Human vs. AI Avatar Presenters in B2C Product Demonstrations: Trust, Engagement, and Purchase Intention

**Title of paper:** Human vs. AI Avatar Presenters in B2C Product Demonstrations: Trust, Engagement, and Purchase Intention **Authors:** - Victor Santos (ORCID: 0000-0003-3399-5414) — Coimbra Business School | ISCAC, Instituto Politécnico de Coimbra; REMIT, Universidade Portucalense - João M. S. Carvalho (ORCID: 0000-0003-0683-296X) — REMIT, Universidade Portucalense Infante D. Henrique - Sílvia Faria (ORCID: 0000-0002-7672-3972) — ISCAP, Instituto Politécnico do Porto; REMIT, Universidade Portucalense **Funding:** FCT — Fundação para a Ciência e a Tecnologia, grant UID/05105/2025 **Target journal:** Journal of Retailing and Consumer Services (Elsevier) **Corresponding author:** Prof. Victor Santos — [e-mail] **License:** Creative Commons Attribution 4.0 International (CC BY 4.0) **Version:** v1.0.0 (2026-05-25) --- ## Overview This Zenodo deposit contains all materials required to reproduce, audit, and extend the empirical findings reported in the manuscript "Human vs. AI Avatar Presenters in B2C Product Demonstrations: Trust, Engagement, and Purchase Intention" submitted to the Journal of Retailing and Consumer Services. Because the Zenodo web interface does not preserve folder hierarchy on upload, all files appear in a single flat list. The table below shows the logical grouping of each file. --- ## File index by logical group ### Group 1 — Analysis (PLS-SEM code and outputs) | File | Description | |------|-------------| | `appendix_b_synthetic_data.py` | Python 3.11 code generating the synthetic datasets (seed = 42). Matches Appendix B of the manuscript. | | `SmartPLS_Human_n145.csv` | Phase 2 — Human condition data, prepared for SmartPLS 4.0 import | | `SmartPLS_AI_n145_CORRECT.csv` | Phase 2 — AI Avatar condition data, prepared for SmartPLS 4.0 import | | `outer loadings.xlsx` | Full outer-loadings table (measurement model) for both phases | | `AI Avatar vs Human Comunicator.xlsx` | Complete PLS-SEM output: structural coefficients, p-values, R², f², bootstrapped CIs (replicates Tables 6, 7, 7B of the manuscript) | ### Group 2 — Dataset (empirical and synthetic data) | File | Description | |------|-------------| | `Pilot_Clean_Dataset_n145.xlsx` | **Anonymised** Phase 2 dataset (n = 145). PROLIFIC_PID and other direct identifiers removed; timestamps reduced to minute precision. | | `synthetic_human_condition_N500.xlsx` | Phase 1 Human-condition synthetic dataset (N = 500). Generated with seed = 42. | | `Synthetic_Sample_Parameters_v14_N500.xlsx` | Generation specifications: demographic distributions calibrated to Eurostat EU statistics; correlation targets per construct. | | `psychometric_summary.xlsx` | Summary table of psychometric indices (α, CR, AVE, HTMT) for both phases. Companion to manuscript Sections 5.1 and 5.4.1. | | `anonymise_xlsx_helper.py` | Anonymisation script kept for transparency (documents the procedure applied to obtain `Pilot_Clean_Dataset_n145.xlsx`). | ### Group 3 — Questionnaire (SoSci Survey instrument) | File | Description | |------|-------------| | `codebook_AI_Avatar_Pilot_2026_2026-05-17_18-52.xlsx` | **Codebook** — official SoSci export listing every variable, response code, and label. Source for Appendix G of the manuscript. | | `variables_AI_Avatar_Pilot_2026_2026-05-17_18-52.csv` | **Variables Overview** — full variable list with technical metadata | | `values_AI_Avatar_Pilot_2026_2026-05-17_18-52.csv` | **Response Code Listing** — all valid response codes per variable | | `structure_AI_Avatar_Pilot_2026_2026-05-25_14-56.json` | **Questionnaire Structure (JSON)** — full topology of the questionnaire (pages, items, ordering) | ### Group 4 — Stimuli (videos and script) | File | Description | |------|-------------| | `youtube_urls.json` | 12 YouTube URLs (5 PT/FR/ES/DE/PL Human + 5 PT/FR/ES/DE/PL Avatar + 1 EN Human + 1 EN Avatar) | | `script_PT_canonical.txt` | Canonical Portuguese script (~250 words) deployed in the recordings | --- ## Live access (no download required) - **Interactive questionnaire (all 5 languages, click flag to enter):** https://www.soscisurvey.de/AI_Avatar_Pilot_2026/ - **Stimulus videos (12 unlisted YouTube clips):** See `youtube_urls.json` in this deposit --- ## Two-phase validation design - **Phase 1 (synthetic, N = 500):** demographically calibrated to Eurostat EU statistics for model identification and psychometric coherence. - **Phase 2 (empirical pilot, n = 145):** Prolific Academic, five EU countries (PT, ES, FR, DE, PL); data collected **15-20 May 2026**. Both phases analysed with PLS-SEM in SmartPLS 4.0. --- ## How to reproduce ### Phase 1 (synthetic) ```bash python appendix_b_synthetic_data.py ``` The script uses `seed = 42` and produces deterministic output. Import the resulting CSVs into SmartPLS 4.0 with the model specification documented in manuscript Section 4.5. ### Phase 2 (empirical) 1. Open SmartPLS 4.0 2. Import `SmartPLS_Human_n145.csv` and `SmartPLS_AI_n145_CORRECT.csv` 3. Apply the model specification (PA, AU, TR, EN, PI as second-order constructs) 4. Run multi-group analysis (MGA) with permutation testing (5000 resamples) Output should reproduce Tables 6, 7, and 7B of the manuscript and the content of Appendix E and F. --- ## Ethical and legal notes - Empirical data collected via Prolific Academic (15-20 May 2026) under GDPR-compliant informed-consent procedure. - All personal identifiers (PROLIFIC_PID, IP addresses, etc.) removed prior to deposit. Timestamps reduced to minute-level precision. - Stimulus videos produced by the authors using HeyGen (avatar visual synthesis) and ElevenLabs (voice synthesis). Voice and likeness rights for the human presenter cleared via written consent. - Eligibility filter applied during recruitment: at least one online purchase in the preceding three months. Compensation paid per Prolific minimum hourly rate. --- ## Pending additions (planned for v1.1.0) The following materials are not yet included in this v1.0.0 release but will be added in a future versioned release at the same Zenodo concept-DOI: - **SoSci project XML export** — a temporary server-side issue at SoSci on the export date prevented its generation. The live SoSci URL above already provides full reviewer access to the deployed instrument in all five languages. - **Synthetic Avatar dataset** — regenerable on demand from `appendix_b_synthetic_data.py` (seed = 42). These omissions do not affect reproducibility of the published analyses, because: 1. The live SoSci URL provides direct inspection of the instrument in all 5 languages. 2. The codebook XLSX, variables CSV, values CSV, and structure JSON jointly document the full questionnaire technical specification. 3. The synthetic Avatar dataset can be regenerated deterministically from the included Python code. --- ## Manuscript-to-SoSci variable mapping For quick reference, the construct-to-variable mapping used in the analyses: | Construct | SoSci codes (Human) | SoSci codes (Avatar) | Manuscript codes | |-----------|---------------------|----------------------|------------------| | Presenter Authority (PA) | P507_01–14 | P707_01–14 | PA1–PA14 | | Presenter Authenticity (AU) | P502_01–14 | P702_01–14 | AU1–AU14 | | Consumer Trust (TR) | P503_01–12 | P703_01–12 | TR1–TR12 | | Consumer Engagement (EN) | P504_01–10 | P704_01–10 | EN1–EN10 | | Purchase Intention (PI) | P505_01–05 | P705_01–05 | PI1–PI5 | | Attention checks | P506_01 | P706_01 | AC_A, AC_B | | Identification | P801, P802, P803 | — | ID01, ID02, ID03 | Demographics: P202–P209 (DM01–DM08). Consent and eligibility: P103 (CO01), P201 (ELEGIBI). See Appendix G of the manuscript for the full bilingual (EN + PT) item wording. --- ## Citation Santos, V., Carvalho, J. M. S., & Faria, S. (2026). Replication package for "Human vs. AI Avatar Presenters in B2C Product Demonstrations: Trust, Engagement, and Purchase Intention" [Data set]. Zenodo. https://doi.org/10.5281/zenodo.XXXXXXX (DOI assigned upon publication of this deposit.) --- ## Contact Prof. Victor Santos ([e-mail]) --- **Funding:** This work was financially supported by Fundação para a Ciência e a Tecnologia (FCT, Portugal) under grant UID/05105/2025, attributed to REMIT — Research on Economics, Management and Information Technologies, Universidade Portucalense Infante D. Henrique. FCT funder DOI: 10.13039/501100001871.