The study began with what the supplied workbook could legitimately support. Its primary measure was follower-normalised like intensity, or FLI, calculated as likes divided by recorded followers and multiplied by 100. This metric was available for 9,968 TikTok records and 10,127 Instagram records. FLI is not a conversion rate, a reach estimate or direct evidence of algorithmic recommendation: it compares recorded likes with recorded account size, and follower counts may have been captured after publication. The workbook also lacked shares, saves, watch time, completion rate and recommendation exposure, while comment counts could not be independently identified, preventing the construction of broader interaction ratios.

A central audit finding concerned the column labelled “views.” Embedded headers in the TikTok sheet described the same column as “comments,” leaving its identity unresolved. Among records with numeric likes and a positive value in that field, likes exceeded the recorded value in 92.7% of TikTok cases and 83.3% of Instagram cases. Those comparisons diagnose the supplied data rather than proving that native platform view counts were wrong, but they rule out treating the field as a reliable exposure denominator. The researchers consequently limited it to diagnostic summaries and did not calculate like-to-view engagement ratios, even where a like count did not exceed the recorded field value.

The analysis used a seven-step protocol covering variable availability, platform validity, timestamp validity, denominator validity, ratio construction, hashtag classification and distributional shape. Rows without account identifiers, embedded headers and Facebook-labelled entries were removed, while account identifiers were standardised by lower-casing and trimming whitespace. The resulting sample represented 3,996 distinct recorded TikTok accounts and 4,691 Instagram accounts, although the workbook lacked verified post identifiers and documentation of the extraction tool, dates, search terms or original inclusion query. The data were therefore treated as dataset-defined records associated with tourism content rather than a probability sample or a controlled collection of verified posts.

Distributional diagnostics produced the study’s clearest platform contrast. The median FLI was 12.63 on TikTok and 1.77 on Instagram, and a Mann–Whitney U test indicated that TikTok records generally occupied higher ranks in the distribution, with a rank-biserial effect of 0.317. The mean told a different story: Instagram’s mean FLI was 922.53, compared with 201.57 for TikTok, because Instagram contained extreme outliers. A supplementary Welch t-test consequently pointed toward higher Instagram means. The result does not support a simple platform ranking: TikTok records displayed stronger typical intensity relative to followers, whereas a smaller number of Instagram posts achieved exceptional values. A higher mean therefore did not mean that the typical Instagram post performed better.

Two sensitivity checks preserved the direction of the median comparison. When FLI was aggregated to the median for each recorded account, TikTok’s account-level median remained higher than Instagram’s, at 18.11 versus 0.98, with a rank-biserial effect of 0.484. A separate analysis removed exact repeated records—900 additional TikTok rows and 311 additional Instagram rows matching earlier entries across eight fields—and again found higher median FLI on TikTok, at 12.50 versus 1.72. These checks strengthen the descriptive contrast, although they do not correct all dependence among posts from the same account, and the remaining record-level significance tests do not account for clustering. Without post identifiers, identical entries also cannot conclusively be distinguished from separate posts carrying identical recorded information.

Hashtag comparisons did not detect a reliable targeted-versus-generic advantage on either platform. Researchers extracted only tokens preceded by a hash symbol, converted them to lower case and classified records as targeted-only, generic-only, mixed or other/uncoded. Targeted-only records had higher descriptive median FLI on both TikTok, at 14.75 versus 12.57 for generic-only records, and Instagram, at 4.05 versus 2.74. Neither difference was statistically significant, with p values of 0.320 and 0.233, respectively. These findings do not establish equivalence or prove that hashtags are irrelevant; exact dictionary matching and uneven extraction coverage, especially for TikTok entries without a hash prefix, limited the scope of the classification.

Account size also related differently to engagement across the platforms. Recorded followers were strongly associated with raw TikTok likes, with a Spearman correlation of 0.660, but only weakly associated with Instagram likes, at 0.087. Descriptive ordinary least-squares models of log-transformed likes and followers yielded coefficients of 0.690 for TikTok and 0.110 for Instagram, with R-squared values of 0.433 and 0.007. These models account for associations rather than causal effects. They show that TikTok could display high typical FLI while raw likes remained strongly linked to audience size, undermining any claim that TikTok circulation operates entirely outside follower networks. The negative correlations between followers and FLI, at −0.330 for TikTok and −0.502 for Instagram, also require caution because followers form the metric’s denominator.

Temporal and textual analysis was restricted because the available records did not support it reliably. Recorded TikTok dates ranged from 2018 to 2025, plus one value parsed as the impossible year 2206, while Instagram included a date before the platform launched in 2010. Because the source of these anomalies was undocumented, the researchers did not conduct period-matched or longitudinal comparisons. Inspection also found apparent caption misalignment in parts of the TikTok sheet, so caption-based sentiment analysis was excluded rather than assigning text to records by assumption. The final results therefore concern platform-conditioned engagement traces, not feed position, recommendation exposure, tourist persuasion, destination effectiveness or causal effects of platform architecture and hashtag use.

The study’s broader message is methodological rather than a verdict on tourism campaigns. Auditing can identify missing variables, inconsistent fields, invalid denominators, ambiguous categories and skewed distributions, allowing researchers to determine where interpretation should stop. It cannot reveal how content was ranked, whether users encountered posts through recommendations, followers, search or external sharing, or which collection process produced a discrepancy. The UAE and Dubai case is also bounded: it lacks variables describing content format, creator type, account verification and institutional versus personal use, all of which could influence both audience size and engagement. The defensible lesson is that platform metrics must be reconstructed and tested before they are treated as comparable evidence of visibility or success.

Subject of Research: Auditing follower-normalised engagement metrics in TikTok and Instagram records associated with UAE and Dubai tourism content.

Article Title: Auditing platform metrics for communication research: a cross-platform case study of TikTok and Instagram engagement

Article References: Salim Humaid Kattarah Al Shamsi, M., & Ragnedda, M. (2026). Auditing platform metrics for communication research: a cross-platform case study of TikTok and Instagram engagement. Frontiers in Communication, 11, Article 1921355. https://doi.org/10.3389/fcomm.2026.1921355

Image Credits: AI Generated

DOI: 10.3389/fcomm.2026.1921355

Keywords: social media metrics, TikTok, Instagram, tourism communication, data auditing, engagement analysis, platform studies, UAE tourism, Dubai tourism, methodological research, Auditing, platform