music videos and brand films, where viewers expect visuals to represent lyrics, values, or memories. The problem is usually meaning rather than resolution, because the audience is looking for intention and stakes. Without this, even beautiful frames can be read as decoration.
• “Where is the human effort or artistry?” Many viewers associate craft with visible work: a witty script with a human touch, expertly staged lighting, unconventional directing, and persuasive performances. AI-driven time and cost savings, which appeal to many businesses, can be interpreted as missed steps. A recent example is the band Nothing More. Fans criticized the band’s AI teaser for the Existential Dread video, accusing them of outsourcing the visuals. The frontman responded publicly, emphasizing that the concept was shaped by an artist and that AI was only part of the process.
• “Why do the people look weird or unsettling?” This sentiment is common for AI-produced brand ads, where viewers expect warm, human faces. In December 2025, McDonald’s Netherlands pulled an AI-generated holiday commercial after a wave of mockery online, with critics calling it “creepy” and pointing to uncanny expressions and warped facial proportions that broke the emotional tone. Once a commercial triggers that “wax-dummy” feeling, viewers stop following the story and start scanning for glitches, and the brand message fades into the background within seconds.
• “Can I trust what I’m seeing?” Suspicion changes interpretation. In entertainment, synthetic imagery can be read as fantasy. In documentaries, education, corporate updates, or anything that sounds like evidence, viewers shift into verification mode and become more skeptical. That is why provenance and disclosure matter, especially as policy is moving toward clearer labeling of AI-generated or manipulated media. The EU AI Act, for example, sets transparency duties for certain synthetic content, including deepfakes to protect viewers and ensure trust.
• “Is this ethically okay?” Viewers are concerned about explicit permission to use real people’s photos and voice recordings in AI clips. In Europe, these concerns translate into real obligations. Using someone’s identifiable image or voice may involve personal data, which requires a legal basis under the GDPR, which is frequently consent in practice. Synthetic videos that qualify as deepfakes may also trigger transparency obligations under Article 50 of the EU AI Act, such as disclosing that the content was created or edited using artificial intelligence.

^{Source: Nano Banana}

• “Are brands using AI to manipulate me?” Marketing tools powered by AI present previously unheard-of possibilities for customer analytics and hyper-personalization at scale. Depending on audience segments, behavior signals, or campaign performance data, generative models can quickly produce multiple ad variations with dynamic visuals, scripts, and CTAs. This makes advertisements more relevant, but it can also come across as intrusive if viewers think their emotions, fears, or insecurities are being targeted without their awareness or consent.
• “Does AI video lessen the value of human emotion?” Given the current rate of AI adoption, social media feeds may soon be flooded with mass-produced perfect videos. This is how human-made work may become less visible and its value reduced. In music and film culture, effort is associated with authenticity, so instant “cinematic” images may appear inflated and disposable. The pushback may appear to be a matter of taste, but it is also a cultural signal about scarcity, credit, and the meaning of craft in an age when content can be generated indefinitely. When the source is hidden, the fear intensifies.
• “Will this replace human jobs and creatives?” Many viewers connect the image to the labor story behind it. Presenting AI as a replacement for directors, animators, editors, or performers inherently leads to resentment, even when the outcome looks strong. Such an approach also elevates the standard: human errors can be perceived as charming, while AI errors are viewed as proof that the shortcut was not worthwhile. The reaction is emotional and economic at once, mixing fear of displacement with a desire to protect culture and careers. How AI work is framed publicly often shapes the response.
• “Is this just a gimmick or clickbait?” AI aesthetics can burn out fast. Rapid morphing, glossy surrealism, and obvious artifacts may earn a first scroll, then quickly feel repetitive. When the effect is not tied to a message, viewers label it as an empty spectacle with no substance behind it. Novelty also depends on context. Oddness can work in surreal art, but in a product ad or corporate message, it can read as evasive or cheap. Coherence matters more than style. If scenes do not connect, or the visuals do not support the claim, viewers conclude the AI effect is covering for a weak idea.
• “Am I being emotionally profiled by algorithms?” This worry is not just about ads getting personal; it is the fear that algorithms can infer mood or vulnerability and use that signal to steer what someone sees next. A concrete example came from the UK, where Network Rail, the public body that runs much of Britain’s rail infrastructure, tested “smart” advertising screens in stations that used camera-based AI to estimate passers-by’s age, gender, and emotional state to measure engagement and improve targeting. The trial drew backlash because people felt analyzed in public without meaningful consent.

^{Source: Nano Banana}

Audiences accept AI videos as a tool, but tend to resist them for connection

Most viewers judge AI videos based on their purpose. If you put a digital avatar into a compliance tutorial or product manual, people will just shrug and hit that “watch” button. But if you integrate the same “virtual presenter” into a brand film that’s supposed to evoke trust in a company, the view-through rate will likely plummet. This happens because in “utility” contexts, viewers reward speed, consistency, and clear delivery. In “connection” contexts, they look for intention, lived experience, and human texture. When AI replaces that texture, resistance may spike.

This does not imply that people either love or hate AI avatars; the decision hinges on the intended outcome. In internal training, the promise is simple: convey the message across, keep it consistent, and make it easy to reuse across teams and languages. According to Colossyan’s reported survey signals, 77% of workers said they would ask an AI avatar more questions, 70% would encourage coworkers to participate, and 54% thought that personalized AI presenters would help them remember information better. The tone here is pragmatic: if the synthetic trainer saves time and reduces admin friction, the algorithmic feel can be acceptable.

The same utility-first acceptance shows up in education. Students increasingly use Gen AI to explain concepts, summarize readings, and draft ideas, but many still worry about over-reliance and what it does to their long-term skills. In Jisc’s Student Perceptions of AI 2025 research, that tension comes through as “use it because it helps” mixed with “not sure it’s beneficial for me.” For neurodivergent and disabled learners, the value can be even clearer: patient, 24/7 support that rewrites dense material and reduces barriers to comprehension. When the goal is clarity, pace matters more than warmth.

Why trust matters so much in AI videos

Many people feel uneasy about AI-generated videos for a reason that has little to do with sharpness or style. It is a trust-and-control problem. When even a small mistake slips into an AI video (e.g., the protagonist’s appearance changing slightly between the frames), viewers judge it more harshly than they would a human-made mishap. Researchers call this algorithm aversion: once people see an algorithm err, confidence drops faster than it does for humans, even when the system is solid on average. That is why one odd frame can flip the reaction from “wow” to “what are they trying to slip past me?”

The demand for clarity is growing because most viewers do not feel equipped to tell what is real. A 2025 Pew survey found 76% of Americans say it matters to know whether pictures, videos, or text were made by AI or by people, yet 53% are not confident they can detect the difference. Policy is moving the same way: the EU AI Act includes transparency duties for certain synthetic content. Although provenance technology is available, it’s still unstable. A Washington Post test found that Content Credentials metadata on a fake video was usually hidden when the clip was uploaded across major platforms.

Real-world campaigns show how quickly trust can erode. Coca-Cola’s AI-led holiday ads drew backlash in 2024 and again in 2025, with critics calling the imagery uncanny and the nostalgia synthetic. Vogue faced outrage after running a Guess ad with an AI-generated model, reigniting fears about consent and jobs. Even advertisers are surprised. True Classic said Meta’s Advantage+ replaced its best-performing photo of a male model in a fleece set with an AI “grandmother.” When this occurs, viewers lose interest in the story and begin to question the process.

^{Source: Nano Banana}

When brands get AI video wrong, the backlash is immediate

Brands have learned the “soulless” penalty is real, and it shows up fast in sentiment. When Toys “R” Us released its Origin brand film created with OpenAI’s Sora, social chatter flipped: CARMA measured positive mentions falling from 12.2% to 3.4%, while negative climbed from 13.5% to 53.4%. Commenters called it “cynical” and “soulless,” pointing to an AI-generated child actor and the lack of real kids playing with toys. The subtext was blunt: you cannot automate nostalgia and expect applause. That is a brand risk, not a rendering problem, and it scales in seconds.

A similar dynamic hit Under Armour’s AI-powered Anthony Joshua spot. The director framed it as a breakthrough, but creatives alleged that parts of the film reused earlier material without proper credit, and the debate moved from “cool tech” to “creative integrity.” CARMA’s tracking, reported by Marketing-Interactive, shows how quickly that shift lands: before the post, conversation was 31.7% positive and 1% negative; after, it fell to 16.1% positive and rose to 7.3% negative. Once attribution becomes the headline, viewers start auditing the process, not the story.

But when they get it right, the reward is huge

When brands use AI video to solve a real constraint, audiences tend to treat it as engineering, not cheating. Headway, an edtech startup, told Business Insider it rebuilt its paid video ads with AI tools such as Midjourney and HeyGen, turning one concept into many fast variations for testing and localization. The company reported a 40% lift in video-ad ROI after shifting to this workflow. The win was not “AI for AI’s sake.” It was speed, volume, and iteration applied to a clear goal: find the message that converts, then scale it. AI handled the churn; humans steered the taste.

Another “got it right” example comes from Cadbury Celebrations in India. For Diwali, the brand used machine learning and a synthetic video of Shah Rukh Khan, a prominent actor and film producer, to allow local shops to create hyperlocal versions of the same ad, with the star appearing to name their store. According to the APAC Effie winners’ brief, thousands of small retailers created their own variants, resulting in a 7.3% increase in brand consideration and 35% sales growth. AI did not replace connection in this case; rather, it amplified it, transforming a single celebrity spot into a collaborative campaign that felt personal and pro-local.

^{Source: Nano Banana}

YOPRST’s playbook for making AI video land with real humans

As an AI video production company, we’ve completed over 30 projects for musicians, tech startups, FMCG companies, and dental clinics. The most important lesson we’ve learned is that AI is not a substitute for emotion. It is a production method that works best when the format aligns with the goal. AI shines in stylized music videos, social media ads, concept clips, explainers, and multilingual training. It struggles when viewers anticipate human proof. Audiences also expect real footage or a clearly labeled hybrid in high-trust industries like healthcare and finance. Follow the tips below to make sure your AI video hits the right chords:

• Start with a human truth and pick AI for the right job. AI is strongest when it visualizes what is hard to shoot: think dream logic, memory distortion, symbolic worlds, impossible transitions, or rapid concept iteration. AI fails when companies expect it to replace the emotional core of a story. Viewers forgive AI stylization when they can sense why a shot exists and what the character wants in that moment. They reject it when the sole purpose is to show the tool’s capabilities. That’s why a strong human-written script should always be the foundation of your video.
• Consider faces as a high-risk surface. If a concept does not require synthetic actors, do not force them into the frame. When control is limited, use abstraction, silhouettes, animation, masked performance, shallow depth of field, or a stylized universe in which the unreal is the point. If you do use faces, keep shots shorter, avoid extreme close-ups, and lock identity with consistent references (i.e., key frames). Plan a safety net for fixes, upscales, and face cleanup before you shoot the whole story. Most uncanny reactions start with eyes and timing, not resolution, so stability beats hyperrealism.
• Do not treat audio as an afterthought. Sound is where most viewers decide whether a piece has soul. A synthetic-looking shot can still land if the pacing, mix, and music feel intentional. Stocky or overly perfect tracks often make the whole video feel cheaper, even if the visuals are cinematic. Dialogue and lip sync are also brutal trust tests, so we keep spoken lines short and prioritize clean phoneme timing. We build the cut on a real vocal, a human performance, or a custom music bed and add foley and ambience to glue shots together. If the ear believes, the eye becomes more forgiving.
• Embed ethics into the production process. If a real person can be recognized by likeness, voice, or name, you need permission and a paper trail. In the EU, that usually means a lawful basis under GDPR for any personal data and clear disclosure for deepfake-style synthetic media under the EU AI Act. If you use a performer, get releases that cover AI generation, training, and usage scope. If you borrow a recognizable visual style, credit sources and avoid copying living artists without consent. Ethics is not a disclaimer at the end; it is decisions made before generation starts.
• Test early with real viewers and watch for the first moment of doubt. AI reactions are often fast and binary. If someone laughs for the wrong reason, pauses to squint at a generated face, or asks whether it is real, the cut is leaking trust. We conduct brief screenings with non-team members, record timestamps where they experienced disorientation or uneasiness, and correct those beats first. We also run A/B tests for video ads’ openers because the first two seconds are crucial. This way, we validate that our clients get a faster return on their AI investments.