AI Content: Why Audiences Say They Hate It but the Data Says They Watch It

In January 2026, YouGov published a survey of 5,000 US internet users. 72% said they have a negative view of AI-generated content. 68% said they try to avoid it. 61% said brands that use AI-generated content lose their trust.

In the same month, Tubular Labs published their Q4 2025 Platform Intelligence Report analyzing 2.3 billion video views. AI-generated and AI-assisted videos on TikTok, YouTube Shorts, and Instagram Reels had an average engagement rate of 5.8%, compared to 4.2% for the platform-wide average across all content. AI content tagged with #AIGenerated or #MadeWithAI averaged 6.4% engagement, higher than the overall average.

Both data sets are credible. Both cover the same time period and roughly the same audience. And they directly contradict each other. Audiences say they hate AI content. Audiences engage with AI content at above-average rates. Both statements are true at the same time.

This article examines why the contradiction exists, what it means for content strategy, and what the data actually tells us about how audiences respond to AI content in practice.

The stated preference vs revealed preference gap

Behavioral economists have a term for what is happening here. Stated preferences are what people say they want when you ask them. Revealed preferences are what people actually do when they make choices. The two frequently diverge.

Daniel Kahneman's work on dual-process theory explains the mechanism. When a survey asks "How do you feel about AI-generated content?" the respondent engages System 2 thinking: deliberate, analytical, and influenced by social norms. The answer reflects what they believe they should feel. When that same person scrolls TikTok at 11pm and watches a video for 30 seconds without checking whether it was made with AI, they are operating on System 1: fast, automatic, driven by stimulus response.

The gap is not unique to AI content. It appears in every category where social desirability influences self-reporting.

The pattern is consistent across domains. When you ask people about categories where there is a "correct" social answer, their stated preferences will overweight that answer. When you measure their actual behavior, the social desirability filter disappears.

The engagement data in detail

The claim that AI content performs well needs specific data, not generalizations.

AI content engagement rates by platform

The pattern: AI content overperforms on short-form entertainment platforms and underperforms on long-form and professional platforms. This split is not random. It reflects a difference in viewing mode.

On TikTok, users scroll rapidly and decide within 0.5-1.5 seconds whether to keep watching. They do not pause to evaluate whether content is AI-generated. They respond to the hook, the visual novelty, and the emotional trigger. AI content that is visually surprising, pattern-breaking, or curiosity-inducing performs well in this environment because the viewing mode is pure System 1.

On YouTube long-form and LinkedIn, users invest deliberate attention. They choose content intentionally, evaluate it while watching, and form opinions about the creator. In this viewing mode, AI content triggers a different response: skepticism about authenticity, questions about the creator's credibility, and the "uncanny valley" discomfort that AI imagery sometimes produces at extended exposure.

The disclosure effect

Does labeling content as AI-generated change performance? The data is mixed.

Sprout Social's 2025 Transparency Report tested 1,200 AI-generated social media posts across 60 brand accounts. Half were labeled "Created with AI" and half were not disclosed.

Disclosure reduces engagement by 18-20% but increases share rate by 22%. The share increase appears driven by novelty sharing ("look at what AI made"). The engagement decrease reflects the System 2 activation that disclosure triggers: once a viewer knows content is AI-generated, they evaluate it through the lens of their stated preferences.

The negative comment rate jumps 425% with disclosure. But the absolute number is small (4.2% vs 0.8% of total comments). The most common negative comments were "this looks AI-generated" (ironic when the label already says so), "why are you using AI," and "this is lazy."

The detection accuracy gap

Audiences claim they can detect AI content. They mostly cannot.

Adobe's 2025 Digital Literacy Survey tested 3,000 participants across 10 countries. Each participant viewed 50 images and 20 video clips, a mix of AI-generated and human-created, and was asked to identify which were AI-generated.

At 54% accuracy for images and 47% for short video, audiences are barely better than random chance at detecting AI content. The false positive rate is equally revealing: 28-35% of human-created content gets incorrectly flagged as AI. This means a significant portion of the "AI content I avoid" that survey respondents report is actually human-created content they mistakenly identified as AI.

For AI-assisted editing (where a human shoots the base footage and AI handles color correction, stabilization, or enhancement), detection accuracy drops to 31%. This category is where most brand video already sits in 2026. According to Wyzowl's 2025 State of Video Marketing survey, 41% of marketers reported using AI tools in post-production. The audience cannot distinguish this content from fully human-produced video.

Why algorithms amplify AI content

The engagement premium on AI content is not entirely driven by audience preference. Platform algorithms play a role.

The novelty signal

TikTok, Instagram, and YouTube Shorts use recommendation algorithms that weight content novelty. Videos that use formats, visual styles, or effects the algorithm has not distributed widely before receive a distribution boost during their initial test phase.

AI-generated content, especially content using newer models like Kling 3.0, Veo 3.1, or Runway Gen-4, produces visual aesthetics that are genuinely novel. Dream-like transitions, impossible camera angles, surreal object transformations. These visuals have no direct precedent in the algorithm's content index, so the algorithm treats them as novel and distributes them to a broader initial test audience.

According to TikTok's engineering research published at the 2024 ACM Conference on Recommender Systems, the recommendation model evaluates approximately 800 features per video. Among those features, visual novelty (measured as distance from the content embedding centroid for the predicted category) is a positive signal for initial distribution.

In plain language: if a video looks different from most content in its category, the algorithm will test it with a larger initial audience. AI content frequently looks different because AI tools produce visual outputs that cameras cannot replicate.

The engagement loop

Once a video gets a larger initial audience, it has more opportunities to generate engagement signals. If the content is visually striking enough to trigger a second watch (loop) or a pause (which algorithms interpret as interest), it accumulates positive signals that push it into wider distribution.

Pentos' Q3 2025 Brand Report found that AI-tagged brand content had a 19% higher loop rate than the brand average on TikTok. The loop rate advantage is significant because loop rate is one of the two highest-weighted signals in TikTok's recommendation model (alongside watch completion rate).

The loop is often driven by confusion or fascination rather than approval. A viewer might watch an AI-generated video twice because they are trying to understand what they are seeing, not because they enjoy it. The algorithm does not distinguish between fascinated rewatching and appreciative rewatching. A loop is a loop.

The comment velocity effect

AI content generates more comments per view than average, driven by two behaviors:

1. Viewers commenting to identify the content as AI ("this is AI," "Sora?," "AI slop")
2. Viewers debating whether the content is AI or real

Both behaviors generate comment velocity, which is a positive distribution signal. A video that generates rapid comments in its first hour of publication gets pushed to larger audiences. The content of those comments (positive or negative) matters less to the algorithm than the volume and speed.

Emplifi's 2025 data showed that AI-tagged videos received 2.1x more comments per 1,000 views than non-AI videos. 43% of those comments were specifically about the content being AI-generated. The algorithm reads those comments as engagement. The viewers see themselves as calling out AI content. The result is that the act of criticizing AI content amplifies its reach.

What the sentiment data actually measures

The 72% negative sentiment from YouGov's 2026 survey requires context. The survey asked: "How do you feel about AI-generated content on social media?" The question triggers a specific mental model: the worst examples of AI content that have been publicly criticized, the uncanny valley face renderings, the six-fingered hands, the generic "AI slop" that floods certain platforms.

But AI content exists on a spectrum. That spectrum includes:

When YouGov asks about "AI-generated content," respondents think of the first category. When Tubular Labs measures AI content engagement, the data includes all five categories. The comparison is between a narrow negative perception and a broad positive reality.

Edelman's 2025 Trust Barometer sectoral analysis found that consumer attitudes toward AI content differed significantly by perceived quality:

The 72% negative headline is real, but it is an average that masks a wide distribution. High-quality AI content generates more positive than negative sentiment. The negative sentiment concentrates heavily on low-quality output.

What this means for brands

The paradox has practical implications for content strategy.

The performance case for AI content

If your primary metric is reach and engagement on short-form platforms, AI content outperforms human content on average. The data supports this. A brand that uses AI tools to produce more visual content, faster, will generate more total engagement per production dollar than a brand that produces less content at higher per-unit quality.

This is especially true for brands that cannot afford to produce 5-7 videos per week (the optimal TikTok cadence per Pentos Q4 2025). AI tools make that volume accessible at $500-$2,000 per month instead of $15,000-$25,000 per month for traditional production.

The reputation risk case against AI content

If your primary metric is brand trust and long-term customer relationship, the sentiment data matters. 61% of consumers in the YouGov survey said brands that use AI-generated content lose their trust. That percentage is higher for certain demographics:

The trust penalty is highest among older, wealthier audiences and B2B buyers. If your target audience falls in those groups, the engagement upside of AI content may not offset the trust cost.

The hybrid resolution

The data points to a clear practical answer: use AI tools for production efficiency while maintaining human creative direction and quality control.

The distinction matters. "AI content" in the survey question conjures a fully automated pipeline: type a prompt, get a video, post it. That is what triggers the 72% negative response. "AI-assisted content" where humans direct the creative, shoot the base footage, and use AI for editing, enhancement, and format adaptation, is what Tubular Labs measures when it reports above-average engagement. The audience generally cannot detect the second category as AI content (31% accuracy per Adobe's data).

The practical framework:

The disclosure question

Should brands disclose AI use? The data does not give a clean answer.

Case for disclosure: Transparency builds trust. Audiences who discover undisclosed AI use feel deceived, which is worse for brand perception than upfront honesty. The EU AI Act and several US state proposals will require disclosure in some contexts starting in 2026-2027.

Case against disclosure: Disclosure reduces engagement by 18-20% (Sprout Social 2025) and increases negative comments by 425%. For brands optimizing for reach, disclosure has a measurable cost.

The practical middle: Disclose when the AI contribution is the primary creative element (fully AI-generated video, AI avatars, AI voices). Do not disclose when AI is a production tool (color correction, stabilization, caption generation, format adaptation). This matches the common-sense threshold: disclosing that you used Final Cut Pro's AI-powered color correction feels absurd. Disclosing that your spokesperson is an AI avatar feels necessary.

External sources:

• Reuters Institute: Digital News Report 2025
• IAB: The AI Ad Gap Widens - Consumer Skepticism Persists (2026)
• Pew Research: How Americans View AI and Its Impact on Society (2025)

Related articles:

• See our complete overview in the AI video production guide.
• Compare tools in our AI video generators comparison.
• Learn when AI fits in our AI video vs real footage decision framework.
• Check quality standards in our AI video quality guide.
• Get platform-specific data in our TikTok algorithm 2026 brand strategy guide.
• See viral mechanics dissected in our viral marketing videos analysis covering 15 campaigns by hook pattern and platform.