The study was conducted in the rural and peri-urban reception area of Botlokwa FM, a licensed community radio station in Limpopo Province that broadcasts primarily in Sepedi and targets low-income listeners with limited access to fixed broadband and formal health services. Botlokwa FM carries a daily health bulletin that regularly promotes provincial SMS hotlines for counselling and appointment booking. One frequently aired Sepedi language bulletin on non-communicable disease screening, which included a real provincial SMS hotline number, was selected in consultation with station staff and re-recorded by the usual presenter for experimental use.

Guided by an integrated framework that treats auditory health literacy as a function of both individual capacities and infrastructural design, the study adopted a cross-sectional, convergent mixed methods design (Creswell & Plano Clark, 2018). The quantitative component used a short survey experiment to test how bulletin structure affects recall of hotline details, while the qualitative component used follow-up interviews to examine listening practices and perceived barriers. This design operationalises the concept of “mode as filter” by treating experimental manipulations of bulletin scripting as interventions in the communicative conditions under which listeners exercise agency in relation to mobile health invitations.

The sampling frame comprised adult Sepedi-speaking residents (aged 18 and older) in three villages within Botlokwa FM’s strongest signal area, identified through coverage maps and consultations with local leaders. No external panel provider was used. Instead, participants were recruited through on-air announcements, WhatsApp messages distributed via existing community groups, and in-person invitations relayed by ward committees and clinic health promoters. This locally grounded recruitment strategy was chosen both for feasibility and to align with the study’s focus on situated listening practices, while acknowledging that it produces a non-probability sample with potential selection bias towards more engaged listeners (Aruleba & Jere, 2022).All participants received an information sheet in Sepedi or English, which explained the study aims, procedures, minimal risks, potential benefits, and how their data would be stored and used, including assurances of confidentiality and anonymised reporting. The information sheet also emphasised that participation was entirely voluntary, that they could decline any question, and that they could withdraw from the study at any time without penalty or consequences for access to services. I explained that joining the study was voluntary, that choosing not to participate or to withdraw later would have no adverse consequences, and that data would be stored securely with identifying details removed from transcripts and analysis files.

Data collection took place in community halls and clinic waiting areas, using small group sessions of six to eight participants. Each session began with informed consent procedures and a brief screening questionnaire to confirm Sepedi as the primary home language and regular exposure to Botlokwa FM. Participants were then randomly assigned, at the individual level, to one of three recorded bulletin conditions: (a) single mention of the SMS hotline number in the middle of the bulletin, (b) single mention at the end, or (c) two clear mentions, one mid-bulletin and one at the end. Of the 120 participants included in the analysis, 40 were assigned to the single mid condition, 40 to the single end condition, and 40 to the double (mid and end) condition, with no attrition after randomisation. Randomisation was implemented using pre-generated lists stratified by village, and facilitators allocated each arriving participant to the next position on the relevant list. Audio files for the three conditions were pre-labelled and loaded in advance so that the correct bulletin version was played for each group without misallocation.

Immediately after listening through standard speakers at typical volume, participants were asked to write down or orally recall the SMS hotline number, followed by a 10-minute filler task and a delayed recall test. Listening took place in small groups, but recall tasks were completed individually and in silence, with facilitators explicitly instructing participants not to talk and seating them so that they could not see one another’s responses. A brief structured questionnaire captured demographics, radio listening patterns, phone type, and prior use of SMS or voice health services. As a form of purposive sampling, I drew interviewees from the experimental participants in a way that reflected diversity in age, schooling, phone type, and recall outcomes. The three experimental groups were designed to be comparable on age, gender, schooling, and phone type, and subsequent checks showed no substantial differences on these characteristics.

The primary outcome was accuracy of SMS hotline recall, coded at three levels: full recall (all digits in correct order), partial recall (at least half of the digits correct in sequence), and incorrect or no recall (McCarthy et al., 2012). Oral recall responses were audio-recorded and transcribed where necessary, and written responses were collected on individual sheets. All responses were then coded into full, partial, or incorrect recall using this predefined scheme, for example, a response with all digits in the correct order was coded as full recall, one with several correct leading digits but missing or reordered final digits as partial recall, and any other response as incorrect. A second coder independently coded a subset of responses, and discrepancies were discussed to check consistency in how the categories were applied.

Secondary outcomes included self-rated clarity of the number announcement (five-point Likert scale) and perceived usefulness of the hotline. Key independent variables were bulletin condition (single mid, single end, double with end), listener age group, highest schooling level, and phone type (basic phone vs smartphone), reflecting the theoretical focus on auditory health literacy and digital inequality (Turchioe et al., 2023).

Quantitative analyses were conducted in Stata 18. Descriptive statistics summarised sample characteristics and recall performance. Chi-square tests assessed differences in full and partial recall across bulletin conditions and demographic groups. To test the propositions that end placement and repetition improve recall, and that these features interact with socio-demographic factors, multinomial logistic regression models were estimated with recall category as the dependent variable and bulletin condition, age, schooling, and phone type as predictors, controlling for self-reported radio listening frequency. Model fit and multicollinearity diagnostics were examined before interpretation.

Interview recordings were transcribed verbatim in Sepedi and translated into English by a trained bilingual research assistant. Transcripts were coded using reflexive thematic analysis with a hybrid deductive inductive approach (Braun & Clarke, 2021). Initial deductive codes captured theoretically informed constructs such as “listening strategies”, “barriers to catching numbers”, “use of others to confirm information”, and “trust in radio-advertised hotlines”, while inductive codes captured emergent practices, for example, relying on younger family members with smartphones. Coding was conducted in NVivo 14, with a second coder reviewing a subset of transcripts to enhance reflexive rigour rather than to achieve positivist reliability (O’Connor & Joffe, 2020). Thematic patterns were then examined in relation to quantitative findings to identify convergences and tensions, particularly where participants’ accounts of difficulty or reliance on repetition could help interpret stratified recall outcomes.

Thematic adequacy was assessed iteratively during analysis. As coding progressed, no substantively new codes emerged in the final interviews, suggesting that the main patterns had stabilised. Themes were also checked across age, schooling, and phone-type subgroups to ensure that they captured variation in listening conditions and constraints rather than reflecting only a narrow subset of participants. Together, these checks indicated that thematic depth was sufficient for the study’s aims, while still recognising that additional interviews might have produced further nuance. In developing the themes, I also examined deviant cases, such as older participants who showed unexpectedly strong recall and younger participants with weak recall and used these cases to refine and nuance our interpretations.

A total of 120 adult Sepedi-speaking listeners participated in the survey experiment. The mean age was 42.3 years (SD = 15.7; range 18-78), with 58.3% identifying as women. Just under half (47.5%) had completed secondary schooling, 34.2% had some secondary education, and 18.3% had primary schooling or less. In line with recent data on rural digital divides, 55.0% owned a basic mobile phone and 45.0% reported owning a smartphone (Aruleba & Jere, 2022; De Leo et al., 2024). Most participants (71.7%) reported listening to Botlokwa FM at least four days per week, and 62.5% recalled hearing health programmes “often” or “very often”. Only 27.5% reported ever using any health hotline (SMS or voice), despite near universal awareness of clinic-based services. This configuration positions the sample as a relevant but not hyper-engaged subset of the station’s core audience and underscores the importance of understanding how on-air design filters practical access to mobile health services. When participant characteristics were examined by bulletin condition, the three experimental groups were very similar in age distribution, gender, schooling, and phone type, with only minor, non-systematic differences. For example, smartphones were slightly more common in the double-mention group, but this imbalance was small and considered when interpreting the regression models. The final analytic sample comprised 40 participants in the single mid condition, 40 in the single end condition, and 40 in the double condition.

Immediate recall performance varied markedly by bulletin condition. In the single mid mention condition, only 17.5% of participants correctly recalled the full SMS hotline number immediately after listening, 31.0% achieved partial recall and 51.5% provided incorrect or no recall. In the single end condition, full recall rose to 32.5%, with 35.0% partial and 32.5% incorrect. In the double condition (mid and end mentions), 52.5% achieved full recall, 30.0% partial recall and only 17.5% incorrect. A chi-square test indicated a statistically significant association between bulletin condition and immediate recall category, χ² (4, N = 120) = 24.87, p < .001. Short delay recall after 10 minutes showed the same ordering, although full recall declined in all groups (single mid 10.0%; single end 21.0%; double 38.0%).

Multinomial logistic regression confirmed that, relative to the single mid condition and controlling for age, schooling, phone type and radio listening frequency, participants in the double condition had substantially higher odds of full versus incorrect recall (adjusted odds ratio [AOR] = 5.12, 95% CI [2.01, 13.06], p < .001), while those in the single end condition also showed improved performance (AOR = 2.34, 95% CI [1.01, 5.40], p = .048). Older age and lower schooling were independently associated with reduced odds of full recall, especially in the single mid condition, pointing to stratified auditory health literacy that mirrors documented digital health inequalities (Turchioe et al., 2023; Aruleba & Jere, 2022). Basic phone users performed worse than smartphone users in all conditions, although this difference attenuated in the double condition, suggesting that favourable bulletin design partially mitigates device-related disadvantage.

Self-rated clarity of the number announcement was highest in the double condition (mean = 4.32, SD = 0.71 on a five-point scale) compared with the single end (mean = 3.88, SD = 0.83) and single mid conditions (mean = 3.45, SD = 0.96), F(2, 117) = 8.27, p < .001. Perceived usefulness of the hotline did not differ significantly across conditions, indicating that structural changes improved recall without altering perceived relevance. Taken together, these results provide clear empirical support for the proposition that end placement and repetition function as infrastructural levers that enhance or constrain listeners’ capacity to act on mobile health invitations in auditory environments.

Quotations are reported using only non-identifying descriptors (participant code, broad age band, and phone type) and exclude village names or other direct identifiers. Interview data deepened the interpretation of these quantitative patterns by foregrounding everyday listening practices and constraints. Across age groups, participants described radio listening as a background activity embedded in household and income-generating work, with frequent interruptions and shared listening settings. Many emphasised that “numbers are the most difficult part”, explaining that “the words you understand, but when they say the number once, it is already gone”. This aligns with cognitive research showing that auditory recall of numerically dense information is highly sensitive to load and distraction (McCarthy et al., 2012).

Repetition emerged as a central resource. Participants in the double condition often reported anticipating the second mention after missing digits the first time. One listener explained, “When they repeat at the end, I prepare my mind, sometimes I even pick up a pen”. Others relied on collective strategies, such as asking younger family members with smartphones to re-listen via station streams or social media, underscoring how auditory health literacy is distributed across households rather than purely individual (Turchioe et al., 2023). Basic phone users highlighted their inability to replay bulletins or screenshot numbers, which made them more dependent on precise on-air design and on shared listening arrangements.

Trust in radio-advertised hotlines was generally high, but several participants expressed frustration that “they talk fast like they are late”, implicitly critiquing production rhythms that prioritise brevity over comprehensibility for numerically dense content. These qualitative insights suggest that bulletin scripting and pacing are experienced as part of the communicative infrastructure that either supports or undermines listeners’ agency. They therefore reinforce the conceptualisation of community radio health bulletins as “modes as filters” in which auditory information processing, socio-economic position, and media design converge to produce methodological voice bias in who can practically join SMS based health publics.