Codebook

This codebook documents the survey instrument used in the Trust in AI Commerce Report v1. Every variable in the published dataset is paired here with its exact question wording, response options, value labels, universe rules, and derivation logic where applicable.

A reader who has both this codebook and the dataset can reproduce every statistic in the published report.

Study Summary

For full methodology — sampling design, screener logic, weighting, integrity flags, and limitations — see the Methodology entity or §7 of the published report.

File Structure

The published dataset.csv contains:

• 3,638 rows — all dispositions retained for screen-out auditability. Filter to Status = "Complete" to reproduce the n=1,463 analysis basis.
• 146 columns — the full Alchemer export minus 7 columns stripped for privacy (see below).

Privacy Strips

Per Product.ai Research's data-publication standard, the following 7 columns were removed before public release. All other 146 columns are published as Alchemer exported them.

How to Read This Codebook

Question wording in the column headers is exactly as shown to respondents in the Alchemer instrument. Each variable entry below documents:

• Column header — the CSV column name (matches what's shown to respondents, abbreviated where useful)
• Variable type — single-select, multi-select, ordinal scale, free text, derived
• Response set — full list of value labels
• Universe — which respondents saw the question (skip-logic notes)
• Notes — derivation rules and reporting caveats

Conventions

Matrix questions (e.g., the 18-category Purchase Research Confidence Index) appear as one column per row of the matrix, with the row label prepended to the question stem. The cell value is the response option chosen for that row.

Multi-select questions appear as one column per selectable option. The cell value is the option label if selected, blank otherwise.

Numeric values in scale columns (2, 3, 5, 6) are unlabeled mid-points on a 7-point Likert with verbal anchors at 1, 4, and 7.

Mixed-format encoding correction (apply before computing means)

The raw Alchemer export mixes text labels and numeric values on the same column for three endpoint-anchored scale questions: PC-1 (last-purchase confidence, 0–10 scale), AT-3 (AI satisfaction, 1–7 scale), and AT-7 (review trust, 1–7 scale). Endpoint and midpoint positions carry text labels; intermediate positions carry numeric codes. A naive mean computation that does not remap the text labels will silently drop the highest-frequency endpoint clusters from the average and understate the result. The §7.9 Limitations note in the report documents the corrected published values; the remap rules used to compute those corrected means are restated below for replicators working directly from the raw dataset.

PC-1 — "For that purchase, how confident were you that you found the best product for your needs before completing the purchase?" (0–10 scale)

Apply this mapping before averaging:

Compute the mean across the full non-missing column. Correct value: 6.95 / 10 (n = 1,453).

AT-3 — "How satisfied were you with the AI assistant's help?" (1–7 scale; AI users only)

Compute the mean across all AI users (Q12 = "Yes"). Correct value: 5.46 / 7 (n = 623).

AT-7 — "How confident are you that online product reviews accurately reflect product quality?" (1–7 scale; full sample)

The AT-7 question was shown to all respondents (n = 1,462 non-missing), not filtered to review users. The report's cited mean uses the actual-review-user denominator — respondents who selected "Online customer reviews" in PC-4 (n = 728) — and produces a corrected value of 4.92 / 7. The full-sample mean is 4.66 / 7 (n = 1,462) if a broader denominator is required.

Implementation note for replicators

In Python, the canonical remap is:

REMAP_0_10 = {"Not at all confident": 0, "Neutral": 5, "Extremely confident": 10}
REMAP_1_7  = {"Not at all confident": 1, "Not at all satisfied": 1, "Neutral": 4,
              "Extremely confident": 7, "Extremely satisfied": 7}

def to_num(v, remap):
    if v is None: return None
    if isinstance(v, (int, float)): return v
    return remap.get(str(v).strip())

Apply REMAP_0_10 to the PC-1 column; apply REMAP_1_7 to AT-3 and AT-7. Filter the dataset to Status == "Complete" for the 1,463-row analysis basis before averaging.

A. Administrative & Metadata

Columns Alchemer auto-generates for every response. These are present in the published CSV in their original form (except the 7 privacy strips noted above).

B. Screening — Eligibility & Anchor (Q1)

Block code: PC (Purchase Confidence)

Screener rule. Respondents who selected Under 18 were disqualified out of the survey at this point. The complete sample is 18+.

Note on encoding. The age bands as exported use inconsistent dash characters (some en-dash –, some hyphen -, and one band reads 44-54 rather than 45-54). These are preserved as exported. Analysts deriving age bands should treat 35-44, 44-54, 55-64, 65 or older as the four upper bands; the 44-54 label is interpreted as ages 45–54 (one-year overlap with the prior band reflects the exported labeling, not the underlying assignment).

C. Shopping Mode (Q2)

Screener rule. Respondents who selected I shop only in physical stores were disqualified — the panel anchor is online shoppers.

D. Category Purchases — Past 12 Months (Q3)

Multi-select. 18-category × selected/not-selected. Cell value is the category label when selected, blank when not.

Question stem. "Which of the following have you purchased for yourself or your household in the past 12 months? (Select all that apply)"

Universe. All respondents who passed screening (Q1 + Q2).

E. Recent ≥$50 Purchase Context (Q4–Q6)

F. Research Tools Used for That Purchase (Q7)

Multi-select. Cell value is the option label when selected, blank when not.

Question stem. "Which of the following did you use to research that purchase? (Select all that apply)"

G. Post-Purchase Feelings (Q8)

Multi-select. Cell value is the option label when selected, blank when not.

Question stem. "After completing that purchase, which of the following best describes how you felt? (Select all that apply)"

H. Purchase Research Confidence Index — PRCI (Q9)

18-category Likert matrix. Each column is one category; cell value is one of the 7-point scale labels.

Question stem. "In general, how confident are you in your ability to find the best product when shopping online in each of the following categories?"

Scale labels. Not at all confident (1), 2, 3, Neutral (4), 5, 6, Extremely confident (7).

I. Retailer-Trust Statement (Q10)

Question stem. "How much do you agree or disagree with the following statement: …"

Used to compute net retailer trust (agree+strongly agree minus disagree+strongly disagree).

J. Information Trust Hierarchy — IH (Q11)

7-source Likert matrix. Each column is one information source; cell value is one of the 7-point scale labels.

Question stem. "When researching a product you may buy online, how much do you trust information from each of the following sources?"

Scale labels. Do not trust at all (1), 2, 3, Neutral (4), 5, 6, Trust completely (7).

K. AI Usage Block (Q12–Q14)

This block produces two of the headline statistics in the report:

• 43% AI usage in past 90 days (Q12 = Yes ÷ full sample) — anchor: #stat-ai-usage-90-day
• 86% verification rate (Q14 ∈ {Yes-always, Yes-sometimes} ÷ AI-user subsample) — anchor: #stat-ai-verification-rate

L. AI Trust Index — ATI (Q15)

18-category Likert matrix. Each column is one category; cell value is one of the 7-point scale labels.

Question stem. "How much do you trust AI-generated product recommendations in each of the following categories?"

Scale labels. Do not trust at all (1), 2, 3, Neutral (4), 5, 6, Trust completely (7).

M. AI Autonomy Threshold — AT (Q16)

Headline statistic anchor. #stat-ai-autonomy-threshold — 42% selected $0 — I would always verify.

N. Review Trust (Q17)

O. Promo Code Block — Savings Behavior (Q18–Q35)

This block belongs to the SC (Savings Behavior) instrument block and powers SimplyCodes-side findings in the Checkout Gap industry report. It is documented here for codebook completeness; PAI-side Trust Report findings do not anchor on these variables.

O.1 Promo code search behavior (Q18)

O.2 Promo-source trust ranking (Q19) — 7-source ranking matrix

Question stem: "When you do search for a promo code, where do you trust finding one the most? Rank the following choices in terms of trust: 1st - most trust; 7th - least trust"

O.3 Promo-code reliability sentiment (Q20)

O.4 Savings-tools trust matrix (Q21) — 5-tool Likert matrix

Question stem: "How much do you trust each of the following savings tools to actually save you money?"

O.5 Recent promo experience (Q22–Q23)

O.6 Promo-failure mode (Q24) — multi-select

Question stem: "Thinking about the most recent time a promo code did not work, what happened? (Select all that apply)"

O.7 Post-failure behavior (Q25) — multi-select

Question stem: "When that code did not work, what did you do next? (Select all that apply)"

O.8 Promo-experience consequences (Q26–Q32)

P. Demographics (Q33–Q35)

Q. Completion Flag

Derived Variables & Composite Scores

The report references several composite measures constructed from the raw variables above.

PRCI Composite (Purchase Research Confidence Index)

Definition. Mean of 18-category PRCI scale responses (cols 61–78). Per-respondent: convert each Likert label to 1–7, take simple average across the 18 categories. Per-category: take mean across all respondents in the relevant universe.

Anchors. #stat-prci-overall (full-sample average), #stat-prci-by-category-range (per-category min/max).

ATI Composite (AI Trust Index)

Definition. Mean of 18-category ATI scale responses (cols 90–107). Same construction as PRCI.

Anchors. #stat-ai-trust-by-category-range.

IH Ranking (Information Trust Hierarchy)

Definition. Per-source mean trust score across all completes (cols 80–86). Ranked descending.

Anchors. #stat-trust-hierarchy-ranking.

Net Retailer Trust

Definition. Percentage of completes who agreed or strongly agreed with the retailer-trust statement (col 79) MINUS percentage who disagreed or strongly disagreed. Neutrals excluded.

Anchors. #stat-net-retailer-trust.

AT Bands (AI Autonomy Threshold)

Definition. Categorical bands on col 108 response. Reported "wouldn't trust over $X" calculations sum bands above each threshold.

Anchors. #stat-ai-autonomy-threshold.

Verification Rate

Definition. Among AI users (col 87 = Yes; n=623), the percentage who answered Yes - always OR Yes - sometimes on col 89.

Anchors. #stat-ai-verification-rate — 86%.

Canonical Statistics → Variable Map

Every statistic anchored in the published report maps to a specific variable or composite. This table aligns the report's #stat-* anchors to the codebook columns that drove them.

For the full Glass Box footnote format that anchors each stat back to its denominator, see §7 of the published report or the productai:glassBoxFootnoteFormat block in the methodology entity.

Release Notes

• v1.0 (2026-06-23) — Initial publication alongside the Trust in AI Commerce Report v1.

Contact

Questions about specific variables, derivation rules, or replication: research@product.ai.