The campaign showed a burger and a woman resembling Queen Letizia. Comments concentrated on another part of the message: gluten and traces. Research from the University of Alicante, published on 30 June 2026, reconstructs that conversation around a 2023 VICIO launch. Its frequency table records “gluten” 497 times and “trazas”, meaning traces, 297 times. The case offers a concrete reading of social listening: an activity designed to win attention can end up organising discussion around the conditions attached to using the product.
The paper appears in Redmarka. Revista de Marketing Aplicado and is written by Estela González Barrachina, Christian Fortanet van Assendelft de Coningh, Tatiana Hidalgo-Marí and Cristina González Díaz. It examines reception of an Instagram post through word frequency, polarity and emotion classification. Its results describe the language in the extracted comments. For marketing, the interest lies in identifying which concerns became prominent and how a selection of tools turns a discussion thread into structured information for interpreting the relationship between a message and its audience.
A 2023 post examined in research published in 2026
The researchers place VICIO's founding in Barcelona in 2020, by Aleix Puig and Oriol de Pablo, and describe a proposition built around product, technology and brand. Their account highlights a combination of traditional and digital communication with a creative, provocative personality. At the end of 2023, the period to which the study relates, the Instagram account had approximately 200,000 followers. That historical reference explains the choice of channel within the research and places the scale of the community then being observed.
The business profile presented in the paper includes technology and logistics management alongside Glovo. It also recalls VICIO receiving the 2023 The Best Digital Restaurant Virtual award within the awards promoted by specialist publication DiegoCoquillat.com. That recognition concerned digital management in the restaurant sector.
The selected post dates from 23 October 2023. The carousel combined images and videos, with an advertising banner as its cover announcing a gluten-free burger. The woman holding the product resembled Letizia. According to the context reported in the study, the brand presented the launch as the outcome of three years of research responding to demand for a gluten-free option. The announcement brought together two registers: immediate recognition of an image with public associations, and a product proposition addressed to a specific need.
The response prompted clarifications that the academic paper reconstructs from reporting published at the time. VICIO explained that the woman was actor Aida de Sàrraga and attributed the warning about possible traces to human errors that could occur during preparation. That explanation forms part of the company's historical response. Within the conversation studied, the traces question acquired a significance of its own: the expectation created by the product description coexisted with questions about the information accompanying the advertisement and what it meant for the people commenting on it.
The study places this discussion within Instagram’s different uses by audiences and professionals. Drawing on IAB Spain’s 2025 research, it cites entertainment at 81%, social interaction at 66% and access to information at 55% among user purposes. For professionals, the cited figures are 53% for strengthening the brand and 44% for customer service. These are figures quoted by the academic authors as channel context, involving different purposes and profiles. They provide a backdrop to the choice of a comments-based study: an account communicating a product can encounter entertainment, information-seeking and requests for attention in the same public space. VICIO’s selected conversation contained specific questions about traces and product information, while its mention table identified associations and other accounts being addressed. The authors’ introductory context helps explain why the investigation combines vocabulary analysis with a separate examination of mentions, rather than reading comment volume alone as a measure of the announcement’s reception.
Why the most-commented post was excluded
The sampling design is particularly revealing. The authors reviewed 136 posts from that year up to the study's observation point, using comments as the comparative measure. The most-commented one, with 9,711, sought ambassadors and asked users to mention other people. It was excluded because that call to action shaped the content of the responses. The second-ranked post, with 1,753 comments, was selected instead: the gluten-free launch. The research thereby distinguishes interaction induced by participation mechanics from a conversation that could contain assessments of a product and the way it was communicated.
The extraction tool, PhantomBuster's Instagram Post Commenters Export, recovered 1,169 comments. The researchers explain the difference from the 1,753 total by the function's exclusion of replies to comments and GIF-format contributions. The analysed collection is therefore the text that this function could export under those conditions. That matters when interpreting the available conversation: a reply within a thread or a visual reaction can be part of the public experience of the post while remaining outside the material entering textual analysis.
The data were exported to Excel in two columns, identifier and text, and then processed in RStudio. To analyse language, the authors removed the identifier column and retained comment content. Preparation included removing punctuation, numbers, links and surplus spaces, alongside standardising upper- and lower-case text. They then split the text into words. That sequence establishes a useful separation between the record of a contribution and the unit eventually counted in the charts: a word can recur several times, while one comment can supply many different words.
The 1,169 comments produced 14,528 word observations, initially grouped into 2,801 terms. Further cleaning used a list of 608 Spanish functional words and left a table of 2,384 terms; the total frequency presented in that table is 5,980 occurrences. These represent different stages of corpus preparation. Vocabulary selection determines what remains available for later analysis: connectors and other expressions with little thematic value are removed, while terms associated with the product or assessments of its advertising retain their weight in the count.
The vocabulary that changed the advertisement's centre
“Gluten” and “trazas” clearly lead the table. They are followed at a distance by “salud”, health, with 80 occurrences; “contener”, contain, with 64; “hamburguesa”, burger, with 63; and “enfermedad”, disease, with 62. “Publicidad”, advertising, records 61; “celíacos”, people with coeliac disease, 52; and “personas” and “vergüenza”, people and shame, 50 each. The concentration shows which message elements recurred in the material. The ingredient and the possibility expressed by the warning stand ahead of more general product references.
The corpus’s ten most frequent words
Frequencies of the leading terms in the study’s cleaned table, derived from 1,169 comments exported from a 2023 VICIO post.
occurrences
Tap or hover for details. All values are also available in the data table.
Collapse notes
Table 1, page 133. Word counts; a comment can supply multiple words. The 2023 corpus includes 1,169 comments and excludes replies and GIFs. It does not represent all customers or Spain’s population.
Explore the data table
| Category | Word occurrences |
|---|---|
| gluten (gluten) | 497 occurrences |
| trazas (traces) | 297 occurrences |
| salud (health) | 80 occurrences |
| contener (contain) | 64 occurrences |
| hamburguesa (burger) | 63 occurrences |
| enfermedad (disease) | 62 occurrences |
| publicidad (advertising) | 61 occurrences |
| celíacos (coeliac) | 52 occurrences |
| personas (people) | 50 occurrences |
| vergüenza (shame) | 50 occurrences |
Other words add perspective. “Letra” ranks 24th and “pequeña” 29th, while “reina”, queen, appears in 63rd place. The authors interpret that distribution as greater attention to the banner's informational message than to its creative or symbolic device. This is a finding specific to the case: the recognisable image could provide an entry into the conversation, while the most repeated words concerned the detail shaping interpretation of the product. Visual prominence and the subject attracting questions do not necessarily coincide.
The researchers also produced a word cloud containing terms appearing at least 35 times, fifteen in total. Alongside gluten and traces were concepts such as contamination and cross-contamination. That representation reinforces the table's thematic reading, while the ranking provides the scale of individual terms. In their discussion, the authors connect this vocabulary with concern about information accompanying a product aimed at consumers with specific needs. It is an interpretation of the conversation that helps identify the subject towards which criticism was directed.
More directed mentions than campaign hashtags
Hashtags had a small presence: four different tags and five appearances. Two belonged to #singluten, while #MarketingFail, #publicidadengañosa and #trazas appeared once each. The contrast with mentions is marked: the study records 184 different accounts mentioned and 249 appearances overall. These expressive tools serve different purposes. A hashtag can group discussion around a topic, whereas a mention points towards a specific interlocutor. In this collection, the authors observe a conversation more oriented towards addressing identifiable actors than organising itself through hashtags.
VICIO's account, @ganasdevicio, received 20 mentions. It was followed by @faceceliacos with 17, @celiacs_catalunya with 14 and @celiacosmadrid with seven. Institutional accounts also appeared, including @sanidadgob and @consumogob with four and three occurrences respectively. The table additionally includes other brands and media outlets. This map shows towards whom contributions were directed: the business itself, associations connected with coeliac disease and other public accounts. Naming an account in the thread describes an action by the commenter, independently of any eventual response from that recipient.
The remaining named entries give that map further detail. The paper records @ocuconsumidores twice and @a3noticias once, alongside other restaurant-sector brands: @ardeburgers appears three times and @pinksmadrid twice. The authors read the spread of mentions as an effort to address actors with a possible response or influence, including the brand, associations, public bodies and media. These references came from the original comment text, which retained the @ symbol for this part of the research. Hashtags were extracted separately, preserving the # symbol, before the two frequency tables were assembled.
What lexical polarity measures
To classify vocabulary, the team chose a dictionary method rather than machine-learning alternatives, citing ease of application, low computational cost and interpretability. It used Spanish versions of the AFINN and NRC lexicons adapted by Juan Bosco Mendoza Vega. AFINN assigns words an orientation and intensity on a scale from minus five to five; NRC links them with eight emotions. These offer different readings of the same material: the first organises positive or negative connotation, while the second broadens the range of expressive categories.
The authors describe specific processing for each dictionary. For AFINN, they aligned relevant columns to match terms, assign connotations and count frequencies, then separated the most repeated words within each polarity. For NRC, they removed the general positive and negative categories included in the resource and retained the eight emotions. That choice gives the third stage a different purpose from the second: detailing emotional associations rather than repeating the same binary division.
The aggregate AFINN result reports 340 lexical units with negative connotations and 259 with positive connotations, out of a stated total of 599. The negative side is larger in that count. The comparison concerns vocabulary classified by the dictionary rather than a vote by 599 consumers. That distinction changes its commercial interpretation: the chart reveals an orientation in the selected language, while complete opinions still depend on the sentences, questions and relationships between words composing each contribution.
The study’s AFINN lexical classification
The study reports 340 negative and 259 positive lexical units, 599 in total in its AFINN classification. These are the original reported counts.
lexical count
Tap or hover for details. All values are also available in the data table.
Collapse notes
Figure 4 and text, page 137. Counts of vocabulary classified by AFINN, with 599 units overall. These categories describe words rather than a proportion of customers.
Explore the data table
| Category | Reported lexical count |
|---|---|
| Negative | 340 lexical count |
| Positive | 259 lexical count |
The AFINN breakdown restores content to the aggregate. In the negative category, “enfermedad” leads with 62 occurrences, followed by “mal”, bad, with 40; “moda”, fashion, with 26; “triste”, sad, with 21; and “vergonzoso”, shameful, with fifteen. On the positive side, “gracias”, thanks, appears 25 times; “seguro” and “favor”, safe or sure and favour, 22 each; and “grande”, great or large, 18. The researchers associate some criticism with concern about presenting coeliac disease as a fashion, while also identifying appreciation for introducing a gluten-free option.
The emotion classification supplies another distribution. The study reports 710 lexical assignments to fear, 566 to sadness, 468 to disgust and 381 to anger. Trust records 276; joy, 139; anticipation, 138; and surprise, 123. The figure orders categories into which the dictionary places vocabulary. A word can belong to several emotions: the analysis itself puts “enfermedad” within fear, disgust, anger and sadness. The chart's interest consequently lies in the profile of those assignments and the terms explaining them, rather than dividing participants into eight mutually exclusive groups.
NRC emotional associations of the vocabulary
Published counts for the eight NRC emotional categories applied to the comment corpus concerning VICIO’s post.
lexical assignments
Tap or hover for details. All values are also available in the data table.
Collapse notes
Text and Figures 6–7, pages 138–139. A word can belong to multiple emotions. These are dictionary associations rather than exclusive groups of commenters.
Explore the data table
| Category | Reported lexical assignments |
|---|---|
| Fear | 710 lexical assignments |
| Sadness | 566 lexical assignments |
| Disgust | 468 lexical assignments |
| Anger | 381 lexical assignments |
| Trust | 276 lexical assignments |
| Joy | 139 lexical assignments |
| Anticipation | 138 lexical assignments |
| Surprise | 123 lexical assignments |
The NRC detail also illustrates why context remains important. “Vergüenza” records 50 occurrences under surprise, “real” 32 under trust, “anuncio”, advertisement, 19 under anticipation, and “seguro” 19 under joy, according to the published associations. The categories come from the dictionary, whereas a writer's intention depends on the statement. The paper recognises ambiguity, multiple meanings and the difficulties of working with a translated lexicon. Reading tables alongside text keeps the substance of the discussion visible behind an apparently straightforward emotional label.
The commercial value lies in identifying what requires a response
The paper's discussion places informational clarity at the centre of this case. The brand uses provocative communication, but this launch invited scrutiny of a condition particularly relevant to its intended audience. Mentions of traces, small print, disease and coeliac consumers make that tension specific. For a team analysing digital conversation, the most useful finding goes beyond polarity's sign: the vocabulary identifies the issue motivating the reaction, while mentioned accounts show towards whom requests for attention were directed.
The work acknowledges its focus on one product, one post and one platform. Its selection prioritises a highly commented publication, and its extraction excludes replies and GIFs. It is a case study with that configuration, examining a historical communication activity. The authors propose extending the analysis across channels and moments, including the brand's later response, to understand how the dialogue develops. This opens a specific research question: how the content of contributions changes when a company clarifies the details that concentrated the initial doubts.
Who's who and how to read the figures
VICIO is the restaurant brand examined; the academic paper places its founding in Barcelona in 2020. The four authors belong to the University of Alicante, and Redmarka published their study on 30 June 2026. PhantomBuster supplies comment extraction and RStudio the processing environment. AFINN and NRC are the dictionaries used for classification. These tools belong to the research described, performing different functions within the collection and analysis sequence.
The 1,753 comments concern the selected post's volume as reported by the study; 1,169 is the exported and analysed collection. Word occurrences count repetitions of vocabulary, and emotional categories can share terms. The figures describe the studied corpus, its vocabulary and the associations assigned by the dictionaries. With those units, the case follows how reaction to an advertisement combines questions about what is offered, assessments of its communication and attempts to obtain a response.
In this article
Marian Vasile
Marian Vasile is a MarTech and growth professional. His work spans product marketing, SEO, paid advertising and deep online tracking, including behavioral analytics and attribution.
