Citing qualcode.ai

"Open-ended survey responses (N = [total responses]) were coded using qualcode.ai (https://qualcode.ai), an AI-assisted qualitative coding platform. The system employs a dual-rater methodology where two independent large language models ([OpenAI model name] and [Anthropic model name]) code each response, enabling calculation of inter-rater reliability metrics. Models were configured with temperature 0.0, following provider recommendations for classification tasks (OpenAI, 2024; Anthropic, 2024).

Categories were defined by the research team based on [theoretical framework / prior research / inductive analysis]. [N] training examples were provided to guide the AI classification. Inter-rater agreement between the two AI raters was [interpretation] (Cohen's κ = [value]; Krippendorff's α = [value]; percent agreement = [value]%).

Disagreements (n = [count], [percentage]%) were reviewed and reconciled by [author initials / "two independent researchers"]. Final category assignments were used for subsequent analysis."

"Responses were coded using qualcode.ai in zero-shot mode, where category assignments were based solely on the category names and descriptions provided by the research team. No training examples were used. This approach tests whether the categories are sufficiently clear for unambiguous classification."

"Disagreements were resolved through [independent review by two coders / discussion until consensus / majority vote among three reviewers]. Reconciled classifications were [used as additional training data for a second coding pass / retained as final codes]."

"Coding categories were developed using an AI-assisted inductive approach via qualcode.ai's auto-suggest feature. Two independent large language models (OpenAI GPT-5.2 and Anthropic Claude Opus 4.5) analyzed a random sample of [N] responses to identify emergent themes. Categories identified by both models were flagged as high-confidence; categories identified by only one model were flagged as low-confidence.

The AI-suggested categories were reviewed and refined by the research team. [Describe modifications: e.g., "Three categories were merged due to conceptual overlap, and two low-confidence categories were removed as not relevant to the research questions."] The final coding guide consisted of [N] categories with researcher-authored descriptions and inclusion/exclusion criteria.

This coding guide was then applied using qualcode.ai's dual-rater classification, where [OpenAI model name] and [Anthropic model name] independently coded all [N] responses..."

"Initial coding categories were generated using qualcode.ai's auto-suggest feature, which employs dual independent AI analysis (OpenAI GPT-5.2 and Anthropic Claude Opus 4.5) to identify themes from a sample of [N] responses. Categories agreed upon by both models were retained with minor editorial refinements to category names and descriptions. The research team validated that all categories were conceptually distinct and relevant to the research questions before proceeding with classification."

"AI-suggested categories served as a starting point for coding guide development. Two independent large language models identified [N] initial categories from a sample of responses. The research team substantially revised these suggestions based on [theoretical framework / domain expertise / prior literature], resulting in a final coding guide of [N] categories that [describe how they differ: e.g., "better aligned with established constructs in the literature" or "reflected more granular distinctions relevant to our research questions"]."

"This research used AI-assisted coding via qualcode.ai. The AI models ([OpenAI model name] and [Anthropic model name]) performed initial coding of open-ended survey responses based on researcher-defined categories. All AI-generated codes were subject to inter-rater reliability assessment, and disagreements were resolved by human researchers.

Survey response data was processed via OpenAI and Anthropic enterprise APIs under data processing agreements. No participant data was used for AI model training. All processing complied with [GDPR / institutional ethics requirements]."

"This research used AI-assisted coding via qualcode.ai. Coding categories were initially generated using dual AI analysis (OpenAI GPT-5.2 and Anthropic Claude Opus 4.5), then reviewed and refined by the research team. Classification was performed by two independent AI models ([OpenAI model name] and [Anthropic model name]) with inter-rater reliability assessment. All AI-suggested categories and AI-generated codes were subject to human review.

Survey response data was processed via OpenAI and Anthropic enterprise APIs under data processing agreements. No participant data was used for AI model training. All processing complied with [GDPR / institutional ethics requirements]."

"Open-ended response data will be processed using qualcode.ai, an AI-assisted coding platform. Data is transmitted to OpenAI and Anthropic APIs for classification under enterprise data processing agreements that prohibit use of data for model training. Data is processed in EU data centers (Frankfurt) and deleted within 30 days. No identifying information is required for the coding process; responses can be anonymized before upload if required."

Qualitative Coding of Open-Ended Responses

Participants' responses to the open-ended question "What factors influenced your decision?" (N = 1,247) were coded using qualcode.ai (Version 1.0; https://qualcode.ai), an AI-assisted qualitative coding platform employing dual-rater methodology.

A coding scheme with eight categories was developed based on the Health Belief Model (Rosenstock, 1974): perceived susceptibility, perceived severity, perceived benefits, perceived barriers, cues to action, self-efficacy, knowledge, and other. Category descriptions and five training examples per category were provided to the system.

Two independent large language models ([OpenAI model name] and [Anthropic model name]) coded each response. Inter-rater agreement was substantial (Cohen's κ = 0.74; Krippendorff's α = 0.76; percent agreement = 82.3%). Disagreements (n = 221, 17.7%) were reviewed by two authors (AB, CD) who reached consensus on final classifications.

Response data was processed in compliance with GDPR via EU data centers. No data was used for AI model training per enterprise API agreements.

Qualitative Coding of Open-Ended Responses

Participants' responses to the open-ended question "How has remote work affected your wellbeing?" (N = 2,156) were coded using qualcode.ai (Version 1.0; https://qualcode.ai), an AI-assisted qualitative coding platform.

Coding categories were developed using an AI-assisted inductive approach. Two independent large language models (OpenAI GPT-5.2 and Anthropic Claude Opus 4.5) analyzed a random sample of 300 responses to identify emergent themes. Categories identified by both models were flagged as high-confidence. The dual-AI analysis initially suggested 12 categories; 8 were identified by both models (high-confidence) and 4 by only one model (low-confidence).

The research team reviewed all suggestions, merging two overlapping categories ("work-life balance" and "boundary management" into "work-life boundaries") and removing one low-confidence category ("commuting") as tangential to the research questions. The final coding guide consisted of 10 categories with researcher-authored descriptions.

Classification was performed using qualcode.ai's dual-rater methodology, where two independent models ([OpenAI model name] and [Anthropic model name]) coded each response. Inter-rater agreement was substantial (Cohen's κ = 0.71; Krippendorff's α = 0.73; percent agreement = 79.8%). Disagreements (n = 436, 20.2%) were reviewed and reconciled by two authors (EF, GH).

Response data was processed in compliance with GDPR via EU data centers. No data was used for AI model training per enterprise API agreements.