The traditional, and more frequently used, measures of advertising effectiveness ascribe to memorability (the ad as a whole, message, any details) and attitude (towards the ad and the brand). However, these measures are basic, often not specific enough, and in some cases may be deemed superficial. For example, they do not capture attention and engagement effectively and accurately. Especially with respect to visual ads (still and video, by different channels), those measures do not reflect attention to and impressions of various visual elements or features and their composition in the ad’s image or ‘frames’ while consumers are viewing the ad. In the past twenty years a collection of implicit measuring techniques for advertising effectiveness have been added to explicit measures (survey-based, self-report) in the toolbox of researchers; the newer methods include eye-tracking, facial expressions, biometrics, and neurophysiological methodologies such as EEG and magnetic resonance imaging (MRI).
A general distinction can be made between: (1) ad testing in which measures are taken during the planning, design and creation of ads for assessing consumer response to the ad pre-publicity and predicting ad effectiveness; (2) measuring ad effectiveness post-publicity for evaluating performance of the ad in the ‘real-world’, usually among larger audiences exposed to the ad concerned. On either occasion researchers may use qualitative and quantitative, explicit and implicit measuring techniques. In recent years the utilisation of implicit measures has been increasing. Yet this should not exclude the continued use of explicit measures, and there can actually be great value in applying those different measurement classes in parallel (complementary or inter-related).
- Venkatraman and his colleagues (2015, [1]) constructed a research plan with the goal of linking traditional measures (explicit, self-report) with newer advanced measuring methods in advertising research (implicit, independent). They combined traditional measures (e.g., recall, liking, purchase intent) with implicit measures based on direct response (Implicit Association Task), eye-tracking, biometrics (e.g., heart rate, skin conductance), and neurophysiological methods, specifically EEG and functional MRI. They tested and compared between the measures in regard to 30 second TV ads. Subsequently, they investigated to what extent the advanced implicit measures explain variance in advertising elasticity (i.e., sensitivity of sales to scale of advertising) beyond traditional measures (fMRI came out as the most effective).
This post will look into insights learned on consumer attention and engagement when viewing ads in different visual formats and channels. We will be doing so with the help of a webinar presented by Affectiva (Media Analytics) at iMotions, titled: “The Changing Face of Advertising: What 100,000 Ads Reveal about Effective Campaigns“ (November 2025). The focal method of interest is the detection and analysis of facial expressions in response to ads. While eye-tracking is concerned with measuring aspects of attention, facial expressions are detected for identifying emotional responses, that is, tracing (types of) emotions experienced by viewers. Emotions are linked to engagement. In over 15 years Affectiva tested thousands of ads using advanced facial analysis technology in which it specializes. This webinar marks the achievement of reaching above 100,000 ads tested and evaluated based on facial expressions. The presenters review some of the key findings and insights on consumer responses and the effectiveness of visual ad campaigns, addressing digital advertising and the use of AI.
- The research firm iMotions is grounded in eye-tracking techniques and analysis, yet it additionally employs other types of sensor-based measurements (e.g., biometrics, EEG). Affectiva, which joined iMotions in early 2025, brought its expertise in facial expression measurement and analysis, aided by AI-based visual analytic tools. Measurement for capturing facial expressions is enabled with a camera (e.g., web-camera, embedded in a laptop or smartphone) or any other optical sensor.
The initial presentation by Serena Pang (Research Manager) examines how the reaction of audiences to advertising has changed from 2014 to 2025. For a start, ads have induced increasing emotional engagement over this period, as reflected from facial expressions of viewers. However, it is also revealed that the valence of emotions evoked has turned from being in the positive dimension to the negative dimension of emotions (based on classification of expressed emotions). These trends were traced across countries (shown for top 10 countries).
Getting into more detail, there has been an increase in most (types of) facial expressions, shown for example of brow raise, brow furrow (frown), confusion, jaw drop, and mouth downturn. Yet there has not been a rise in occurrence of smiles over the years (a slight nominal decrease). Pang clarifies that it is not that consumers-viewers are smiling less but that they experience negative emotions more frequently. Furthermore, the range of emotions (’emotional palette’) induced by TV ads widened between 2014 and 2025. Thus, for example, the incidence of joy, the highest in 2014, was moderated by 2025. On the other hand, the incidence of the other emotions displayed, which may be characterised as neutral or negative, has increased over these years — they include sentimentality, sadness and confusion (the greater increases, and getting closer to the level of joy in 2025), followed by contempt, anger, fear, surprise, and disgust (the latter five least occurring in both 2014 & 2025).
Graham Page (Global Managing Director) reviews and discusses further issues arising from their analyses (part of them conducted by Kantar research group) over those years. In opening his review, he concurs with Pang with regard to the emotional palette, suggesting that the findings do not mean that people are necessarily sadder, rather that the video (TV) ads induce more varied emotions. He shows, for instance, what happens to the frequency of smiles during the progression of screening two video ads (Cadbury, from 2018 and 2024) in comparison: the flow and level of incidence of smiles for the two ads are quite close and similar (i.e., observing how the shape of those curves changes from start to end of the video clip). In particular, in both cases the rate of smiles increases and reaches its top level in the ending scene (more salient in 2018). People continue smiling in response to ads. While there is overall evidence of less positivity, Page remarks that it does not mean that viewers are more miserable (note that COVID-19 happened in between those two points in time).
Nonetheless, from the more concrete business perspective, figures on sales performance associated with ads suggest that “increased negativity may not be a good thing”. Page shows a matrix comparing the proportions of high performing ads across two dimensions of positivity and negativity of ads:
| Less Positive Expressions | More Positive Expressions | |
| Less Negative Expressions | 48% | 59% |
| More Negative Expressions | 30% | 39% |
Therein, ads eliciting more positive expressions and less negative expressions exhibit the highest proportion of high sales performing ads. Page concludes accordingly that more positive emotions (overall) are more sales effective. Notwithstanding, this table suggests even more pointedly that eliciting the less negative expressions has the stronger positive effect (compared to more positive expressions) on sales performance of the video ads. That is, particularly less negative emotions could be more conducive for sales effectiveness of advertisements.
Digital Advertising
Digital advertising is on the rise, becoming more prevalent and gaining more exposure. Page demonstrates key data-driven observations. Firstly, and prominently, attention and engagement are not the same: Engagement is not correlated with attention, whereby receiving attention (looking) does not necessarily result in getting the viewer engaged with the ad (seeing). Hence, optimising for attention does not guarantee that consumers-viewers will also react and engage with the ad more intensively or deeply (refers to “all formats” digital).
- From the viewpoint of Page and Pang, engagement appears clearly to be conceived as emotionally driven, which corresponds with the focus of Affectiva on facial expressions as manifests of emotions. However, engagement may also have a cognitive driving force. In particular, engagement of cognitive nature may be generated by special interest in the ad content and brand, intrigue or curiosity, and resolving cognitive challenges (e.g., identifying an analogy, interpreting a metaphor, understanding a joke or humorous punch, solving a riddle or puzzle).
- In drawing on the theoretical framework configured by MacInnis and Jaworski (1989, [2]) for information processing from advertisements, it is suggested that engagement with the ad, and moreover the advertised brand, can be expected to entail both cognitive and emotional responses as elements of brand processing. The forms, scope and degree of cognitive and emotional responses are determined by the level of processing (six levels posited), being moderated by attention and capacity (contingent on motivation as initial trigger). Following this model, engagement may genuinely come about at the top three levels of processing (moderate, high and very high) wherein processing becomes more detailed, comprehensive and deeper. Nevertheless, it is noted that cognitive responses may evoke emotional responses in consequence (e.g., thrill, surprise, enjoyment, confusion).
Page brings up additional related insights: Attention and engagement seem to conflict with each other, whereby attention falls as the ad video goes longer whereas engagement grows as the video gets longer. Hence, while attention depletes within the first 50 seconds, engagement builds up as viewers watch the ad video longer, generating more (emotional) responses. Using more dynamic sensory stimuli, in image and sound, tends to capture more attention (e.g., dynamic visuals and music retain attention, a still ‘slide’ image loses attention). Especially important is the narrative of a video ad for attracting viewers and getting them more engaged. That can be achieved by developing a story in the video, demonstrating relevance and novelty, or creating an emotional punch. Interestingly, Page shows that comparatively, levels of attention and engagement for TV and digital-online ads (2023-2025) are similar, yet valence of emotions in digital ads falls lower than in TV ads. He points further to shifts in length of TV ads (shorter) and in pace of TV and digital ads, faster and more dynamic (global, 2014 to 2024).
In summary of the discussion on digital advertising: Capturing greater attention is not sufficient, and what works to achieve it may not work well for increasing engagement and could actually harm it. It is advised for ads to be designed in ways that more often raise smiles in their ending scenes. It is also noted in the presentation of Page that styles of “digital” ads may not fit well in all formats and channels of advertising (e.g., length, speed, narrative).
Impact of AI
Pang notes that the use of AI in advertising is increasing during the design of ads in different visual formats, relating in particular to Generative AI (Gen AI) tools. There is concern, however, that the impact of AI may involve a weakening of the emotional connection of consumers with the ad and the brand advertised.
Pang compares between video ads that apply Gen AI with ads that do not use this technology with respect to its impact on emotions expressed by viewers (on TV and digital channels). Primarily, video ads that apply Gen AI gain overall greater engagement than non-Gen AI ads, though the valence is less positive with response to Gen AI ads (e.g., higher levels of expressed confusion, sadness and brow furrow). Indeed, viewers express smiles and surprise in response to the Gen AI ads, but she remarks that the balance with regard to Gen AI is overall a little more negative.
We come now to a critical issue about the implications of using Gen AI, which may also raise some controversy. When comparing the performance of ads, Pang shows that in ads where the use of Gen AI is more obvious (i.e., the intervention of AI is easier to identify), their performance tends to be lower compared with ads wherein the application of Gen AI is not obvious (i.e., appearing seamless to viewers, better integrated into visuals). When Gen AI is obvious, ads ‘shift’ to the lower performance class, whereas ads wherein Gen AI is not obvious ‘shift’ to the medium and high classes:
| Gen AI Obvious | Low 35% | Medium 34% | High 30% |
| Gen AI not Obvious | Low 17% | Medium 43% | High 40% |
Pang clarifies that when a video is created by Gen AI in a way too obvious to consumers-viewers it may look awkward to them and raise disbelief (evoke ‘nose wrinkle’). The viewers may still smile while watching, yet the advertiser has to make sure that the audience is “smiling with the ad and not at the ad”.
Pang cautions in particular of video ads in which AI is visibly distracting or disturbing to viewers. There is, however, a potential fault in ads wherein AI is hard to discern by a ‘naked eye’. On the one hand, it is reasonable that when the ‘touch’ of Gen AI in images is obvious, it would be met with lower credibility, perhaps resentment, from viewers, and would garner lower trust. On the other hand, ads wherein consumers remain unaware of the use of Gen AI might be implicated in causing deception and undesirable illusions that mislead viewers into forming mistaken beliefs. The visibility of AI has to be moderated to balance between seamless ‘natural’ appearance of the visual images and respect for transparency and authenticity.
- As an example for good illustration, watch the humourous video ad for “Liquid Death: Mountain Water” (@ time 35:00-36:30 min:sec). A curve drawn during screening depicts how exhibition of smiles increases, starting more hesitantly before peaking in the last third of video clip (near the punchline moment), then fall steeply with the final slide. The joke in the ad starts funny, but then it feels like stretching its point too long with harsher scenes. Pang demonstrates that viewers who said they most enjoyed the ad (“top box” responses) exhibited or surfaced more smiles than the other viewers — the former viewers found the ad sincerely enjoyable and funny as corroborated by their smiles, whereas all others were smiling much less all along. Pang notes that this ad ranked at bottom 15% of all ads on ‘enjoyment’ and concludes that the ad was apparently funny for some viewers but was overall “polarizing”.
- The final slide presented the tagline “Arrestingly Refreshing” and stated beneath “Made with AI”, a fair disclosure. Some viewers might have been disappointed and felt uncomfortable about AI, yet it seems more likely that too many viewers just did not like the rather cynical, even crude humour, which does not appear to be in good taste. Could the Gen AI agent not have a fine sense of humour, and does not know when to stop?
The research approach of facial expression analysis is based on interpreting facial expressions (facial coding) and associating them with corresponding classes of emotions. However, the approach is not free of critiques. In analysing facial expressions, a question has been raised whether they are universal, that is, the same expressions are identifiable across facial features and socio-ethnic groups and linked to the same emotions. The inference of emotion evoked may be influenced by contextual factors such as situation, culture, time and place, and also individual-level factors such as personality and goals. Affectiva is reasonably applying AI-based algorithms and methods to overcome such challenges, with aim to improve the accuracy of interpreting or coding the facial expressions and inferences of the underlying emotions.
The webinar presentation by Page and Pang of Affectiva – iMotions provides us with an instructive and revealing consumer-driven portrayal of advertising and the developments in this domain over the past ten years. It reinforces mainly the view that the practice of advertising is a mixture of art (creation) and science (research). Advertisers and advertising professionals have to keep watching closely after crucial aspects of consumers’ reactions (e.g., attention, engagement), consider the fit and effectiveness of strategies and styles of ads in different formats and channels (e.g., TV, digital — online, mobile, social media), and take care to avoid pitfalls in use of advanced technologies (e.g., Gen AI). There shall be no dull moment for researchers and practitioners.
Ron Ventura, Ph.D. (Marketing)
References:
[1] Predicting Advertising Success Beyond Traditional Measures: New Insights from Neurophysiological Methods and Market Response Modeling; Vinod Venkatraman, Angelika Dimoka, Paul A. Pavlou, Khoi Vo, William Hampton, Bryan Bollinger, Hal E. Hershfield, Masakazu Ishihara, & Russell S. Winer (2015), Journal of Marketing Research, 52 (August Special: Neuroscience & Marketing), pp. 436-452
[2] Information Processing from Advertisements: Toward an Integrative Framework; Deborah J. MacInnis and Bernard J. Jaworski (1989); Journal of Marketing, 53 (October), pp. 1-23
A Gateway to Consumer and Customer Behaviour 