Team Members
Ellen Gustafsson, Federico Irace, Kata Horváth, Lawrence Paul Cabrera, Nicole Sanchotene, Andrea Russo (Project Leader)
1. Introduction to the project
Digital platforms frame the consumption of content, both directly and through their infrastructures, sustaining the distribution of targeted advertisements. This project aims to map patterns of advertisements on YouTube, focusing on how targeting strategies vary across age, gender, location and education, through an approach informed by digital methods. Students will draw from a dataset of ads recommended to users on Italian YouTube, which includes the date the ad was served and the user to whom it was recommended. Through a qualitative-quantitative approach, the project aims to analyze how advertisements vary across different demographics and examine the degree of personalization in online advertising, considering whether it operates through individual targeting or broader demographic categories, such as gender or age. The project also explores which kinds of advertisements can be considered ‘omnivorous’ (i.e., aimed at all consumers indiscriminately) and which are ‘specialized’ (i.e., directed toward a specific segment of the population) in their targeting strategies.
2. Research Questions
What kind of advertisements are different categories of demographics exposed to on YouTube?
What are the strategies behind firms’ ad targeting choices?
3. Methodology
Sample
The primary source of data originated from the AlgoFeed Project (https://algofeed.unimi.it/), which provides a structured infrastructure for collecting online platform data through voluntary data donation by users. This approach enabled us to capture a broad and ecologically valid pool of YouTube video advertisements as they appeared in real user environments. The sample of advertisements was developed through voluntary contributions from Italian YouTube users participating in AlgoFeed’s data donation project.
From the broader data collected via AlgoFeed, we were given a random sample of 500 YouTube advertisements for in-depth analysis. The dataset consists of a sample of YouTube advertisements that users reported encountering on the platform. For each video, the dataset includes metadata such as the video ID, URL, and title, along with additional information like duration and automatically extracted named entities. The most important element for our analysis is the categorical tag that we assigned to each video, indicating the intent of the advertisement (e.g., Brand) or the sector of the promoted product, service, or experience (e.g., Food, Fashion, Technology).
This means the dataset combines both structured information (video identifiers, durations, tags) and semi-structured content (titles and entities), allowing us to examine advertisements not only by their descriptive attributes but also by how they are positioned within broader sociological categories of consumption and promotion.
Tagging
Following the collection of the random sample of 500 YouTube advertisements without any accompanying sociodemographic data, the next step of the methodology involved the development of a codebook for systematic categorization.
To analyze these ads systematically, we developed a tagging scheme that allowed us to categorize each video according to its primary purpose and content. Our first distinction was based on intent: we separated ads that aimed to promote a company’s activity and overall identity from those that advertised a specific product, service, or experience. The former were grouped under the category “Brand”, capturing cases where the emphasis was on the business itself, its reputation, recognition, or values, rather than on a concrete offering.
For the remaining advertisements, we adopted a sectoral approach, assigning tags according to the use and type of product, service, or experience being promoted. These categories may indeed reflect the different domains of consumption and cultural practice that advertising engages with. In this way, our tagging scheme combines both an intent-based distinction and a content-based classification, allowing for a meaningful analysis of how advertising may target different aspects of everyday life.
| Tag | Description |
| Brand | Ads that highlight the business and overall image, values, or identity of a company rather than selling a specific product or service. e.g., L’ORÉAL PARIS | Porta in Scena il Tuo Valore |
| Car | Covers advertisements for automobiles, motorcycles, and other motorized means of transport. e.g., Hyundai Tucson HEV 15s NoPromo 1080×1920 Reels NoLogo |
| Consumable | Refers to everyday items that are quickly used and regularly replaced. e.g., POLVERE SU TUTTI I MOBILI? NON CON SWIFFER DUSTER! | SWIFFER. SE LO PROVI, TI CATTURA! |
| Digital Product | Applies to goods or services that exist online, such as apps, software, or online services. e.g., Work Anywhere Device IT Adobe Document Cloud |
| Domestic | Covers durable household goods and furnishings that are typically longer-term investments. e.g., DIVANO 15 SEC YOUTUBE |
| E-commerce | Promotions from online shopping platforms. e.g., Amazon Shipping (IT) 2023 |
| Event | Ads for specific happenings such as exhibitions or fashion shows. e.g., KENZO SS24 RUNWAY SHOW BY NIGO HIGHLIGHT |
| Fashion | Clothing, accessories, and style-related promotions. e.g., PINKO presents Deepest Desires – SS24 Shoes Collection #PINKOSS24 #PINKODeepestDesires #Shoes |
| Food | Food and beverage-related products. e.g., Nuovo Lindt Choco Wafer |
| Hobby | Products and services linked to leisure and personal interests (e.g., travel, cards, board games). Ads appeal to creativity, enjoyment, and lifestyle enrichment. e.g., Bring on Ecuador with G Adventures! |
| Media | Refers to visual media products and services such as television shows, films, and online videos. e.g., The Killer | Netflix Italia |
| Music | Covers songs, albums, and music streaming platforms. e.g., Kollektiv Turmstrasse: YAP Short Clip |
| No Data | Used when the available information about the video was incomplete, unclear, or impossible to classify. e.g., ios_gg_450ac250bc55f27ca9e23f34300d6686_video_0915L007_IT_iOS_GI_Expand_Video_0911.mp4 |
| Self Care | Products and services for personal wellbeing, beauty, and health. e.g., Minéral 89 Crema Idratante 72h |
| Sport | Covers advertisements related to sports as a form of entertainment, focusing on events or sports broadcasting. e.g., F1® is in the air 🇶🇦🏎️ |
| Technology | Covers advertisements for technological devices and hardware that are primarily physical products. e.g., Galaxy Z Flip5 | Samsung |
| Video Game | Promotions for video games, consoles, or gaming services. e.g., Game of Empires: Warring Realm |
Analysis
Only after completing the coding process did we reintroduce sociodemographic data from the AlgoFeed participants into the analysis. This sequencing was a deliberate methodological choice aimed at reducing bias and avoiding circular reasoning. By first working exclusively with the advertisement sample on its own terms, we avoided the temptation to let preconceptions about targeting practices or user profiles shape the coding of ad content. When demographic data such as age groups, gender distribution, or education levels were finally considered, it was solely as a secondary interpretive layer. This allowed us to examine whether certain types of advertisements clustered more strongly around specific user demographics, while ensuring that the foundational categorization of ads remained independent of these factors.
| Province | N |
| Naples | 231 |
| Milan | 60 |
| Varese | 48 |
| Cremona | 37 |
| Bergamo | 36 |
| Avellino | 20 |
| Pavia | 13 |
| Caserta | 13 |
| Salerno | 10 |
| Monza and Brianza | 10 |
| Lecco | 7 |
| Benevento | 6 |
| Brescia | 6 |
| Como | 1 |
| Lodi | 1 |
| Region | N |
| Campania | 280 |
| Lombardy | 219 |
| Parents Education | N |
| High School Diploma | 229 |
| Middle School Diploma | 166 |
| Bachelor’s Degree | 34 |
| Vocational Qualification | 31 |
| Master’s Degree (or Single Cycle) | 15 |
| PhD | 13 |
| Postgraduate Master (I or II level) | 10 |
| No Degree | 1 |
| Education Level | N |
| High School Diploma | 247 |
| Bachelor’s Degree | 162 |
| Master’s Degree (or Single Cycle) | 54 |
| Middle School Diploma | 22 |
| PhD | 11 |
| Postgraduate Master (I or II level) | 3 |
| Age Group | N |
| 35–40 | 231 |
| 25–34 | 136 |
| 18–24 | 132 |
| Gender | N |
| Female | 295 |
| Male | 204 |
4. Findings
After completing the stages of classification and analysis, we observed that the advertisements, when grouped by category, were distributed as shown in the table below.
| Tag | N |
| No data | 96 |
| Brand | 92 |
| Media | 50 |
| Food | 47 |
| Self care | 39 |
| E-commerce | 32 |
| Digital product | 27 |
| Domestic | 21 |
| Fashion | 20 |
| Technology | 20 |
| Video game | 20 |
| Consumable | 10 |
| Car | 6 |
| Hobby | 6 |
| Event | 6 |
| Sport | 4 |
| Music | 3 |
| Home (error) | 1 |
The distribution of gender across advertisement tags shows that Female and Male users appear in roughly similar proportions across most categories, though females are slightly more represented in No data, Brand, Media, and Fashion, and significantly more in Self care, E-commerce, and Food. Males are somewhat more represented in Domestic and Video games. Many categories, especially those with lower counts such as Sport, Music, show balanced and/or low representation between genders. The Brand tag, defined as ads that highlight the overall image, values, or identity of a company rather than selling a specific product, is present for both male and female groups, with a tilt toward females. Most significant differences, with a higher representation of females is visible in the categories Food, Self care, and E-commerce, and for male in Domestic and somewhat in Video Game. Notable is that females are targeted more times with ads in Technology.
Age-based distributions showed clustering of ads. The 18–24 and 25–34 groups appeared equally in multiple categories such as Media, Food, Self Care, Hobby, and Event. Ads in Technology and Car categories were more evenly distributed between 25–34 and 35–40 . Only E-commerce is exposed almost equally to the three age groups.
The distribution of education shows that High School Diploma and Bachelor’s Degree dominate across multiple tags, especially in Brand and Fashion. Ads in Technology and Video Games also show contributions from Master’s Degrees, while Middle School Diploma and PhD are less common. The Sport and Consumable categories are more evenly distributed across lower and mid-level educational backgrounds. Overall, educational segmentation is present, with higher education levels more represented in technology-related categories and lower levels in lifestyle-related ones.
Parental education is most frequently reported as High School Diploma or Middle School Diploma across nearly all categories. These dominate in No data, Brand, Media, Self care, Car and Food. Only in the Food category do individuals with a middle school diploma dominate the ads they are exposed to. Higher parental education (Bachelor’s Degree, Master’s Degree, PhD) appears in smaller proportions, more visible in Technology, Video games. The higher educational level of Masters Degree is only represented in Brand, Media, Food, Technology, and Video Game.
When looking at regional distribution, Campania and Lombardy are the two regions represented. Campania has higher counts in No data, Media, Food, E-commerce, Digital Product, Domestic, Technology, Video Game, Fashion, and Hobby. Lombardy is more represented in Brand, Car, Sport, and Music. Categories like Fashion and Consumable are relatively evenly distributed between the two.
At the provincial level, Naples dominates the distribution across nearly all categories, followed by Milan. Other provinces such as Caserta, Salerno, Avellino, and Cremona appear but with much smaller counts. Naples has particularly high representation in No data, Brand, Media, Food, Self Care, Digital Product, Technology, Video Game and Consumable. Milan appears most significantly in Brand, Self Care and Consumable. Overall, the data shows a strong concentration of ads in a few provinces, with many provinces barely represented.
5. Discussion
The analyzed sample revealed a predominance of advertisements aimed at institutional brand promotion, rather than campaigns exclusively focused on the immediate sale of products or services. This finding suggests that advertising on YouTube, for the audiences considered, largely functions as a strategy to reinforce brand reputation and visibility, operating as a form of continuous presence in the digital environment. Rather than solely stimulating direct consumption, these advertisements appear to serve the purpose of reminding users of the brand’s existence and relevance, consolidating its position in the collective imagination while fostering familiarity and trust over time. This dynamic aligns with Carah and Brodmerkel’s (2021) observation that brands engage with, shape, and capitalize on “algorithmic culture,” leveraging platform infrastructures not only to sell products but to sustain cultural visibility and authority.
Two further points emerge from the distributional patterns observed across tags once sociodemographic variables were introduced post-coding, with implications for how personalization appears to operate on Italian YouTube. First, several categories that might be expected to be strongly gendered, such as Technology or Fashion, showed only modest skews rather than sharp separations, while “omnivorous” classes like Food and Media remained broadly distributed. Second, the prevalence of the Brand and No Data tags highlights interpretive limits but also analytic opportunities. The Brand tag’s prominence may indicate a strategic emphasis on reputation and trust-building in a cluttered attention economy, while the No Data bucket – kept deliberately to ensure coding transparency in our analysis – acts as a reminder that classification granularity is co-determined by creative disclosures, metadata quality, and extraction pipelines, not only by coder judgment.
6. Conclusion
Unlike television, for instance, where the entire audience receives the same advertisements, on the internet, ads are “tuned” to each user based on socio-psychological profiles built in real time through data collection (Carah et al. 2024; Beauvisage et al. 2024). The analyses carried out make it possible to assess, at least partially, how demographic criteria — such as age, gender, or education — are mobilized to promote advertisements to specific audiences. This is a relevant contribution, given that it is currently not possible to precisely observe how the distribution of ads occurs for each individual on the platform. An important limitation, however, is that the research cannot capture the full range of parameters used by recommendation algorithms, which likely include contextual and behavioral variables in addition to the demographic data analyzed. Future research could address this challenge by combining broader datasets, controlled browsing experiments, or algorithmic auditing techniques, thereby enabling a more detailed understanding of how different dimensions of user profiles influence the personalization and distribution of advertisements.
7. References
– Beauvisage, T., Beuscart, J. S., Coavoux, S., & Mellet, K. (2024). How online advertising targets consumers: The uses of categories and algorithmic tools by audience planners. New Media & Society, 26(10), 6098-6119. https://journals.sagepub.com/doi/10.1177/14614448221146174
– Carah, N., and Brodmerkel, S. (2020). Critical perspectives on brand culture in the era of participatory and algorithmic media, Sociology Compass, 2-12. https://doi.org/10.1111/soc4.12752
– Carah, N., Hayden, L., Brown, M. G., Angus, D., Brownbill, A., Hawker, K., .& Robards, B. (2024). Observing “tuned” advertising on digital platforms. Internet Policy Review, 13(2), 1-26. https://policyreview.info/articles/analysis/observing-tuned-advertising-digital-platforms
– Punziano, G.; Gandini, A.; Caliandro, A.; Airoldi, M.; Padricelli, G.M.; Acampa, S.; Trezza, D.; Crescentini, N.; Rama, I. 2024. Eliciting and Retrieving the Feedback-Loop. Exploring Elicitation Interview Techniques for Detecting Algorithmic Feedback on Social Media and Cultural Consumption. In: 6th International Conference on Advanced Research Methods and Analytics (CARMA 2024). Valencia, 26-28 June 2024. https://doi.org/10.4995/CARMA2024.2024.17835
– Punziano, G.; Gandini, A.; Caliandro, A.; Airoldi, M.; Padricelli, G.M.; Acampa, S.; Trezza, D.; Crescentini, N.; Rama, I. 2024. The Algofeed project. A methodological proposal to assessing the effects of algorithmic recommendations on platformized consumption.. In: 6th International Conference on Advanced Research Methods and Analytics (CARMA 2024). Valencia, 26-28 June 2024. https://doi.org/10.4995/CARMA2024.2024.17834