More Than Just Access: Generative AI as Communication Intermediary for Blind and Low-Vision Users
Protik Dey, Mohd Saifuzzaman, Taslima Akter
cs.HC · cs.AI · cs.ET
Quick verdict
Timely interview study (n=19) framing GenAI as a communication intermediary for BLV users; valuable framing, but thin evidence and limited novelty.
Summary
This arXiv preprint reports a qualitative interview study of how blind and low-vision (BLV) people use generative AI tools such as ChatGPT, Google Gemini, Be My AI, and Seeing AI as 'communication intermediaries' that stand in for reading, describing, and asking another person for help. The authors conducted semi-structured Zoom interviews with 19 BLV adults in the United States, recruited mainly through the National Federation of the Blind and selected from 33 respondents based on GenAI experience; all were primary screen-reader users, and two authors analyzed the data using reflexive thematic analysis. Findings form four themes: BLV users treat GenAI answers with calibrated skepticism, and verification paradoxically depends on prior knowledge (§4.1); substitution of interpersonal help is deliberate, with low-stakes tasks delegated to GenAI and high-stakes ones reserved for humans (§4.2); inaccessible interfaces block communication entirely, acting as a threshold condition (§4.3); and privacy disclosure is a conscious trade-off mediated by trust in companies (§4.4). Main limitations: only five participant quotes support the four themes, the sample excludes infrequent users and magnification users, and the contribution overlaps substantially with prior work the authors themselves cite (refs 33 and 35).
Review complete
Assessment at a glance
4
risk flags identified
● high 0
● medium 2
● low 2
5 strengths · 5 weaknesses · 5 author questions
AI text signal = Binoculars zero-shot detector score over 20 sampled passages (Fast-DetectGPT co-reported) — absence does not prove human authorship, and a flag does not imply misconduct.
Levels reflect verification complexity — not a quality score. Evolve never rates your paper.
Key claims under review
- BLV users treat GenAI outputs with calibrated skepticism, and verification paradoxically depends on prior knowledge, constraining epistemic autonomy exactly when independent judgment matters most (§4.1).
- Substitution of human help is deliberate: BLV users delegate low-stakes tasks to GenAI and reserve high-stakes ones for people, trading reduced social burden against sincerity and accountability (§4.2, §5).
- Accessibility functions as a threshold condition — an inaccessible interface eliminates the communicative exchange entirely rather than degrading it (§4.3).
- Privacy disclosure to GenAI is a conscious but incompletely informed trade-off, with corporate ownership serving as an imperfect and unevenly applied trust proxy (§4.4, §5).
- Current GenAI systems place the burden of detecting failure on BLV users, motivating explicit non-visual uncertainty signaling and standardized cross-platform policy protections (§5, §6).
Strengths
Clear conceptual framing of GenAI as a 'communication intermediary'
§1-§2 articulate the shift from access tool to substitute for interpersonal help, linking epistemic autonomy (unverifiable outputs) and relationships (who is trusted, what is disclosed, who is accountable) — a framing that organizes the entire paper and its recommendations.
Methodological transparency on procedure
§3 reports eligibility criteria, IRB approval, recruitment channels (NFB plus snowball), interview length (45-60 min), compensation ($30 gift card), and explicitly justifies omitting inter-rater reliability with citations [25, 28], consistent with reflexive TA norms.
The threshold-condition insight is genuinely useful
Recommendations are linked to findings rather than generic
Sharp accountability asymmetry analysis in §5
Weaknesses
Unusually sparse evidence density
Sampling bias is introduced but never acknowledged
Limited novelty over cited prior work
Reflexivity claimed but positionality absent
Scope narrower than claims
Full Review Report
$1.99 · one-timeUnlock the complete preliminary analysis Evolve produced for this paper.
Report preview
Scope & Claims
This arXiv preprint (2609.15696, cs.HC, v1 Sep 2026) reports a qualitative interview study examining generative AI tools — ChatGPT, Google Gemini, Be My AI, Seeing AI — as "communication intermediaries" for blind and low-vision (BLV) users: channels that increasingly stand in for reading, describing, and asking another person for help. The stated contributions are (a) an account of how BLV users experience the substitution of GenAI for interpersonal help, (b) identification of what shapes willingness to rely on it, and (c) design and policy implications (§1, §6). Central claims: BLV users exercise calibrated skepticism toward GenAI outputs (§4.1); substitution is deliberate, with low-stakes tasks delegated to AI and high-stakes ones reserved for humans (§4.2, §5); accessibility is a threshold condition rather than a quality factor (§4.3); and disclosure is a conscious but incompletely informed trade-off (§4.4).
The remaining sections cover Methodology Assessment, Statistical Evidence, Results & Reproducibility, Limitations & Risks, and Verdict & Recommendations.
- All 5 strengths & 5 weaknesses, analyzed in detail
- 4 risk flags with severity, section & reviewer questions
- Methods & statistical evidence assessment
- 5 questions a peer reviewer would ask
- Full six-dimension evaluation report & recommendations
One-time purchase · this paper only
One-time purchase, this paper only. AI-generated preliminary analysis — not peer review.
Frequently asked questions
What does this paper find about blind users and ChatGPT?
Based on interviews with 19 blind and low-vision adults in the US, the paper finds that BLV users deliberately use generative AI as a stand-in for asking people: they delegate low-stakes tasks (reading labels, drafting emails, describing scenes) to tools like ChatGPT, Be My AI, and Seeing AI, but reserve high-stakes matters (medical, financial, navigation-critical) for trusted humans. They treat AI answers with calibrated skepticism, worry about what they disclose to these systems, and cannot use tools whose interfaces their screen readers cannot operate.
How many participants were in the study and how were they recruited?
Nineteen blind or low-vision adults, all US residents, English speakers, and primary screen-reader users. Most were recruited through the National Federation of the Blind, with some snowball sampling; 19 were selected from 33 applicants based on their generative AI experience. Participants received a $30 Amazon gift card, and interviews lasted 45-60 minutes over Zoom under university IRB approval.
What design recommendations does the paper make for accessible AI?
Five: (1) explicitly announce stalls, failures, and low-confidence answers in non-visual form; (2) treat accessibility as a persistent, system-wide requirement rather than an afterthought; (3) offer task-sensitive modes that are more uncertainty-forward for high-stakes domains like medical, financial, or navigation queries; (4) communicate data retention and sharing practices at the point of disclosure, in screen-reader-accessible terms rather than in a privacy policy; (5) establish standardized, enforceable baseline privacy and accessibility protections across GenAI products so individual users need not vet each company.
Is the paper's data or code publicly available?
No. The paper mentions IRB approval, informed consent, and an interview protocol, but includes no data availability statement, no shared codebook, and no supplementary materials, so the qualitative analysis cannot be independently audited.
About this review
This is an AI-generated preliminary review by Evolve.Pub, produced from the openly available text of the paper. It is not peer review and does not represent the authors' views. The canonical record of the paper is at the publisher link above.
- Extracted text appears complete (abstract through references, §1-§7), but any figures or tables were not present in the extraction to analyze.
- Only five participant quotes (P8, P11, P15, P17, P19) are visible in the provided text; additional supporting data could exist in an appendix not captured by extraction.
- This is an arXiv v1 preprint (Sep 2026); no peer-review record is visible, so judgments reflect the manuscript as submitted.
Go deeper on this paper
Wrote a paper? Stress-test it before reviewers do.
Upload your manuscript for the same structured scrutiny — reasoning, statistics, evidence, figures, and reproducibility.