Table 1
Theme prevalence by year.
| MAIN THEME | THEME | 2022 (%) | 2023 (%) | 2024 (%) | 2025 (%) |
|---|---|---|---|---|---|
| Main Theme 1: Access, Capability & Distribution | Economic access barriers and institutional privilege | 5 (15.6%) | 13 (13.0%) | 5 (5.1%) | 1 (1.0%) |
| Main Theme 1: Access, Capability & Distribution | Skills and effective-use gap | 2 (6.3%) | 14 (14.0%) | 1 (1.0%) | 8 (7.8%) |
| Main Theme 1: Access, Capability & Distribution | Access constraints and uneven availability | 3 (9.4%) | 6 (6.0%) | 44 (44.9%) | 53 (52.0%) |
| Main Theme 1: Access, Capability & Distribution | Openness, model governance, and contested narratives | 2 (6.3%) | 8 (8.0%) | 0 (0.0%) | 0 (0.0%) |
| Main Theme 2: Political Economy, Data & Compute Power | Digital colonialism, data extraction and expropriation | 3 (9.4%) | 14 (14.0%) | 5 (5.1%) | 7 (6.9%) |
| Main Theme 2: Political Economy, Data & Compute Power | Platform, compute and market concentration | 5 (15.6%) | 4 (4.0%) | 1 (1.0%) | 1 (1.0%) |
| Main Theme 2: Political Economy, Data & Compute Power | Environmental externalities | 0 (0.0%) | 3 (3.0%) | 2 (2.0%) | 1 (1.0%) |
| Main Theme 2: Political Economy, Data & Compute Power | IP and copyright conflict | 0 (0.0%) | 0 (0.0%) | 1 (1.0%) | 1 (1.0%) |
| Main Theme 3: Societal & Educational Impacts | Labor market disruption and sectoral shifts | 0 (0.0%) | 3 (3.0%) | 14 (14.3%) | 11 (10.8%) |
| Main Theme 3: Societal & Educational Impacts | Education integrity and covert use | 0 (0.0%) | 0 (0.0%) | 1 (1.0%) | 1 (1.0%) |
| Main Theme 3: Societal & Educational Impacts | Governance, rights, privacy and societal risks | 1 (3.1%) | 0 (0.0%) | 0 (0.0%) | 0 (0.0%) |
[i] Note. Values indicate the number and percentage of posts in each year in which the relevant theme appeared. Percentages are calculated using the yearly retained corpus as the denominator: 2022, n = 32; 2023, n = 100; 2024, n = 98; 2025, n = 102. Because multiple labeling was possible, column totals do not sum to 100%
Table 2
Main Theme 1: Access, Capability & Distribution — Themes, Example Codes, and Illustrative Quotes.
| THEME | EXAMPLE CODE | ILLUSTRATIVE NORMALIZED EXCERPT (EN) | POST ID |
|---|---|---|---|
| Economic access barriers and institutional privilege | Elite institution advantage | Prompt engineering is becoming a new elite skill in the GenAI world. This creates a divide in productivity. | 88 |
| Paywall subscription barrier | ChatGPT access is “free” but the real cost is in compute and data ownership. That’s the divide. | 23 | |
| Private public school gap | Private schools: 52% GenAI usage vs public schools: 18%. New class divide in education. | 1 | |
| Skills and effective-use gap | Age divide older adults | There is a GenAI divide by age: older adults will be left behind unless we invest in support and design. | 33 |
| Skills capacity gap | ChatGPT is impressive, but let’s be clear: it doesn’t make everyone equally capable. Skills still matter. | 11 | |
| Access constraints and uneven availability | Infrastructure device gap | AI for all? Not without devices, stable internet, and electricity. The divide is infrastructure. | 22 |
| Global South general | GenAI won’t close the global digital divide. It will widen it. The Global South will be consumers, not producers. | 14 | |
| Latin America lac focus | Generative AI in Latin America — digital divide: access, language, data sovereignty. | 145 | |
| Openness, model governance, and contested narratives | Democratization claim | AI is being sold as democratizing knowledge, but access and control remain centralized. | 7 |
| Techno-solutionism critique | Technology doesn’t automatically reduce inequality. GenAI may amplify it without policy and redistribution. | 17 |
Table 3
Main Theme 2: Political Economy, Data & Compute Power — Themes, Example Codes, and Illustrative Quotes.
| THEME | EXAMPLE CODE | ILLUSTRATIVE NORMALIZED EXCERPT (EN) | POST ID |
|---|---|---|---|
| Digital colonialism, data extraction and expropriation | Data extraction expropriation | Generative AI is a new form of extraction: taking data, value, and labor from the many to benefit the few. | 9 |
| Digital colonialism | Generative AI = digital colonialism: data extraction, water use, copyright violations. | 2 | |
| Platform, compute and market concentration | Compute concentration GPU | Whoever controls compute (GPUs) controls GenAI. This is an inequality machine. | 12 |
| Platform power bigtech | GenAI benefits will accrue to Big Tech unless we rethink governance, competition, and public infrastructure. | 19 | |
| Environmental externalities | Energy carbon intensity | GenAI’s hidden costs: energy use and emissions. The divide includes who bears the environmental burden. | 28 |
| Water intensity | Water use for AI is not abstract. Communities will compete with data centers. That’s part of the AI divide. | 27 | |
| IP and copyright conflict | Copyright/IP violation | Training GenAI on copyrighted works without consent is expropriation. The benefits accrue to platforms. | 29 |
Table 4
Main Theme 3: Societal & Educational Impacts – Themes, Example Codes, and Illustrative Quotes.
| THEME | EXAMPLE CODE | ILLUSTRATIVE NORMALIZED EXCERPT (EN) | POST ID |
|---|---|---|---|
| Labor market disruption and sectoral shifts | Labor market change | GenAI will change labor markets fast. Those without access and skills will be displaced first. | 24 |
| Professional services shift | GenAI is reshaping professional services. The productivity gap will widen between adopters and non-adopters. | 26 | |
| Education integrity and covert use | Teacher unaware use | Teachers: your students are already using GenAI. The divide is between those who can use it well and those who can’t. | 31 |
| Governance, rights, privacy and societal risks | Regulation governance | Regulation matters: without governance, GenAI will widen inequalities and harm rights. | 41 |
Table 5
Frame prevalence by year (most salient frames shown, post-level presence, proportions).
| FRAME | 2022 n (%) | 2023 n (%) | 2024 n (%) | 2025 n (%) |
|---|---|---|---|---|
| Geographic divides & tool availability | 3 (9.4%) | 5 (5.0%) | 34 (34.7%) | 40 (39.2%) |
| Labor market disruption | 0 (0.0%) | 3 (3.0%) | 21 (21.4%) | 21 (20.6%) |
| Political economy/extraction | 1 (3.1%) | 15 (15.0%) | 6 (6.1%) | 7 (6.9%) |
| Skills/effective-use gap | 3 (9.4%) | 15 (15.0%) | 1 (1.0%) | 0 (0.0%) |
| Compute/platform concentration | 3 (9.4%) | 4 (4.0%) | 3 (3.1%) | 0 (0.0%) |
| IP/copyright conflict | 0 (0.0%) | 0 (0.0%) | 1 (1.0%) | 1 (1.0%) |
| Environmental externalities | 0 (0.0%) | 5 (5.0%) | 3 (3.1%) | 1 (1.0%) |
[i] Note. Values indicate the number and percentage of posts in each year in which the relevant frame appeared. Percentages are calculated using the yearly retained corpus as the denominator: 2022, n = 32; 2023, n = 100; 2024, n = 98; 2025, n = 102. Because multiple frames could be assigned to a single post, column totals do not sum to 100%.
Table 6
Supplementary year–year similarity of frame-share vectors (cosine similarity)
| YEAR | 2022 | 2023 | 2024 | 2025 |
|---|---|---|---|---|
| 2022 | 1.00 | 0.96 | 0.81 | 0.74 |
| 2023 | 0.96 | 1.00 | 0.82 | 0.75 |
| 2024 | 0.81 | 0.82 | 1.00 | 0.97 |
| 2025 | 0.74 | 0.75 | 0.97 | 1.00 |
[i] Note. Cosine similarity values are used only as supplementary descriptive indicators of within-corpus patterning and should not be interpreted as inferential statistics or as evidence of platform-wide similarity.
Table 7
Legitimation & rhetoric by year (proportions).
| DISCURSIVE MECHANISM | 2022 n (%) | 2023 n (%) | 2024 n (%) | 2025 n (%) |
|---|---|---|---|---|
| Authority/report citation | 0 (0.0%) | 0 (0.0%) | 16 (16.3%) | 13 (12.7%) |
| Stats/metrics | 0 (0.0%) | 0 (0.0%) | 1 (1.0%) | 3 (2.9%) |
| High-certainty modality | 8 (25.0%) | 11 (11.0%) | 0 (0.0%) | 2 (2.0%) |
| Intensifiers/hype | 4 (12.5%) | 5 (5.0%) | 1 (1.0%) | 0 (0.0%) |
| Binary oppositions | 0 (0.0%) | 3 (3.0%) | 0 (0.0%) | 1 (1.0%) |
[i] Note. Values indicate the number and percentage of posts in each year in which the relevant discursive mechanism appeared. Percentages are calculated using the yearly retained corpus as the denominator: 2022, n = 32; 2023, n = 100; 2024, n = 98; 2025, n = 102. Because multiple discursive mechanisms could appear in a single post, column totals do not sum to 100%.
Table 8
Top frame co-occurrences as supplementary descriptive indicators.
| FRAME A | FRAME B | CO-OCCURRENCE (POST) | EDGE JACCARD |
|---|---|---|---|
| Geographic divides & tool availability | Labor market disruption | 26 | 0.26 |
| Political economy/extraction | Environmental externalities | 5 | 0.15 |
| Geographic divides & tool availability | Political economy/extraction | 5 | 0.05 |
| Compute/platform concentration | Environmental externalities | 3 | 0.19 |
| Political economy/extraction | Compute/platform concentration | 3 | 0.08 |
[i] Note. Edge Jaccard values are reported as supplementary descriptive indicators of frame co-occurrence within the retained corpus. They are not used as inferential statistics.
Table 9
Descriptive change in frame co-occurrence: 2022–2023 versus 2024–2025 (Δ edges).
| FRAME A | FRAME B | 2022–23 | 2024–25 | Δ |
|---|---|---|---|---|
| Geographic divides & tool availability | Labor market disruption | 2 (1.5%) | 24 (12%) | +22 |
| Political economy/extraction | Environmental externalities | 1 (0.8%) | 4 (2.0%) | +3 |
| Political economy/extraction | IP/copyright conflict | 0 (0.0%) | 2 (1.0%) | +2 |
| Environmental externalities | IP/copyright conflict | 0 (0.0%) | 2 (1.0%) | +2 |
| Political economy/extraction | Compute/platform concentration | 1 (0.8%) | 2 (1.0%) | +1 |
| Compute/platform concentration | Environmental externalities | 1 (0.8%) | 2 (1.0%) | +1 |
[i] Note. Values indicate the number and percentage of posts within each period in which the two frames co-occurred. Percentages are calculated using the total number of retained posts in each period as the denominator: 2022–2023, n = 132; 2024–2025, n = 200. The Δ column reports the raw difference in co-occurrence counts and is intended as a descriptive indicator of within-corpus patterning, not as an inferential test.