Submitted:
06 November 2026
Posted:
09 November 2026
You are already at the latest version
Abstract
AI-assisted legal information may help people identify services and obtain preliminary guidance, but its accessibility and safety cannot be inferred from availability alone. This study examines perceptions of legal-service accessibility, human support and professional oversight using 385 adult, AI-aware survey records from Bangladesh. The sample comprises law students, graduates, advocates or legal professionals and other respondents; the workbook does not identify actual justice seekers or legal-aid users. The analysis retains individual questionnaire items and uses paired endorsement comparisons and exploratory rank correlations instead of treating heterogeneous items as validated scales. Of 385 respondents, 276 (71.7%) agreed that AI-assisted legal aid could help people identify where to seek assistance, 249 (64.7%) endorsed possible access for people far from providers, 253 (65.7%) endorsed retaining human support when AI cannot address a user’s needs, and 330 (85.7%) supported responsible AI use to complement rather than replace legal-aid professionals. The last statement exceeded endorsement of human support by 20.0 percentage points. The statements differ in content, so that difference is not a preference ranking or evidence of the relative effectiveness of safeguards. Associations of general accessibility and conditional trust in professional oversight with two accessibility items were small and uncertain after adjustment for multiple comparisons. These self-reported, cross-sectional findings do not establish actual access, service quality, improved legal outcomes or causality. The article integrates technology-acceptance and trust perspectives with people-centred access to justice and proposes future evaluation with identified legal-aid users and observed referral outcomes.
Keywords:
artificial intelligence
; legal aid
; access to justice
; human support
; service accessibility
; Bangladesh
1. Introduction
Access to justice requires more than formal rights. A person may need to recognise a legal problem, find an appropriate service, understand a procedure and obtain assistance that fits their circumstances. Cost, distance, time, language, disability and uncertainty about where to seek help can affect these steps. A people-centred approach examines users’ needs and the quality and accessibility of the response, not simply the existence of an institution.1
AI-supported tools can provide preliminary information, help organise a question and offer possible next steps outside ordinary office hours. These capabilities may be useful for navigating legal services, yet a plausible response is not necessarily accurate or current. A study of large language models and US legal questions found substantial legal hallucination in the tasks examined; its findings do not establish an error rate for every AI service or for Bangladeshi law.2 Incorrect or incomplete guidance can have serious consequences if a user misunderstands a deadline, shares sensitive information or fails to obtain qualified assistance.
The problem is whether respondents perceive AI-assisted legal aid as making particular forms of information and service navigation easier to access, and how those perceptions relate to human support and professional oversight. General support for AI complementing professionals is not the same as a functioning referral process or an assurance that a qualified person will be available when a tool reaches its limits. This article compares responses to these distinct propositions while recognising that differently worded statements cannot rank the causal effectiveness of safeguards.
The original study title refers to justice seekers. The supplied workbook records law students, law graduates, legal professionals and other respondents but does not identify whether anyone has sought legal assistance. The revised title therefore describes the topic and Bangladesh setting without claiming that the sample consists of justice seekers. Respondents’ perceptions are valuable as stated views; they do not measure the accessibility of a particular legal-aid service or an actual legal outcome.
The study describes perceptions of AI-assisted legal-aid navigation, availability and human support in the recorded sample. It asks: How often do respondents endorse different accessibility and oversight statements? How do support for AI complementing professionals and support for a human fallback coexist within respondents? How are general legal-information accessibility and conditional trust in professional oversight associated with two specific service-accessibility items? What can a perception survey contribute to an access-to-justice research agenda?
2. Literature Review and Theoretical Foundation
The literature connects perceived utility and trust with the broader question of whether a person can obtain meaningful legal assistance. This section distinguishes preliminary digital information, human assistance and the outcomes needed to evaluate access to justice.
2.1. AI Information and Access to Justice
AI-assisted legal information in this study refers to a potential tool for preliminary guidance or navigation. It does not imply that an AI system is a lawyer, has assessed a person’s complete circumstances, or can provide representation. Information about where to seek help can be a useful first step, but an answer is meaningful only if it is understandable, relevant, accurate and connected to an available service.
A study of legal-aid professionals in the United States reported potential uses of generative AI, including preliminary research, drafting and translation, alongside concerns about risks and support needs.3 That study involved professionals and a different institutional setting; it cannot establish what Bangladeshi justice seekers experience. The present questionnaire similarly measures beliefs about potential services rather than the results of deploying a named application.
People-centred justice research distinguishes legal needs from the ability to obtain a timely and appropriate response. An accessible interface can reduce an initial information barrier but cannot by itself prove that a user receives advice, meets a deadline, obtains representation or resolves a dispute.4 An evaluation of an AI-enabled referral should trace the journey from a question to a human response where required.
2.2. Accessibility and Its Dimensions
The shared questionnaire contains statements on locating assistance, access outside office hours and access for people far from providers. These are distinct aspects of possible accessibility. Finding an office does not demonstrate that the office is open, that the person is eligible for assistance or that the service can address their needs. Likewise, availability of general information at night is different from access to a qualified person at night.
The sample’s responses to these statements are not direct measurements of travel time, cost, use, disability access, connectivity or legal outcomes. A user might perceive a service as convenient while facing barriers involving language, device access or privacy. A valid empirical evaluation should identify which barrier is being measured and distinguish perceptions from observed service experience.5
2.3. Trust, Oversight and Responsible Use
Trust concerns a willingness to rely on another actor under conditions of uncertainty.6 In legal information, source identification, limitations and meaningful professional oversight may help a user assess whether further verification is required. The questionnaire asks whether such features would increase trust or willingness; it does not show that a particular legal AI tool has them or that they make the tool accurate.
Human oversight must have operational substance. A legal professional should be able to inspect relevant information, identify error and exercise independent judgment. A hand-off to human assistance is different from simply labelling a system as supervised. Confidentiality, accessibility of the human service and whether a referral is completed also matter. These issues should be evaluated in the context of an actual deployment rather than assumed from a user interface.7
2.4. Theoretical Foundation
The Unified Theory of Acceptance and Use of Technology 2 (UTAUT2) discusses determinants of use including expected performance, effort and facilitating conditions.8 It offers a vocabulary for perceived usefulness and support, but the questionnaire does not measure the full model, and this study does not test UTAUT2’s latent paths. Trust Theory adds attention to reliance and vulnerability.9 A people-centred access-to-justice framework asks whether the tool helps the person reach an appropriate and effective response.10
The perspectives address different outcomes. A respondent may consider an AI tool convenient without showing that it improves a legal remedy; a person may favour professional oversight without having an opportunity to use a service. The study therefore analyses selected items separately and uses the theories as interpretive lenses, not as established causal mechanisms.
2.5. Bangladesh Legal Context
The Legal Aid Services Act 2000 establishes a framework for legal aid in Bangladesh, including the National Legal Aid Services Organization.11 Its scope, eligibility rules and application to a particular service require assessment of the relevant provisions and facts. The present survey does not show that any particular AI tool is part of the national legal-aid system or meets legal or professional obligations.
International principles on legal aid in criminal justice emphasise accessible legal assistance and safeguards.12 They do not make every digital information service a substitute for qualified advice. A future Bangladesh-specific study can examine actual pathways through public and other legal-aid services, with a clear distinction between enforceable requirements, institutional practices and recommendations.
2.6. Analytical Expectations and Measurement
The ALSA block combines service navigation, time and distance access, human fallback and support for AI complementing professionals. The PUA and TAIL blocks similarly mix general accessibility, potential usefulness, conditional trust and oversight. These items are related but should not automatically be interpreted as indicators of a single validated construct.13 This article compares individual responses and uses exploratory associations rather than estimating unsupported latent paths.
Three paired comparisons examine support for complementing professionals against human fallback and distance access, and service navigation against distance access. Four exploratory correlations examine PUA1 and TAIL2 against ALSA1 and ALSA3. The comparisons are not claimed to be preregistered. Their wording and scope differ, and no item is a direct measure of a real service’s effectiveness.
3. Methodology
The methodology describes the dataset, the limits of its sample categories and the analysis. It distinguishes stated perceptions from observed access, professional advice and legal outcomes.
3.1. Design and Population
The study analyses a cross-sectional questionnaire completed by adults identified as AI-aware. The unit is an individual record; no legal-aid office, AI tool, referral, court case or legal outcome is observed. Current-status categories include law students, law graduates, advocates or legal professionals and other relevant respondents. The workbook does not identify a respondent’s status as an actual justice seeker, their experience with legal aid or their use of an AI legal service. The findings therefore describe the recorded sample rather than a representative population of justice seekers.14
3.2. Sample, Data Provenance and Eligibility
The workbook contains 385 records, each coded as consenting, adult eligible and aware of AI; respondent IDs are unique. The supplied files do not document recruitment dates, sampling frame, invitations, response rate or independent duplicate-screening procedures. A target of 385 is consistent with a conventional sample-size calculation under assumptions, but it does not establish random sampling or a national margin of error.15
The researcher identified the numerical questionnaire responses as real survey data. The Codebook describes the identifier sequence and optional contextual comments as synthetic. Their creation and relationship to the source records should be documented; the contextual-text field was excluded. An identifier label does not independently establish the origin of numerical responses, so transparent provenance and underlying fieldwork records remain important.
3.3. Questionnaire and Measures
Items use five response categories from strongly disagree to strongly agree. ALSA1 concerns whether AI legal aid could help people understand where to seek assistance; ALSA2 concerns information outside normal office hours; ALSA3 concerns people far from providers; ALSA4 concerns available human support when AI cannot meet a user’s needs; and ALSA5 concerns responsible use of AI to complement rather than replace professionals. PUA1 asks whether AI legal information can make basic guidance easier to access. TAIL2 asks whether oversight by a qualified legal professional would increase trust. Exact item wording is given in Appendix A.
These statements are hypothetical or normative. They do not measure actual waiting time, referral completion, cost, legal-aid eligibility or legal outcomes. No item is treated as a validated stand-alone measure of an entire theoretical construct. The supplied materials do not document item adaptation, cognitive interviews or a pilot; none is assumed.16
3.4. Data Collection and Screening
The focal numerical items are complete in the available 385 rows. The dataset codes consent and eligibility but does not establish administration mode, locations of recruitment, fieldwork dates, invitations, exclusions or response rate. Optional contextual text is excluded. Current status and location are reported as recorded and are not used to infer legal need or prior experience with legal services.
3.5. Statistical Analysis
For each focal item, the distribution and percentage endorsing agree or strongly agree are reported. Neutral responses are included in non-endorsement and are not classified as opposition. Pointwise Wilson intervals are descriptive uncertainty summaries; population interpretation depends on sampling assumptions not established in the workbook.17
Three paired contrasts compare endorsement of ALSA5 with ALSA4 and ALSA3, and ALSA1 with ALSA3, using exact McNemar tests and Holm correction across the three contrasts. Percentile intervals are calculated from whole-respondent bootstrap resamples and are pointwise. These items ask different questions and are not forced choices, so their differences do not rank the value of AI, human support or a policy intervention.18
Four exploratory Spearman correlations assess PUA1 and TAIL2 with ALSA1 and ALSA3. Holm adjustment covers those four p-values; bootstrap intervals are pointwise. These are single ordinal items measured simultaneously. The coefficients do not show that usefulness or oversight causes a change in access, nor do non-significant tests establish no association.19
3.6. Measurement Limitations
The questionnaire groups different concepts under common block labels. Internal consistency alone cannot establish a valid scale, and composite analysis may obscure differences between availability, navigation and human support.20 The item-level approach keeps the basis of each result visible. The questions also differ in scope: support for a general principle of complementarity is not identical to actual access to a human when a tool cannot help. Respondents may interpret “legal aid” or “qualified legal professional” in different ways.
3.7. Ethical Considerations
The authors state that formal ethics approval was not obtained because their institutional setting did not provide an accessible formal review mechanism applicable to this non-interventional adult questionnaire research. They report informed consent, voluntary participation, an anonymous questionnaire, an opportunity to discontinue and aggregate reporting. The contextual-text field was excluded. Responsibility for ethical conduct remained with the researchers; their full statements are in Section 8.
4. Results
The results describe the 385 coded records. They concern perceived possibilities and safeguards, not observed legal-aid access or the experience of a verified group of justice seekers.
4.1. Respondent Profile
Table 1 presents recorded current status and location. These fields do not establish legal-aid use or a respondent’s legal need.
Law students account for 227 records (59.0%), law graduates for 73 (19.0%), advocates or legal professionals for 58 (15.1%) and other respondents for 27 (7.0%). The analysis therefore uses the mixed, law-oriented sample as its stated scope.
4.2. Perceptions of Accessibility and Human Support
Table 2 reports the ALSA items separately. Their topics range from navigating to assistance to supporting a human fallback and the principle of complementing professionals.
Service navigation was endorsed by 276 respondents (71.7%), information outside office hours by 265 (68.8%) and improved access for people far from providers by 249 (64.7%). Human support when AI cannot address a need was endorsed by 253 (65.7%), and responsible use to complement rather than replace professionals by 330 (85.7%). These are endorsements of different statements, not evidence of an observed referral or successful legal outcome.
4.3. Usefulness and Conditional Trust
Table 3 presents the PUA and TAIL responses. The items on source identification and professional oversight are conditional trust statements rather than assessments of a named system.
General access to basic legal guidance was endorsed by 257 respondents (66.8%). Conditional trust under professional oversight was endorsed by 266 (69.1%); clear limitations by 325 (84.4%) and caution about acting without verification by 329 (85.5%). The statements describe what respondents believe or prefer, not whether safeguards exist in practice.
4.4. Paired Contrasts and Joint Responses
Table 4 compares selected endorsements within respondents. “First only” and “second only” include neutral answers on the item that was not endorsed.
Support for complementing professionals exceeded endorsement of human fallback by 20.0 percentage points, and exceeded the distance-access statement by 21.0 points. Service-navigation endorsement exceeded the distance item by 7.0 points. The results describe response distributions for differently worded items; they do not show that one safeguard is more effective or that people oppose an unendorsed statement.
Table 5 shows joint endorsement of complementarity and human fallback in the same respondents.
A total of 225 respondents endorsed both statements, while some endorsed only one. This distribution does not establish whether an actual human service is available or whether a respondent has tried to obtain help.
4.5. Exploratory Associations
Table 6 presents four Spearman rank correlations between general accessibility or conditional trust in oversight and two specific accessibility items. P-values are adjusted across four exploratory tests; confidence intervals are pointwise.
The correlations are small and none provides clear evidence against a null association after Holm adjustment. This is not evidence that the variables are unrelated. The items concern distinct propositions and were measured simultaneously; their associations cannot establish the effects of an AI tool, professional oversight or a legal-aid programme.
5. Discussion
The findings distinguish general support for an assistive role from views on particular access and support conditions. Interpretation must remain within the recorded mixed sample and the questionnaire’s hypothetical wording.
5.1. Complementarity and Human Fallback
Most respondents endorsed responsible AI use to complement rather than replace legal-aid professionals. A smaller proportion endorsed the separate statement that human support should remain available when AI cannot meet a user’s needs. The paired contrast is a difference between response distributions, not evidence that respondents reject human assistance or that complementarity is achievable without it. ALSA5 expresses a general principle while ALSA4 describes a particular support condition, and neutral responses contribute to non-endorsement.
The joint distribution shows that 225 respondents endorsed both. Their answers are compatible with support for an assistive role and human fallback, but do not establish how either feature would operate. A tool could claim to complement legal professionals while making a referral difficult or leaving the user without an available provider. Future studies should specify referral availability, response time, cost and responsibility before comparing alternatives.21
5.2. Navigation, Time and Distance
The navigation and distance items invite different judgments. Knowing where to seek assistance may be feasible through a digital tool even when the person cannot afford travel, obtain connectivity or secure an appointment. Similarly, information outside office hours is not equivalent to qualified human advice being available at that time. The survey does not record these downstream outcomes.
A people-centred assessment should therefore follow a user journey: identification of a legal need, receipt of understandable information, appropriate referral and contact with a qualified service when needed. This approach avoids equating a positive perception of convenience with the resolution of a legal problem.22
5.3. Trust and Legal-Information Quality
Conditional trust in professional oversight and the general-accessibility item have small exploratory correlations with the specific navigation and distance items. After correction, the analysis does not establish clear associations. The result is compatible with a range of relationships and should not be read as proof that professional oversight is unimportant. The questionnaire also does not specify a particular tool or measure whether an oversight process works.
The risk of incorrect legal information makes verification important. Dahl and colleagues documented legal hallucinations under the US legal-question conditions they studied. Their findings motivate testing a specific tool against current Bangladeshi legal sources; they cannot be transferred as a numerical error rate for the present setting.23 Professional review should include the capacity to identify errors and direct a person to appropriate help, while protecting confidentiality.
5.4. Bangladesh Legal-Aid Context
Bangladesh’s Legal Aid Services Act 2000 establishes a legal-aid framework, including an organisation to carry out the Act’s purposes.24 The questionnaire does not identify an AI service operated under this framework, and no conclusion about compliance or eligibility follows from the survey. A particular user’s legal-aid entitlement depends on applicable law and facts. The study’s recommendations should be understood as proposed ways to evaluate service design, not descriptions of current institutional practice.
The scope of the sample also matters. Legal students and professionals may imagine a service differently from someone who has recently experienced a legal problem. Future work should identify users and non-users, record the type of legal need and document recruitment. It should also include participants whose language, disability, connectivity and location affect access, without assuming that a digital interface solves those barriers.
5.5. Measurement, Provenance and Inference
The small correlations and paired differences are not a test of a full acceptance model. Single items have different wording, and all measures are self-reported at one time. A composite could conceal distinctions between a stated intention, a general view of accessibility and a conditional trust statement. The study preserves those distinctions and does not claim causal effects.25
The numerical responses were identified by the researcher as real survey data, while the Codebook labels the identifier and contextual-text fields synthetic. Documentation should explain how these fields were prepared and how the survey records were assembled. Recruitment dates, response handling and consent records should also be retained. This information supports reproducibility and permits assessment of selection without inventing a sampling process.
5.6. Integrated Interpretation
UTAUT2 helps frame anticipated usefulness and facilitating conditions; Trust Theory highlights reliance under uncertainty; and a people-centred justice perspective asks whether an appropriate response is obtained.26 The survey cannot validate a model connecting these elements. Its contribution is a descriptive separation of possible information access, professional complementarity and human fallback that can guide a more rigorous evaluation of real services.
6. Theoretical and Practical Implications
The implications are proposals informed by item-level responses and relevant literature. They are not evidence that any intervention, referral system or oversight arrangement improves access.
6.1. Theoretical Implications
First, perceived digital availability should be kept distinct from access to appropriate legal assistance and from legal outcomes. A question about information outside office hours measures a different proposition from whether a person can reach a qualified provider. Second, professional oversight and human fallback should be specified separately: oversight concerns review of the system or its outputs, while fallback concerns whether a user can obtain personal assistance when the system is insufficient.
Third, future studies could develop separate measures of perceived utility, calibrated trust, referral accessibility and actual service experience. The full UTAUT2 model would require appropriate measures of its constructs; the present survey does not justify claims that its paths or the broader access-to-justice framework are confirmed.27
6.2. Practical and Policy Implications
A provider considering AI-assisted legal information could identify the organisation responsible, state the jurisdiction and limits of the answer, and provide links or contact routes to authoritative sources and human assistance. It should avoid implying that a general response is a substitute for case-specific advice. Human support needs to be reachable and appropriate to the user’s problem; a referral link alone is not a completed referral.
Evaluation should use realistic, locally relevant scenarios and include accuracy checks against current legal sources. It could measure whether users understand the answer, protect confidential information, reach a qualified service and receive a response. Tests should involve people with different levels of literacy, access and legal need, and should investigate whether a tool excludes some users. Legal-aid institutions and providers remain responsible for meeting applicable duties in their own context.28
7. Conclusion, Limitations and Future Research
Among 385 recorded respondents, 71.7% endorsed the proposition that AI-assisted legal aid could help people identify where to seek assistance; 64.7% endorsed potential access for people far from providers; 65.7% endorsed human fallback; and 85.7% endorsed responsible AI use to complement professionals. The 20.0-point paired difference between complementarity and human fallback is a difference between distinct statements, not a ranking of their causal impact. Exploratory associations did not establish clear relationships after adjustment for four tests.
The study is limited by cross-sectional self-report, an undocumented sampling and fieldwork process, a mixed law-oriented sample, the absence of an identified justice-seeker or legal-aid-user variable and heterogeneous individual items. The numerical responses were identified by the researcher as real, while the Codebook describes the identifier and contextual-comment fields as synthetic; the preparation of these fields should be documented. The survey does not measure actual service access or legal outcomes.
Future research should recruit and identify people with defined legal needs, record the dates and procedures of fieldwork, and separate general AI familiarity from experience of a legal-aid service. A service evaluation could vary source information, human oversight, referral availability and follow-up, then measure comprehension, referral completion, accuracy and user experience. Interviews and accessible user testing could reveal barriers that a general agreement scale cannot capture. A validated study with a defined population would permit more informative tests of the theoretical framework.
Ethical Approval and Oversight
Formal ethical approval was not obtained because, at the time of the study, the authors' institutional setting did not provide an accessible formal review mechanism applicable to this non-interventional adult questionnaire research. The study nevertheless followed participant-protection safeguards, including informed consent, voluntary participation, the opportunity to discontinue, an anonymous questionnaire and aggregate reporting. The methodological safeguards are detailed in Section 3.7.
Informed Consent
Informed consent was obtained from adult participants before questionnaire completion. The questionnaire explained the study purpose, voluntary participation, anonymity and the option to discontinue. All analysed records contain affirmative consent and adult-eligibility responses.
Data Availability
The numerical data supporting the reported analyses are held by the research team. Requests for de-identified data and analytical specifications may be directed to the corresponding author. Any sharing is subject to participant confidentiality and applicable institutional requirements. The contextual-text field was excluded from the analysis.
Conflicts of Interest
The authors declare that they have no conflicts of interest.
Author Contributions
Tanjila Jerin: Lead student researcher; conceptualisation; preparation of study materials; organisation and curation of survey data; manuscript development. The project was conducted under the supervision of Prof. Dr Kazi Abdul Mannan. Kazi Abdul Mannan: Conceptual and methodological guidance; supervision; interpretation of statistical findings; critical review and editing of the manuscript.
Generative AI Disclosure
Generative AI assisted manuscript drafting, language editing, organisation and the development of analytical code. Statistical estimates were computed programmatically from the numerical survey data. Responsibility for the study's analysis, interpretation and reporting rests with the authors.
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 |
|---|---|
| ALSA1 | AI-assisted legal aid could help people understand where to seek legal assistance. |
| ALSA2 | AI could make legal information available outside normal office hours. |
| ALSA3 | AI-assisted legal aid could improve access for people living far from legal service providers. |
| ALSA4 | Human support should remain available when AI legal aid cannot address a user's needs. |
| ALSA5 | I would support responsible use of AI to complement, not replace, legal aid professionals. |
| PUA1 | AI-based legal information can make basic legal guidance easier to access. |
| PUA2 | AI legal assistants can reduce the time needed to obtain preliminary legal information. |
| PUA3 | AI-enabled services can be useful for people who face barriers to conventional legal services. |
| PUA4 | AI legal tools can improve the convenience of obtaining basic legal information. |
| PUA5 | I would consider using an AI legal service for preliminary legal guidance. |
| TAIL1 | I would trust AI legal information more when its sources are clearly identified. |
| TAIL2 | I would trust an AI legal service more if a qualified legal professional provides oversight. |
| TAIL3 | I believe AI legal tools should clearly state their limitations. |
| TAIL4 | I am cautious about acting on AI-generated legal information without verification. |
| TAIL5 | Transparent explanations increase my willingness to use AI legal services. |
Source: Group 2 Codebook. Answers range from strongly disagree to strongly agree.
Items are analysed separately and do not establish a single validated latent construct.
Table of legislation
Legal Aid Services Act 2000 (Bangladesh Act No 6 of 2000)
| 1 | Organisation for Economic Co-operation and Development, Toolkit for Access to Justice and People-Centred Justice Systems (OECD 2025) <https://www.oecd.org/en/publications/toolkit-for-access-to-justice-and-people-centred-justice-systems_aecf7f78-en.html> accessed 27 October 2026. |
| 2 | Matthew Dahl and others, ‘Large Legal Fictions: Profiling Legal Hallucinations in Large Language Models’ (2024) 16 Journal of Legal Analysis 64. |
| 3 | Colleen V Chien and Miriam Kim, ‘Generative AI and Legal Aid: Results from a Field Study and 100 Use Cases to Bridge the Access to Justice Gap’ (2025) 57 Loyola of Los Angeles Law Review 903. |
| 4 | OECD, Toolkit (n 1). |
| 5 | OECD, Toolkit (n 1). |
| 6 | Roger C Mayer, James H Davis and F David Schoorman, ‘An Integrative Model of Organizational Trust’ (1995) 20 Academy of Management Review 709. |
| 7 | Dahl and others (n 2); Chien and Kim (n 3). |
| 8 | Viswanath Venkatesh, James YL Thong and Xin Xu, ‘Consumer Acceptance and Use of Information Technology: Extending the Unified Theory of Acceptance and Use of Technology’ (2012) 36 MIS Quarterly 157. |
| 9 | Mayer, Davis and Schoorman (n 6). |
| 10 | OECD, Toolkit (n 1). |
| 11 | Legal Aid Services Act 2000 (Bangladesh Act No 6 of 2000). |
| 12 | United Nations General Assembly, ‘United Nations Principles and Guidelines on Access to Legal Aid in Criminal Justice Systems’ UNGA Res 67/187 (20 December 2012) UN Doc A/RES/67/187. |
| 13 | Godfred O Boateng and others, ‘Best Practices for Developing and Validating Scales for Health, Social, and Behavioral Research: A Primer’ (2018) 6 Frontiers in Public Health 149; Klaas Sijtsma, ‘On the Use, the Misuse, and the Very Limited Usefulness of Cronbach’s Alpha’ (2009) 74 Psychometrika 107. |
| 14 | Erik von Elm and others, ‘The Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) Statement: Guidelines for Reporting Observational Studies’ (2007) 4 PLOS Medicine e296. |
| 15 | William G Cochran, Sampling Techniques (3rd edn, John Wiley & Sons 1977). |
| 16 | Boateng and others (n 13). |
| 17 | Cochran (n 15). |
| 18 | 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). |
| 19 | Holm (n 18); 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. |
| 20 | Sijtsma (n 13). |
| 21 | Chien and Kim (n 3); OECD, Toolkit (n 1). |
| 22 | OECD, Toolkit (n 1). |
| 23 | Dahl and others (n 2). |
| 24 | Legal Aid Services Act 2000 (n 11). |
| 25 | Sijtsma (n 13); Wasserstein and Lazar (n 19). |
| 26 | Venkatesh, Thong and Xu (n 8); Mayer, Davis and Schoorman (n 6); OECD, Toolkit (n 1). |
| 27 | Venkatesh, Thong and Xu (n 8); Boateng and others (n 13). |
| 28 | Dahl and others (n 2); Chien and Kim (n 3). |
References
- Altman, DG; Bland, JM. Absence of Evidence Is Not Evidence of Absence. BMJ 1995, 311, 485. [Google Scholar] [CrossRef] [PubMed]
- Boateng, GO; et al. Best Practices for Developing and Validating Scales for Health, Social, and Behavioral Research: A Primer. Frontiers in Public Health 2018, 6, 149. [Google Scholar] [CrossRef] [PubMed]
- Chien, CV; Kim, M. Generative AI and Legal Aid: Results from a Field Study and 100 Use Cases to Bridge the Access to Justice Gap. Loyola of Los Angeles Law Review 2025, 57, 903. [Google Scholar]
- Cochran, WG. Sampling Techniques, 3rd edn; John Wiley & Sons, 1977. [Google Scholar]
- Dahl, M. ‘Large Legal Fictions: Profiling Legal Hallucinations in Large Language Models’. Journal of Legal Analysis 2024, 16, 64. [Google Scholar] [CrossRef]
- Efron, B; Tibshirani, RJ. An Introduction to the Bootstrap; Chapman & Hall, 1993. [Google Scholar]
- Holm, S. ‘A Simple Sequentially Rejective Multiple Test Procedure’. Scandinavian Journal of Statistics 1979, 6, 65. [Google Scholar]
- Mayer, RC; Davis, JH; Schoorman, FD. An Integrative Model of Organizational Trust. Academy of Management Review 1995, 20, 709. [Google Scholar] [CrossRef]
- McNemar, Q. ‘Note on the Sampling Error of the Difference between Correlated Proportions or Percentages’. Psychometrika 1947, 12, 153. [Google Scholar] [CrossRef] [PubMed]
- Organisation for Economic Co-operation and Development. Toolkit for Access to Justice and People-Centred Justice Systems (OECD 2025) accessed 27 October 2026. Available online: https://www.oecd.org/en/publications/toolkit-for-access-to-justice-and-people-centred-justice-systems_aecf7f78-en.html (accessed on 27 October 2026).
- Sijtsma, K. ‘On the Use, the Misuse, and the Very Limited Usefulness of Cronbach’s Alpha’. Psychometrika 2009, 74, 107. [Google Scholar] [CrossRef] [PubMed]
- United Nations General Assembly. United Nations Principles and Guidelines on Access to Legal Aid in Criminal Justice Systems. December 2012; 67/187.
- Venkatesh, V; Thong, JYL; Xu, X. ‘Consumer Acceptance and Use of Information Technology: Extending the Unified Theory of Acceptance and Use of Technology’. MIS Quarterly 2012, 36, 157. [Google Scholar] [CrossRef]
- von Elm, E. ‘The Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) Statement: Guidelines for Reporting Observational Studies’. PLOS Medicine 2007, 4, e296. [Google Scholar] [CrossRef] [PubMed]
- Wasserstein, RL; Lazar, NA. ‘The ASA Statement on p-Values: Context, Process, and Purpose’. American Statistician 2016, 70, 129. [Google Scholar] [CrossRef]
Table 1.
Recorded respondent status and location.
| Measure | Category | Count (%) |
|---|---|---|
| Status | Law student | 227 (59.0%) |
| Law graduate | 73 (19.0%) | |
| Advocate or legal professional | 58 (15.1%) | |
| Other relevant respondent | 27 (7.0%) | |
| Location | Dhaka | 150 (39.0%) |
| District town | 86 (22.3%) | |
| Other city corporation | 78 (20.3%) | |
| Upazila-rural area | 61 (15.8%) | |
| Other | 10 (2.6%) |
N = 385. Status and location are reported as recorded; legal-aid use and justice-seeker status are not identified.
Table 2.
Perceived legal-aid access and support.
| Item | Counts 1 / 2 / 3 / 4 / 5 | Endorsed n (%) | 95% CI (%) |
|---|---|---|---|
| ALSA1 | 1 / 14 / 94 / 187 / 89 | 276 (71.7%) | 67.0–76.0 |
| ALSA2 | 0 / 17 / 103 / 183 / 82 | 265 (68.8%) | 64.0–73.3 |
| ALSA3 | 1 / 14 / 121 / 180 / 69 | 249 (64.7%) | 59.8–69.3 |
| ALSA4 | 0 / 18 / 114 / 176 / 77 | 253 (65.7%) | 60.8–70.3 |
| ALSA5 | 0 / 3 / 52 / 169 / 161 | 330 (85.7%) | 81.9–88.9 |
N = 385. Endorsed means agree or strongly agree. Non-endorsement includes neutral responses. Wilson intervals are pointwise. Exact wording for focal items is in Appendix A.
Table 3.
Perceived usefulness and conditional trust.
| Item | Counts 1 / 2 / 3 / 4 / 5 | Endorsed n (%) | 95% CI (%) |
|---|---|---|---|
| PUA1 | 0 / 11 / 117 / 176 / 81 | 257 (66.8%) | 61.9–71.3 |
| PUA2 | 1 / 20 / 104 / 171 / 89 | 260 (67.5%) | 62.7–72.0 |
| PUA3 | 1 / 17 / 97 / 185 / 85 | 270 (70.1%) | 65.4–74.5 |
| PUA4 | 0 / 20 / 104 / 175 / 86 | 261 (67.8%) | 63.0–72.3 |
| PUA5 | 0 / 14 / 125 / 168 / 78 | 246 (63.9%) | 59.0–68.5 |
| TAIL1 | 0 / 14 / 111 / 173 / 87 | 260 (67.5%) | 62.7–72.0 |
| TAIL2 | 0 / 24 / 95 / 182 / 84 | 266 (69.1%) | 64.3–73.5 |
| TAIL3 | 0 / 5 / 55 / 172 / 153 | 325 (84.4%) | 80.5–87.7 |
| TAIL4 | 0 / 3 / 53 / 163 / 166 | 329 (85.5%) | 81.6–88.6 |
| TAIL5 | 0 / 3 / 45 / 176 / 161 | 337 (87.5%) | 83.9–90.5 |
N = 385. Endorsed means agree or strongly agree. Non-endorsement includes neutral responses. Wilson intervals are pointwise. Exact wording for focal items is in Appendix A.
Table 4.
Paired endorsement comparisons.
| Contrast | First only / second only | Difference pp | 95% CI pp | Holm p |
|---|---|---|---|---|
| ALSA5 − ALSA4 | 105 / 28 | 20.0 | 14.3–25.5 | <.001 |
| ALSA5 − ALSA3 | 114 / 33 | 21.0 | 15.3–27.0 | <.001 |
| ALSA1 − ALSA3 | 90 / 63 | 7.0 | 0.8–13.2 | 0.035 |
N = 385; pp = percentage points. Exact McNemar tests; Holm adjustment across three tests. Intervals are pointwise percentiles from 10,000 respondent-level bootstrap resamples. Wording differs across items.
Table 5.
Joint endorsement of complementarity and human fallback.
| Response configuration | Count (%) |
|---|---|
| Both endorsed | 225 (58.4%) |
| Complementarity only | 105 (27.3%) |
| Human fallback only | 28 (7.3%) |
| Neither endorsed | 27 (7.0%) |
N = 385. Neither includes neutral responses; these categories describe item combinations, not fixed respondent types.
Table 6.
Exploratory associations with accessibility items.
| Items | Spearman ρ | 95% CI | p | Holm p |
|---|---|---|---|---|
| PUA1 / ALSA1 | 0.078 | -0.025–0.174 | 0.127 | 0.467 |
| PUA1 / ALSA3 | 0.067 | -0.029–0.168 | 0.190 | 0.467 |
| TAIL2 / ALSA1 | 0.025 | -0.075–0.117 | 0.621 | 0.621 |
| TAIL2 / ALSA3 | 0.080 | -0.017–0.181 | 0.117 | 0.467 |
N = 385. Four correlation p-values adjusted by Holm. Pointwise percentile intervals use 2,000 respondent-level resamples and are not adjusted for multiplicity.
Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content. |
© 2026 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).
Copyright: This open access article is published under a Creative Commons CC BY 4.0 license, which permit the free download, distribution, and reuse, provided that the author and preprint are cited in any reuse.