International Research Journal of Business and Social Science
Journal Home
OPEN ACCESS
AI Literacy and Migrant Rights Information in Bangladesh A Survey of AI Aware Respondents
| Most. Irin Jahan Emo ORCID: https://orcid.org/0009-0007-2030-2873 Kaifa Yeshmin ORCID: https://orcid.org/0009-0008-9084-582X Fahmida Faria Iva ORCID: https://orcid.org/0009-0009-3679-9495 Department of Law Shanto-Mariam University of Creative Technology Dhaka, Bangladesh |
| Prof. Dr Kazi Abdul Mannan Department of Business Administration Faculty of Business Shanto-Mariam University of Creative Technology Dhaka, Bangladesh Email: drkaziabdulmannan@gmail.com ORCID: https://orcid.org/0000-0002-7123-132X Corresponding author: Most. Irin Jahan Emo: rahanamst82@gmail.com |
Int. Res. J. Bus. Soc. Sci. 2026, 12(4); https://doi.org/10.64907/xkmf.v12i4.irjbss.05
Submission received: 10 September 2026 / Revised: 28 September 2026 / Accepted: 30 September 2026 / Published: 02 October 2026
Download (PDF)
Abstract
Accurate and accessible information can help people navigating overseas employment, while generative artificial intelligence (AI) creates new questions about the quality and verification of migration-related advice. This study examines responses to statements about AI literacy and migrant-worker-rights information in 385 adult, AI-aware survey records from Bangladesh. The mixed sample includes law students, graduates, legal professionals and other respondents; prospective overseas workers are not identified in the workbook. The questionnaire asks about self-reported ability to locate rights information with AI, recognition of the need to verify migration information, awareness of misleading recruitment information and support for pre-departure AI-literacy training. It does not test knowledge of specific migrant-worker rights or observe use of an AI migration service. Item distributions, joint endorsement, paired comparisons and exploratory ordinal correlations are reported. Of 385 respondents, 256 (66.5%) endorsed the ability-to-locate statement, 341 (88.6%) endorsed the need to verify AI-generated migration information, and 267 (69.4%) supported AI-literacy training in pre-departure preparation. A total of 231 respondents (60.0%) endorsed both locating information and the need for verification; 110 endorsed verification without endorsing locating information. These answers do not establish actual verification or training effectiveness. The article distinguishes locating information, verifying its authority, knowing a legal right and obtaining an effective remedy. Future research should identify prospective workers, use rights-specific scenarios and document recruitment and consent procedures.
Keywords: generative artificial intelligence; migrant worker rights; AI literacy; verification; pre-departure information; Bangladesh
1. Introduction
People considering overseas employment may need reliable information on recruitment, employment contracts, migration procedures, working conditions and avenues for assistance. Information may be provided by public agencies, employers, recruiters, informal intermediaries and digital services. The reliability and relevance of that information can vary with the destination, employment arrangement and stage of migration. An accessible answer is therefore not necessarily an accurate or actionable answer.
Generative artificial intelligence (AI) can produce fluent summaries and respond to questions in natural language, but it may also produce plausible inaccuracies. UNESCO’s guidance for generative AI identifies accuracy and human oversight as issues requiring attention.[1] In migration contexts, a misleading answer about a contract, fee or complaint process could affect consequential decisions. Verification requires not only recognising that an AI output may be wrong but also knowing which authority or qualified adviser can assess the particular claim.
Bangladesh’s Ministry of Expatriates’ Welfare and Overseas Employment publishes information and provides links to services including complaint mechanisms and visa verification.[2] The International Labour Organization (ILO) describes digital overseas-employment services spanning registration, training and complaints.[3] These sources provide context and potential verification routes; the present survey does not show whether respondents used them or whether a particular route resolved a complaint.
The proposed study focuses on AI literacy and awareness of migrant-worker rights among prospective overseas workers. The workbook, however, categorises respondents as law students, law graduates, advocates or legal professionals, and other relevant respondents. It does not identify which respondents are prospective migrants or migrant workers. The title and conclusions therefore cannot attribute the full sample’s views to prospective overseas workers.
The focal MWR items also ask different things. One concerns self-reported ability to use AI to locate migrant-rights information; another concerns understanding that AI-generated migration information should be verified. Other items address awareness of misleading recruitment information, a belief about AI literacy identifying inconsistencies, and support for training before departure. None asks respondents to identify a specific right or apply the law to a scenario. The study can report these self-assessments and views, but cannot measure actual knowledge of rights or the effect of training.
Research and policy discussions often link information access with safer migration. Yet the sequence from finding information to verifying it, understanding an applicable right and obtaining an effective remedy involves several steps. The ILO’s work on fair recruitment and access to justice describes recruitment and remedies as parts of a wider migration process.[4] A survey statement that information can be located does not establish that these later steps are available or successful.
This article focuses on the relationship between self-reported ability to locate rights information and recognition that migration advice requires verification. Its distinctive contribution is a joint-response analysis that separates locating information from verification and a practical distinction between these activities and demonstrated rights knowledge. The analysis is distinct from a study of decision confidence in migration choices or trust in remittance platforms, although the same Group 11 questionnaire contains items relevant to those topics.
The objectives are to describe responses to the five migrant-rights-information statements; examine how self-reported ability to locate information coexists with recognition of the need for verification; explore associations with selected AI-literacy statements; and identify implications for future rights-specific research. The study asks: How often are the focal statements endorsed? How do location and verification responses coexist? What are the estimated associations between information-location responses and selected AI-literacy and training statements? What additional measures would be needed to examine migrant-worker rights awareness directly?
These questions concern the recorded sample and do not estimate the level of knowledge among prospective migrants in Bangladesh. They also do not test whether an AI system improves migration decisions, reduces recruitment fraud or changes access to justice.
2. Literature Review and Theoretical Foundation
This section distinguishes the content of a legal right from the processes through which a user finds, evaluates and acts on information. Theoretical perspectives are used to frame the questions rather than to claim a validated causal model.
2.1 Migration information and rights
Labour migration involves decisions and legal relationships before departure, in transit, at the destination and on return. Relevant information may concern recruitment terms, contracts, fees, working conditions and complaint processes. Bangladesh’s Overseas Employment and Migrants Act 2013, as amended, supplies part of the domestic legal context, while destination-country rules and the facts of employment may also matter.[5] A general statement about migrant rights cannot identify all applicable entitlements in every case.
The ILO has described pre-departure orientation materials covering legal issues and working conditions, while noting the importance of rights information.[6] More recent work describes digital arrangements intended to support overseas-employment processes and recruitment oversight.[7] These sources identify possible institutional resources but do not establish that respondents in the present workbook used them or that a particular service is accurate for every individual case.
The distinction between a right, information about a right and a remedy is essential. A person may find an explanation of contract terms but still need to know whether those terms apply, how to document a breach, which institution has authority and whether assistance is accessible. The questionnaire does not test these steps. Consequently, the findings should be interpreted as perceptions of information-related capabilities and precautions rather than as observed legal empowerment.
2.2 AI literacy, information search and verification
AI literacy encompasses knowledge of what AI systems can and cannot do, the ability to evaluate outputs and an understanding of responsible use. Long and Magerko’s account describes AI literacy as a set of competencies rather than a single belief.[8] The present AIL items concern self-reported awareness of plausible inaccuracies, verification, bias, sensitive information and human oversight. They do not directly test a user’s ability to identify an incorrect migration claim.
MWR1 asks whether respondents can use AI to locate migrant-rights information. It does not ask them to locate a named statute or assess an actual output. MWR2 asks whether they understand the need to verify AI-generated migration information with authoritative sources. A person may recognise the need to verify while lacking access to an appropriate source; conversely, agreement with a self-assessment is not proof that verification occurred. These items are therefore analysed separately.
A concrete verification task would identify the claim, applicable jurisdiction and date. For example, a question about a recruitment fee or a contract term may need to be checked against current rules, a specific contract and reliable advice. AI-generated text may be a starting point for further inquiry but cannot be assumed accurate, current or applicable merely because it is fluent. The survey records views about verification and does not establish the correctness of any generated answer.
2.3 Information adoption theory
The information adoption model developed by Sussman and Siegal examines how the perceived quality and credibility of information can influence its adoption in mediated organisational contexts.[9] It provides a conceptual starting point for asking how a person evaluates AI-generated migration information. The original model was developed in a different setting; applying it to prospective migrants would require measuring relevant information quality, source credibility, perceived usefulness and adoption in that population.
The current workbook does not present an identified message, its source, an adoption decision or a measure of argument quality. It cannot test the full model or show that a respondent adopted migration advice. The model is used as an interpretive lens for future research, not as a verified mechanism in this study. A later study could compare how participants assess an AI answer with and without verifiable sources or a referral to a qualified human adviser.
2.4 Trust and institutional credibility
Trust involves a willingness to rely on another party under conditions of uncertainty. Mayer, Davis and Schoorman’s model distinguishes characteristics relevant to trust from trust itself and the subsequent act of risk-taking.[10] In migration services, a person may assess the competence and integrity of a provider, the accuracy of information and the availability of recourse. Trust in an AI tool, in a government agency and in a recruiter are not automatically the same.
The Group 11 instrument includes trust statements about remittance and legal-information platforms, but this article does not treat them as a direct measure of trust in a named institution. Nor does it claim that trust mediates the relationship between AI literacy and rights awareness. The focal statements do not identify a platform or an information provider, and no temporal or experimental design establishes such a pathway.
Trustworthy information requires more than persuasive presentation. A service can state that it is reliable while leaving its sources, update date or complaint process unclear. Conversely, an official source may be difficult for a user to access or interpret. A useful evaluation would consider accessibility and the ability to check or challenge information alongside the credentials of the source.
2.5 Human capital and training
Human capital theory considers education and training as investments in knowledge and skills.[11] It can help frame questions about whether AI-literacy training equips people to evaluate migration information. But training availability, quality, language and time costs matter; a belief that training should be included before departure does not establish that a programme was delivered or effective.
MWR5 asks whether AI-literacy training should be included in pre-departure preparation. The statement is a preference about a possible programme, not a measure of a respondent’s actual training or a pre/post assessment. The ILO’s descriptions of pre-departure orientation suggest that training can cover multiple migration and rights topics, but those sources do not evaluate the programme proposed by this questionnaire.[12]
A future training study should specify its content and intended audience, measure comprehension before and after instruction, and compare outcomes where feasible. It should also examine whether participants can use relevant official sources and contact qualified help in the languages and conditions they encounter. Training should not be treated as a substitute for enforceable rights, fair recruitment or accessible remedies.
A migration-related message can be evaluated along several dimensions. Its factual accuracy matters, but so do its date, geographic scope, relevance to an individual’s employment arrangement and the identity of the organisation responsible for it. A statement that was correct for one destination or recruitment channel may be misleading when applied to another. A system that produces a clear summary without making those boundaries visible can create confidence that is not supported by the underlying source.
These dimensions are particularly important where a person must make decisions before having access to all relevant documents. A recruitment offer may contain claims about pay, accommodation, fees or the duties of an employer. Checking a claim may require comparing the written contract, current official guidance and information from a person with relevant expertise. The questionnaire does not collect contracts, destination details or examples of disputed claims; these steps are proposed as a framework for future evaluation rather than as events observed in the dataset.
A person’s capacity to verify information depends partly on the institutions and resources available to them. An authoritative webpage may be difficult to find, written in an unfamiliar language or inaccessible on a low-bandwidth device. A human support service may have limited operating hours or require information that the user does not possess. These possibilities show why an individual literacy score alone would not establish effective access to rights information. They are possible conditions for further study; the present workbook does not measure them.
Trust can assist people in navigating complex information, but an unconditional recommendation to trust a platform would not address whether it is competent, accountable or able to correct errors. The information adoption and organisational trust literatures offer useful conceptual distinctions between source credibility, perceived information usefulness and willingness to rely.[13] A migration-specific application would need to consider power differences between the worker, recruiter and provider as well as the availability of alternatives.
A future assessment can use short scenarios that specify a destination, recruitment channel and employment arrangement. Participants might receive a realistic AI-generated answer and be asked to identify its source, distinguish an unsupported claim from a verifiable one, and find an appropriate official or qualified human source. The task should be developed with legal and migration specialists so that the expected answer reflects current law and acknowledges genuine uncertainty. A score can then distinguish locating a source from interpreting it and choosing a next step.
To evaluate a training programme, researchers could measure these abilities before and after a defined intervention and compare participants with an appropriate control group where feasible. They should also record language, accessibility and prior experience, avoid collecting unnecessary identifying or immigration information, and ensure that participants know how to obtain independent assistance. Improvements in a simulated task should not automatically be treated as proof of safer migration or successful remedies. Those outcomes would require longer-term evidence and careful attention to participant protection.
2.6 Theoretical integration and analytical expectations
Information adoption, trust and human capital each address a different part of the problem: how a message is evaluated, whether a source is trusted and how skills may be developed. These perspectives are complementary, but the survey does not measure a complete model connecting them. It does not establish that training changes knowledge, that trust mediates an association or that an AI output was adopted.
The analysis concentrates on MWR1, MWR2 and related items without combining them into a validated rights-awareness scale. Three exploratory paired comparisons contrast MWR1 with MWR2, MWR3 and MWR5. Four exploratory Spearman correlations compare MWR1 with AIL1, AIL2, MWR4 and MWR5. These were selected for the current analysis and are not presented as preregistered confirmatory tests. Differences in item wording mean that paired contrasts do not rank needs or estimate the impact of an intervention.
3. Methodology
This section describes the supplied data, measurement choices, analysis and ethical information. It distinguishes the proposed population from the respondent categories actually recorded.
3.1 Research design and population
The study uses a cross-sectional questionnaire dataset containing 385 records. The workbook codes all analysed records as consenting, adult eligible and aware of AI. Recorded status categories are law student, law graduate, advocate or legal professional, and other relevant respondent. The other category is not further defined. No field identifies prospective overseas workers, current migrant workers or family members receiving remittances.
The unit of analysis is the individual recorded response. The survey can describe views in this mixed AI-aware sample and explore associations between items. It cannot show whether AI literacy causes rights awareness, whether training improves behaviour or whether a respondent actually used a migration information service. The analysis does not make a national population claim.
3.2 Sampling and provenance
The workbook contains 385 unique respondent identifiers. The supplied materials do not establish recruitment dates, a sampling frame, invitations, survey administration mode, response rate or independent duplicate checks. A sample-size target of 385 can arise from a conventional calculation under specified assumptions, but the number does not establish probability selection or a national margin of error.[14] No sampling weights are applied.
The researcher identifies the numerical answers as real survey responses. The Codebook labels the respondent identifier sequence and optional contextual comments as synthetic. These labels alone do not establish that numerical answers are synthetic, but their creation and linkage to source questionnaires need documentation. The optional contextual text is excluded from analysis. Fieldwork, consent and source-record documentation are needed to permit independent verification.
3.3 Questionnaire and measurement
The instrument uses five ordered response categories from strongly disagree to strongly agree. MWR1 concerns ability to use AI to locate migrant-worker-rights information; MWR2 concerns understanding the need to verify AI-generated migration information; MWR3 concerns awareness of misleading recruitment information; MWR4 concerns the potential role of AI literacy in identifying suspicious or inconsistent information; and MWR5 concerns inclusion of AI-literacy training in pre-departure preparation. Appendix A reproduces the exact statements.
AIL1 concerns understanding that AI can generate plausible but inaccurate information. AIL2 asks about verifying important AI-generated information before relying on it. These are self-reported statements, not tests of objective ability or observed behaviour. All focal numerical items have valid responses in the 385 supplied records. No separate pilot, validated item adaptation, cognitive testing or measurement-model validation is documented in the supplied materials.
The MWR statements address capability, verification, recruitment risk, a belief about literacy and a training preference. They are not averaged into a validated rights-awareness score. A common response scale does not establish a single construct; internal consistency alone cannot establish validity.[15] The survey does not identify which legal rights respondents know or whether a correct answer is applied to a concrete migration situation.
3.4 Descriptive and paired analysis
Each focal item is summarised with counts for all five response categories and the number and percentage selecting agree or strongly agree. Neutral and disagreeing answers remain separate in the distributions. They are combined only as non-endorsement in joint and paired summaries; neutral answers are not treated as opposition.
Three within-person contrasts compare endorsement of MWR1 with MWR2, MWR3 and MWR5. Exact McNemar tests use discordant pairs and Holm adjustment covers the three tests. Pointwise percentile bootstrap intervals for percentage-point differences use 10,000 respondent-level resamples.[16] Because the statements have different wording, these contrasts do not establish that one is more important or that training or verification changed any outcome.
3.5 Exploratory associations
Four Spearman correlations use the original five-category responses to compare MWR1 with AIL1, AIL2, MWR4 and MWR5. Holm adjustment covers the four p-values; percentile bootstrap intervals based on 2,000 respondent-level resamples are pointwise and not adjusted for multiplicity. The comparisons are exploratory and were selected for this analysis, without a claim of preregistration. Correlations do not establish temporal order or causal direction, and statistical non-significance does not prove the absence of a relationship.[17]
3.6 Data handling and uncertainty
The analysis retains all 385 records for focal numerical items, does not use the optional comments, and reports aggregate results. No imputation or weighting is applied. Pointwise Wilson intervals describe uncertainty under conventional assumptions, but undocumented recruitment means they cannot be interpreted as representative national intervals.
Responses can reflect differences in interpretation, familiarity and context. A neutral answer may reflect uncertainty, indifference or a reason not captured by the questionnaire. The paired binary summaries are therefore supplemented by full item distributions and exact wording. The analysis does not classify respondents as migrants based on their location, age or answer to a migration-information item.
3.7 Ethical considerations
The questionnaire materials state that participation is voluntary, no name is required, responses will be analysed in aggregate and participants may stop at any time. The workbook records affirmative consent and adult eligibility for analysed records. These fields do not independently verify the consent process, data-collection conditions or the existence or absence of an applicable ethics review. Formal ethics-review status and oversight must be confirmed by the authors before submission. The contextual-text field was excluded.
4. Results
The results describe responses to statements in the supplied mixed sample. They do not measure the prevalence of rights knowledge among prospective overseas workers or the effectiveness of a migration service.
4.1 Respondent profile
Table 1 presents the recorded age and status categories. The categories do not identify prospective migrants or establish representative sampling.
Table 1. Recorded age and respondent status
| Measure | Category | Count (%) |
| Age | 21-23 | 154 (40.0%) |
| 24-26 | 75 (19.5%) | |
| 18-20 | 62 (16.1%) | |
| 27-30 | 48 (12.5%) | |
| 31+ | 46 (11.9%) | |
| Status | Other relevant respondent | 152 (39.5%) |
| Law student | 138 (35.8%) | |
| Law graduate | 55 (14.3%) | |
| Advocate or legal professional | 40 (10.4%) |
N = 385. Prospective migrant status is not identified; status does not establish a representative national sample.
Of 385 records, 152 (39.5%) are other relevant respondents, 138 (35.8%) are law students, 55 (14.3%) are law graduates and 40 (10.4%) are advocates or legal professionals. The largest age category is 21–23, with 154 records (40.0%). Migration intention, employment destination and use of migration services are not recorded.
4.2 Rights-information statements
Table 2 reports the full distributions of the five MWR statements. They concern different aspects of locating information, verification, recruitment risk and training preference.
Table 2. Migrant-rights information statements
| Item | Counts 1 / 2 / 3 / 4 / 5 | Endorsed n (%) | 95% CI (%) |
| MWR1 | 0 / 12 / 117 / 179 / 77 | 256 (66.5%) | 61.6–71.0 |
| MWR2 | 0 / 2 / 42 / 185 / 156 | 341 (88.6%) | 85.0–91.4 |
| MWR3 | 0 / 9 / 106 / 183 / 87 | 270 (70.1%) | 65.4–74.5 |
| MWR4 | 0 / 15 / 106 / 192 / 72 | 264 (68.6%) | 63.8–73.0 |
| MWR5 | 0 / 12 / 106 / 175 / 92 | 267 (69.4%) | 64.6–73.7 |
N = 385. Endorsement means agree or strongly agree; neutral answers are not opposition. Wilson intervals are pointwise; exact wording is in Appendix A.
The self-reported ability to locate migrant-rights information with AI was endorsed by 256 respondents (66.5%). The need to verify AI-generated migration information was endorsed by 341 (88.6%), and 267 (69.4%) supported including AI-literacy training in pre-departure preparation. These are responses to statements, not a test of legal knowledge or training effectiveness.
4.3 Joint responses on locating and verification
Table 3 shows how the ability-to-locate and verification-need statements coexist within respondents. Each person appears in one category.
Table 3. Joint endorsement of locating and verification
| Response configuration | Count (%) |
| Both endorsed | 231 (60.0%) |
| Locating only | 25 (6.5%) |
| Verification only | 110 (28.6%) |
| Neither endorsed | 19 (4.9%) |
N = 385. Neither includes neutral answers; categories describe responses, not fixed types of people.
A total of 231 respondents (60.0%) endorsed both statements. A further 110 (28.6%) endorsed the need to verify without endorsing the ability-to-locate statement. The latter category includes neutral and disagreement responses on MWR1; it does not establish why those respondents answered as they did or whether they actually verified an AI output.
4.4 General AI-literacy statements
Table 4 provides context for the exploratory associations. AIL1 and AIL2 are self-reported statements about AI inaccuracies and verification, rather than tests of migrant-rights knowledge.
Table 4. General AI-literacy statements
| Item | Counts 1 / 2 / 3 / 4 / 5 | Endorsed n (%) | 95% CI (%) |
| AIL1 | 2 / 21 / 114 / 175 / 73 | 248 (64.4%) | 59.5–69.0 |
| AIL2 | 0 / 2 / 54 / 177 / 152 | 329 (85.5%) | 81.6–88.6 |
N = 385. Endorsement means agree or strongly agree; neutral answers are not opposition. Wilson intervals are pointwise; exact wording is in Appendix A.
The distinction between a general AI statement and a migration-specific statement matters: knowing that AI may produce inaccurate information does not establish that a respondent can determine the legal authority for a particular migration claim. The survey does not provide a scenario in which this ability can be assessed.
4.5 Paired comparisons
Table 5 compares endorsement of MWR1 with three other statements within the same respondents. The p-values are Holm-adjusted for these comparisons and the confidence intervals are pointwise.
Table 5. Paired endorsement comparisons
| Contrast | First only / second only | Difference pp | 95% CI pp | Holm p |
| MWR1 − MWR2 | 25 / 110 | -22.1 | -27.5 to -16.6 | <.001 |
| MWR1 − MWR3 | 80 / 94 | -3.6 | -10.4 to 3.1 | 0.649 |
| MWR1 − MWR5 | 75 / 86 | -2.9 | -9.1 to 3.6 | 0.649 |
N = 385; pp = percentage points. Exact McNemar tests, Holm adjusted across three tests. Bootstrap intervals are pointwise. Items have different wording.
Because the statements ask different questions, their endorsement differences describe response patterns. They do not demonstrate a training effect, the importance of one need relative to another or the reliability of a particular information service.
4.6 Exploratory associations
Table 6 uses the five-category responses to examine associations with MWR1. The p-values are adjusted across four tests and the bootstrap intervals are pointwise.
Table 6. Exploratory associations with information-location ability
| Items | Spearman ρ | 95% CI | p | Holm p |
| MWR1 / AIL1 | 0.080 | -0.016 to 0.180 | 0.119 | 0.356 |
| MWR1 / AIL2 | -0.016 | -0.116 to 0.085 | 0.760 | 0.760 |
| MWR1 / MWR4 | 0.094 | -0.002 to 0.196 | 0.065 | 0.260 |
| MWR1 / MWR5 | 0.077 | -0.019 to 0.167 | 0.132 | 0.356 |
N = 385. Holm adjustment covers four tests; 95% percentile bootstrap intervals are pointwise, based on 2,000 resamples.
All four estimated associations were small (Spearman ρ from −0.016 to 0.094); none met the 0.05 threshold after Holm adjustment. The confidence intervals are pointwise and are not adjusted for multiplicity. The associations describe simultaneous responses in the recorded sample; they cannot show whether general AI literacy causes a person to find accurate rights information or whether training changes verification practices. The estimates and their uncertainty should be interpreted together.
5. Discussion
The main finding is the coexistence of self-reported information-location ability and recognition of the need to verify AI-generated migration information. The data do not establish whether respondents found accurate legal information, verified it or used it to exercise a right.
5.1 Locating information and recognising verification
In the full sample, 66.5% endorsed the ability-to-locate statement, while 88.6% endorsed the need to verify migration information. A total of 231 endorsed both. This pattern is compatible with respondents seeing AI as a potential search aid while recognising limitations in its outputs. The questionnaire does not ask whether the same people located an actual right or checked an answer against an official source.
The 110 respondents who endorsed verification without endorsing ability to locate information are a descriptive response group, not proof of a skills deficit. Their MWR1 answers could be neutral or disagreement, and the survey does not ask why. A future assessment could present a defined migration question and ask participants to locate, identify and explain an authoritative source. It could then distinguish access barriers from difficulty evaluating the information.
The information adoption model draws attention to the quality and credibility of a message.[18] It does not imply that a general agreement with verification necessarily produces correct adoption. In this study the source, message and subsequent action are not observed. Their roles remain questions for future scenario-based or observational research.
5.2 From general AI awareness to rights-specific understanding
The general AIL items concern awareness of AI inaccuracies and self-reported verification, while MWR1 concerns perceived ability to find migrant-rights information. Their association, if present, does not establish that a person understands a particular right. A legally accurate answer depends on applicable law, destination, individual circumstances and the current status of the information.
For example, knowing that a contract may contain employment terms is different from determining whether a particular term is lawful or how to seek help after a discrepancy. The Overseas Employment and Migrants Act 2013 provides part of the Bangladeshi context, but the relevant destination law and facts may also be important.[19] This study does not provide legal determinations for individual cases.
The exploratory correlations are cross-sectional and cannot determine whether AI literacy precedes information-location ability or whether a third factor influences both. Low or uncertain estimates also do not prove that the concepts are unrelated. A better design would use independently validated measures and observe how participants handle a specific information task.
5.3 Recruitment information and practical safeguards
Misleading recruitment information can create legal and financial risks, but the survey’s MWR3 item does not identify a particular misleading claim or an actual recruitment experience. Agreement indicates awareness of a general possibility rather than proof that a respondent encountered fraud. ILO work on fair recruitment and access to justice addresses a broader set of institutional practices and remedies.[20]
The Ministry’s public website lists information and complaint services, and the ILO describes digital overseas-employment processes.[21] These are potential sources to consult and verify at the time of use; the present data do not show whether respondents accessed them or whether a complaint was resolved. A responsible AI information service should distinguish a general answer from legal advice and direct a user to current, relevant sources and appropriate human assistance when needed.
5.4 Pre-departure training and access
The training item was endorsed by 69.4% of respondents. This shows support for including AI literacy in pre-departure preparation as a general proposition. It does not establish programme content, cost, participation or effectiveness. Human capital theory suggests that learning can develop skills, but whether a particular training programme improves rights understanding is an empirical question.[22]
A practical programme could teach participants to compare an AI-generated statement with an official source, identify the date and jurisdiction of that source, protect sensitive information and seek qualified assistance for unresolved questions. Such activities should be evaluated using realistic tasks rather than agreement statements alone. Access considerations include language, connectivity, disability access and the time available before departure. These are potential design considerations, not barriers measured in the present sample.
5.5 Institutional trust and remedies
Information users may need to decide whether to rely on a recruiter, a public agency, a digital platform or an AI-generated answer. Trust in each actor can have different bases and consequences.[23] The present analysis does not measure trust in a named service, evaluate a complaint pathway or show that rights are enforceable in a particular destination.
Access to justice involves more than finding a link: a worker may need understandable procedures, evidence, representation, and a response from an institution with relevant authority.[24] A digital referral may help someone begin that process, but a referral is not itself a successful remedy. Future evaluation should track whether a user can reach the service, obtain an appropriate response and understand the next step.
A useful migration-information service should make clear what it does not know. If the destination, contract type or recruitment arrangement is unspecified, an answer should not imply that one rule applies to every case. The date of the source matters because procedures and legal conditions can change. Where a source is unavailable or contradictory, an explicit statement of uncertainty can support a user’s decision to seek further help. The current survey does not test whether any platform communicates uncertainty in this way.
The quality of a referral also deserves attention. A link to a government page may be a helpful starting point, but some users will require an accessible explanation, an opportunity to ask follow-up questions or assistance with a complaint. A referral can be judged by whether it reaches an appropriate service and whether the person understands the next step. Such outcomes are different from the number of people who agree that verification is necessary. They require user-centred evaluation with relevant safeguards for privacy and voluntary participation.
AI-generated text may combine information from different jurisdictions or fail to identify its source. Checking a claim involves comparing it with an authoritative, current source and, when appropriate, a contract or individual record. A person should also be able to recognise when the evidence is insufficient to answer the question. This process does not guarantee that all errors can be detected, particularly where rules are complex or services are difficult to access. Human advice and institutional accountability remain important alongside digital skills.
Future studies can compare participants’ interpretations of the same migration scenario when they receive an AI answer, an official source, or both. Outcomes could include the ability to identify uncertainty, locate a reliable source and choose an appropriate next step. Any comparison should explain the setting and avoid presenting a simulated recruitment scenario as evidence of an actual violation. The results could inform training materials while preserving the distinction between a research task and legal advice.
5.6 Integrated interpretation
The findings support a distinction between four steps: locating an answer, verifying its authority, understanding a specific right and obtaining help or a remedy. The questionnaire addresses self-reported aspects of the first two and views about training. It does not observe the latter steps. Information adoption, trust and human capital provide useful questions for further work but are not validated as a causal chain here.
The sample limitation is consequential. Most records are law students, law graduates, legal professionals or other relevant respondents; prospective migrant status is not identified. Their views should not be generalised to prospective overseas workers. A subsequent study should recruit and identify that population explicitly, while recognising the diversity of migration destinations, languages and recruitment pathways.
6. Theoretical and Practical Implications
The implications distinguish the observed responses from recommendations for research and service design. No AI intervention or migration service was evaluated.
6.1 Theoretical implications
First, self-reported ability to locate information, recognition of the need to verify, legal knowledge and access to remedies should be measured separately. A single mean across heterogeneous items can obscure their different meanings.[25] A rights-awareness instrument should identify the jurisdiction and test application of a particular rule rather than treating confidence in searching as legal knowledge.
Second, an information adoption study should identify the message, its source and the adoption decision.[26] Trust should identify the actor trusted and the risk taken, rather than being inferred from agreement that information ought to be verified. Third, human-capital arguments about training require direct evidence of the content and outcomes of training.[27] The present data provide an exploratory basis for these questions rather than confirmation of a combined model.
6.2 Practical implications
Pre-departure education could include exercises on checking AI-generated migration information against current official sources, recognising uncertain answers and seeking qualified human help. Programme evaluation should measure whether participants can complete those tasks correctly and retain the skills. Training should be accessible in relevant languages and formats and should not require users to expose unnecessary sensitive information.
AI-enabled information services could provide the source, jurisdiction and date of an answer, identify uncertainty, distinguish general information from professional advice and offer a route to human assistance. These are design proposals; the survey does not evaluate a deployed system or demonstrate that such features cause better outcomes. The Ministry and ILO sources can provide starting points for verified referrals, subject to checking current service details.[28]
6.3 Policy and societal implications
Policymakers can consider AI literacy as one element of broader fair-recruitment and rights-protection measures. Effective protection also depends on recruitment oversight, accessible complaints and remedies.[29] A broad survey endorsement of training does not establish the resources needed or the best programme design.
Consultation with prospective migrants, returning workers, worker organisations, recruiters and relevant public agencies could identify practical information needs. Their experiences should be measured directly rather than inferred from a mixed law-oriented sample. Testing the accessibility of information and referral channels is as important as publishing guidance.
7. Conclusion, Limitations and Future Research
The analysis describes 385 AI-aware records on migrant-rights information. Of these, 66.5% endorsed self-reported ability to locate rights information with AI and 88.6% endorsed the need to verify AI-generated migration information; 231 respondents endorsed both. These results describe the recorded sample and do not demonstrate objective legal knowledge, actual verification or the effectiveness of training.
The principal contribution is to separate locating information and recognising verification from knowing a specific right and obtaining a remedy. This distinction is necessary when interpreting a questionnaire whose items ask about different capabilities and preferences. The findings do not establish a causal pathway from AI literacy to migrant-worker rights awareness.
Limitations include a mixed sample that does not identify prospective overseas workers, undocumented recruitment and dates, cross-sectional self-report, no rights-specific scenario and no observed use of an AI migration service. The researcher identifies the numerical responses as real survey data. The Codebook’s synthetic labels for identifiers and optional comments need documentation against source records; contextual text was excluded. The sample does not support a national estimate.
Future research should document recruitment, consent and data provenance, recruit prospective workers explicitly, and record relevant destinations and information needs. Scenario-based tasks can assess whether participants identify authoritative sources, understand a specific right and know how to seek assistance. Longitudinal or comparative training studies could test whether instruction improves those abilities and whether accessible referral pathways lead to useful support.
Declarations
Ethical approval and oversight
The supplied materials do not establish whether formal ethics approval was obtained or an exemption applied. The authors must confirm the applicable review status and institutional context before submission. The questionnaire states that participation is voluntary, that no name is required and that respondents may stop at any time; the workbook codes all analysed records as consenting adults. The methodological information is described in Section 3.7.
Informed consent
The questionnaire includes a consent statement and the analysed records contain affirmative consent and adult-eligibility responses. The authors should confirm how consent was obtained and whether participants received the stated information in the field.
Data availability
The dataset was supplied for this analysis. The corresponding author should confirm the data custodian, access conditions and any restrictions on sharing de-identified records before submission. The contextual-text field was excluded from the analysis.
Funding and competing interests
Funding and competing-interest information must be confirmed by both authors before submission.
Author contributions
Most. Irin Jahan Emo and Prof. Dr Kazi Abdul Mannan should confirm their respective contributions to conceptualisation, data collection and curation, analysis, manuscript preparation, supervision and final approval. Prof. Dr Kazi Abdul Mannan is identified as co-author and supervisor.
Generative AI disclosure
Generative AI assisted in drafting, language editing, organisation and development of analytical code. Statistical estimates were calculated programmatically from the supplied numerical workbook. The authors should review and approve the analysis, interpretation, references and disclosure in accordance with the target journal’s policy.
Bibliography
Altman DG and Bland JM, ‘Absence of Evidence Is Not Evidence of Absence’ (1995) 311 BMJ 485
Becker GS, Human Capital: A Theoretical and Empirical Analysis, with Special Reference to Education (National Bureau of Economic Research 1964)
Cochran WG, Sampling Techniques (3rd edn, John Wiley & Sons 1977)
Efron B and Tibshirani RJ, An Introduction to the Bootstrap (Chapman & Hall 1993)
Holm S, ‘A Simple Sequentially Rejective Multiple Test Procedure’ (1979) 6 Scandinavian Journal of Statistics 65
International Labour Organization, From Registration to Return: Strengthening Overseas Employment Governance in Bangladesh with ILO Support (17 December 2025) <https://www.ilo.org/resource/article/registration-return-strengthening-overseas-employment-governance-bangladesh> accessed 27 October 2026
International Labour Organization, Fair Recruitment and Access to Justice for Migrant Workers (ILO 2022) <https://www.ilo.org/publications/fair-recruitment-and-access-justice-migrant-workers> accessed 27 October 2026
International Labour Organization, Promising Practices for Fair Recruitment in Bangladesh: Digitised Management of Recruitment Agencies (ILO 2024) <https://www.ilo.org/publications/promising-practices-fair-recruitment-bangladesh-digitised-management> accessed 27 October 2026
International Labour Organization, Development of Pre-Departure Orientation Training Manuals in Bangladesh <https://wwwex.ilo.org/dyn/migpractice/migmain.showPractice?p_lang=en&p_practice_id=105> accessed 27 October 2026
Long D and Magerko B, ‘What Is AI Literacy? Competencies and Design Considerations’ in Proceedings of the 2020 CHI Conference on Human Factors in Computing Systems (ACM 2020)
Mayer RC, Davis JH and Schoorman FD, ‘An Integrative Model of Organizational Trust’ (1995) 20 Academy of Management Review 709
McNemar Q, ‘Note on the Sampling Error of the Difference between Correlated Proportions or Percentages’ (1947) 12 Psychometrika 153
Miao F and Holmes W, Guidance for Generative AI in Education and Research (UNESCO 2023)
Ministry of Expatriates’ Welfare and Overseas Employment, Citizen E-Services <https://probashi.gov.bd/> accessed 27 October 2026
Sijtsma K, ‘On the Use, the Misuse, and the Very Limited Usefulness of Cronbach’s Alpha’ (2009) 74 Psychometrika 107
Sussman SW and Siegal WS, ‘Informational Influence in Organizations: An Integrated Approach to Knowledge Adoption’ (2003) 14 Information Systems Research 47
Wasserstein RL and Lazar NA, ‘The ASA Statement on p-Values: Context, Process, and Purpose’ (2016) 70 American Statistician 129
Appendix A. Analytical questionnaire items
Table A1 reproduces the survey statements so that each estimate can be interpreted against its exact wording.
Table A1. Exact wording of analytical items
| Code | Questionnaire statement |
| MWR1 | I can use AI tools to locate information about migrant-worker rights. |
| MWR2 | I understand the need to verify AI-generated migration information with authoritative sources. |
| MWR3 | I am aware that misleading recruitment information can expose migrants to legal and financial risks. |
| MWR4 | AI literacy can help prospective migrants identify suspicious or inconsistent information. |
| MWR5 | I believe AI-literacy training should be included in pre-departure preparation. |
| AIL1 | I understand that generative AI can produce plausible but inaccurate information. |
| AIL2 | I verify important AI-generated information before relying on it. |
Source: Group 11 Codebook. Response options range from strongly disagree to strongly agree.
Items are analysed separately and do not establish a single validated latent construct.
Table of legislation
Overseas Employment and Migrants Act 2013 (Bangladesh Act No 48 of 2013, as amended)
[1] Fengchun Miao and Wayne Holmes, Guidance for Generative AI in Education and Research (UNESCO 2023).
[2] Ministry of Expatriates’ Welfare and Overseas Employment, Citizen E-Services <https://probashi.gov.bd/> accessed 27 October 2026.
[3] International Labour Organization, From Registration to Return: Strengthening Overseas Employment Governance in Bangladesh with ILO Support (17 December 2025) <https://www.ilo.org/resource/article/registration-return-strengthening-overseas-employment-governance-bangladesh> accessed 27 October 2026.
[4] International Labour Organization, Fair Recruitment and Access to Justice for Migrant Workers (ILO 2022) <https://www.ilo.org/publications/fair-recruitment-and-access-justice-migrant-workers> accessed 27 October 2026.
[5] Overseas Employment and Migrants Act 2013 (Bangladesh Act No 48 of 2013, as amended).
[6] International Labour Organization, Development of Pre-Departure Orientation Training Manuals in Bangladesh <https://wwwex.ilo.org/dyn/migpractice/migmain.showPractice?p_lang=en&p_practice_id=105> accessed 27 October 2026.
[7] ILO, Registration to Return (n 3); International Labour Organization, Promising Practices for Fair Recruitment in Bangladesh: Digitised Management of Recruitment Agencies (ILO 2024) <https://www.ilo.org/publications/promising-practices-fair-recruitment-bangladesh-digitised-management> accessed 27 October 2026.
[8] Duri Long and Brian Magerko, ‘What Is AI Literacy? Competencies and Design Considerations’ in Proceedings of the 2020 CHI Conference on Human Factors in Computing Systems (ACM 2020).
[9] Stephanie Watts Sussman and Wendy Schneier Siegal, ‘Informational Influence in Organizations: An Integrated Approach to Knowledge Adoption’ (2003) 14 Information Systems Research 47.
[10] Roger C Mayer, James H Davis and F David Schoorman, ‘An Integrative Model of Organizational Trust’ (1995) 20 Academy of Management Review 709.
[11] Gary S Becker, Human Capital: A Theoretical and Empirical Analysis, with Special Reference to Education (National Bureau of Economic Research 1964).
[12] ILO, Pre-Departure Orientation (n 6).
[13] Sussman and Siegal (n 9); Mayer, Davis and Schoorman (n 10).
[14] William G Cochran, Sampling Techniques (3rd edn, John Wiley & Sons 1977).
[15] Klaas Sijtsma, ‘On the Use, the Misuse, and the Very Limited Usefulness of Cronbach’s Alpha’ (2009) 74 Psychometrika 107.
[16] Quinn McNemar, ‘Note on the Sampling Error of the Difference between Correlated Proportions or Percentages’ (1947) 12 Psychometrika 153; Sture Holm, ‘A Simple Sequentially Rejective Multiple Test Procedure’ (1979) 6 Scandinavian Journal of Statistics 65; Bradley Efron and Robert J Tibshirani, An Introduction to the Bootstrap (Chapman & Hall 1993).
[17] Douglas G Altman and J Martin Bland, ‘Absence of Evidence Is Not Evidence of Absence’ (1995) 311 BMJ 485; Ronald L Wasserstein and Nicole A Lazar, ‘The ASA Statement on p-Values: Context, Process, and Purpose’ (2016) 70 American Statistician 129.
[18] Sussman and Siegal (n 9).
[19] Overseas Employment and Migrants Act 2013 (n 5).
[20] ILO, Fair Recruitment (n 4).
[21] Ministry of Expatriates’ Welfare (n 2); ILO, Registration to Return (n 3).
[22] Becker (n 11).
[23] Mayer, Davis and Schoorman (n 10).
[24] ILO, Fair Recruitment (n 4).
[25] Sijtsma (n 15).
[26] Sussman and Siegal (n 9).
[27] Becker (n 11).
[28] Ministry of Expatriates’ Welfare (n 2); ILO, Registration to Return (n 3).
[29] ILO, Fair Recruitment (n 4); ILO, Recruiting Agencies (n 7).