Made in the USA — Consumer Study, April 2025

1 · Sponsorship, funding, and who conducted the research (§III.A)

The study was sponsored, funded, designed, fielded, and analyzed by QuestionPunk (app.questionpunk.com), an AI-interview survey platform, as a self-funded research and platform-demonstration project. No external funder, client, political party, campaign, trade association, or advocacy organization sponsored, financed, or reviewed the research. Findings were originally published on the company's LinkedIn page ("Made in the USA: Consumers Still Willing to Pay More," April 2025) and summarized on questionpunk.com.

2 · Purpose of the study

Dual purpose, disclosed to enable appropriate interpretation: (a) substantive — to measure U.S. consumer attitudes toward the "Made in the USA" label during the April 2025 tariff debate: emotional response, category salience, purchase behavior, willingness-to-pay premium, perceived company authenticity, and five-year importance trend; (b) methodological/commercial — to demonstrate the QuestionPunk platform's combination of closed-ended questions with AI-generated personalized follow-up interviews.

3 · Population under study (§III.A)

Adults aged 18 and older residing in the United States. Coverage is limited to members of the Prolific online research panel who elected to take the study — an opted-in, internet-using population that differs from the general adult population in known ways (see §10 and §14).

4 · Sample design and recruitment (§III.A)

Nonprobability, opt-in online access panel with census-matched quotas. Respondents were recruited from the Prolific panel (study id 67fd87af7700bd56b2d4cbb8, study name "Made in the USA," created April 14 and launched April 24, 2025 at 17:20 UTC); every session carried a Prolific participant, study, and session identifier. The Prolific study record, retrieved via the Prolific API in July 2026, confirms the study ran as Prolific's U.S. representative sample product (US_REP_SAMPLE), which allocates places by interlocked age × sex × ethnicity quotas matched to U.S. Census distributions, with a U.S. current-residence screener; 300 places were offered and 299 taken. The original publication's phrase "census-representative by age, gender, and region" overstated the product's scope on one dimension — Prolific's representative sample stratifies age, sex, and ethnicity, not region (see §13, Erratum 4). Census-matched quotas on a volunteer panel do not make a probability sample; the achieved composition (including a 55–64 overweight and 65+ shortfall at the top of Prolific's age strata) is tabled in §10 and governs interpretation.

5 · Incentives (§III.B, disclosed proactively)

Respondents were paid through Prolific upon entering the completion code shown on the survey's final screen. Per the Prolific study record (retrieved July 2026): reward US$1.60 per completed submission, a realized average of US$11.12 per hour based on Prolific's measured average completion time of 8.6 minutes (platform-measured median 7.8 minutes); advertised completion estimate 8 minutes, maximum allowed time 39 minutes. Total study cost including participant payments and Prolific platform fees: US$685.71, or US$2.32 per completed interview.

6 · Mode and instrument (§III.A)

Single web mode on the QuestionPunk platform, presented as a chat-style interface: a welcome message; 8 closed attitude/behavior items (6 single-choice, 1 nine-row agree/disagree matrix, 1 select-up-to-3 with write-in); an AI-moderated follow-up block of 3 personalized questions (§11); and 5 demographic items (age, gender, religion, household income, political views), followed by the Prolific completion code. Question order was fixed; response options were presented in the fixed order shown in Appendix A. Full verbatim wording in Appendix A. English only.

7 · Dates and locations of data collection (§III.A)

April 24, 2025. All 295 completed interviews were collected between 17:22 and 23:23 UTC on April 24 (a single evening, U.S. time). A further 37 sessions that never completed trickled in through April 28. Respondents participated remotely on their own devices. Network-level geolocation was not yet captured by the platform at that time; U.S. residency relied on Prolific's panel verification.

8 · Sample sizes, dispositions, and outcome rates (§III.A)

Because this is a nonprobability opt-in panel sample, AAPOR response rates (RR1–RR6) do not apply; the platform-side disposition funnel is reported instead. (Prolific-side dispositions — how many panelists viewed or declined the listing — were not retained.)

All published estimates are based on the 295 completed interviews; every closed item has a full base of n = 295 (all items were required). Median completion time 7.8 minutes (IQR 6.1–10.5; maximum 31). All 295 completers answered the AI follow-up block (median 3 exchanges; median 354 characters of written answers per respondent).

9 · Precision of estimates (§III.A)

This is a nonprobability sample; a margin of sampling error cannot be calculated and is not reported. Percentages describe the achieved sample of 295 respondents, not precise population point estimates. Subgroup figures cited in the report rest on bases as small as n = 9 (respondents who never check origin labels), n = 16 (the >20%-premium tier), and n = 30 (adults 65+, the group behind the report's 90% figure); anything under n ≈ 50 should be read as a pattern, not a point estimate. No external benchmark of the same construct was available for validation; the achieved-sample demographics are benchmarked against the U.S. Census in §10.

10 · Weighting and achieved sample composition (§III.A)

Published estimates are unweighted. No weighting, imputation, or model-based adjustment was applied in either the 2025 publication or the 2026 re-audit. Achieved composition of the 295 completers, with approximate 2024 Census Bureau distributions of the adult population for reference:

The two most consequential deviations for this topic: the sample under-represents seniors (the group with the highest measured willingness to pay, so headline willingness figures are, if anything, conservative) and under-represents self-described moderates while over-representing both ideological wings.

11 · Artificial-intelligence systems used (AAPOR AI-disclosure guidance)

a. AI follow-up interviewer (2025 fieldwork). After the closed items, the platform generated three personalized follow-up questions per respondent, each conditioned on the respondent's own prior answers. The generation used an OpenAI GPT-4-class chat model called via API with temperature 0. The exact model identifier was set by deployment environment configuration and the April 2025 environment record was not retained; the platform's configured model set in that period comprised gpt-4o, gpt-4o-mini, and gpt-4.1-mini. This is disclosed as a records limitation rather than resolved by guess.

b. Interviewer standing instruction (verbatim, the complete prompt-of-record for the follow-up block):

"Simulate a skilled focus group moderator by analyzing each respondent's full set of answers to the standard survey questions about 'Made in the USA' products. Use this analysis to identify a profile or pattern in their views, and then ask thoughtful, personalized follow-up questions that dig deeper into why they feel that way — uncovering their core motivations, trade-offs, or values that aren't visible in the original responses."

c. 2026 re-analysis. The July 2026 re-audit and the thematic coding of all 295 AI-interview transcripts were performed by Anthropic Claude models (Claude 5 family) under the direction and review of the research lead: closed-ended tabulations were recomputed deterministically in Python from raw records; interview coding used one full coding pass (Claude Sonnet 5) over all 295 interviews against the fixed topic/stance codebook published with the archive, with an independent second pass on a 20% systematic sample (n = 59; every 5th interview): stance agreement 90% (Krippendorff's α = 0.83); primary-topic agreement 80% (α = 0.74). The topic distribution is accordingly presented at whole-percent precision and labeled as satisfying only the tentative-conclusion reliability threshold (0.667 ≤ α < 0.80); the stance distribution meets the 0.80 standard. Quotations in the revised report are lightly copy-edited for spelling and punctuation only (wording never changed, omissions marked with ellipses); each was programmatically matched to its stored source answer and the edit diff reviewed.

d. AAPOR AI-checklist items (per the Responsible AI Integration in Survey Research task-force report, May 2026):

12 · Data quality, editing, and exclusions (§III.A)

• Deduplication: zero duplicate Prolific participant IDs among the 295 completed interviews. Fifteen participant IDs appear more than once among started sessions (re-entries that never yielded a second complete); they do not affect published estimates.
• Fraud screening: the platform's automated fraud-scoring and geolocation systems post-date this study; no respondent-level fraud scores exist. Identity assurance relied on Prolific's panel verification.
• Exclusions: none. No responses were removed or edited. Zero synthetic, preview, or test responses are included in the analytic file (the survey's three preview sessions ran under a separate draft survey id).
• Partials: the 37 abandoned sessions are excluded from all published estimates (original and revised); including their answered items changes no published figure by more than 1 point.
• Write-ins: the category, gender, religion, and politics items accepted write-ins; write-ins were retained verbatim and reported in an "other" line where applicable, not recoded.

13 · Errata and the 2026 re-audit (§III.A)

In July 2026 the complete response data were re-extracted from the production database and every published figure was recomputed from raw records. Results:

• Reproduced exactly: the four statistics of the original LinkedIn article — 21% unwilling to pay any premium (21.0%); 53% open to paying up to 10% more (exactly 53.6%; the published "53%" is the sum of the independently rounded components, 32% + 21% — a rounding convention noted here since 53.6% rounds to 54%); 87% of those willing to pay more than 10% viewed the label's promoters as authentic (46/53 = 86.8%); 44% of those unwilling to pay a premium viewed them as authentic (27/62 = 43.5%).
• Erratum 1 — sample description. The original publication reported "300 U.S. adults." The correct figures are 332 sessions started, 295 completed; all percentages were and are computed on the 295 completes.
• Erratum 2 — a garbled syndication. The questionpunk.com case-studies summary rendered the willingness figure as "53% Willing to Pay 10%+ More." The correct statement (as in the original article) is "53% willing to pay up to 10% more"; the share willing to pay more than 10% is 17.9%. The summary conflated the percentage with the coincidentally equal count of 11%+ payers (n = 53).
• Erratum 3 — a benchmark citation. The original article cited Budget Lab at Yale projections of "46% short-term … electrical equipment price increases." We could not verify that figure in the April 15, 2025 Budget Lab report (its short-run figures include apparel ≈65% and textiles ≈45%; later Budget Lab vintages put electrical equipment/electronics near 17–18%). The revised report cites the apparel and textiles figures only.
• Erratum 4 — recruitment description. "Census-representative by age, gender, and region" mislabeled one stratification dimension: the Prolific study record confirms the U.S. representative-sample product, which stratifies age, sex, and ethnicity — not region. See §4 and the achieved-composition table in §10.

14 · Limitations (§III.A)

(1) Nonprobability opt-in panel; no design-based inference; professional panelists may differ systematically from the general public. (2) Achieved sample over-represents ages 55–64, under-represents 65+ and moderates (§10); estimates are unweighted. (3) Single-day snapshot during a fast-moving tariff news cycle; results are time-stamped to April 24, 2025. (4) Willingness-to-pay is self-reported and hypothetical; stated premiums typically exceed revealed ones. (5) Agree/disagree matrix items are subject to acquiescence bias — agreement was high on logically opposed statements; the report interprets only relative ordering. (6) The AI moderator occasionally mischaracterized respondents' positions in its question framing (disclosed in §11.d), which may color the tone of some follow-up answers. (7) No geolocation records; U.S. residency rests on panel verification. (8) 2026 thematic coding is single-vendor AI with a 20% reliability sample and no human second coder; topic-level percentages carry ±several points of coding noise.

15 · Availability of materials and contact

The full instrument (Appendix A), the revised analytical report, and this disclosure are published together. The item-level dataset (closed-ended responses with demographics, stripped of Prolific identifiers) and the 2026 coding data (codebook, per-interview codes, reliability sample) are archived and available to bona-fide researchers on request. Contact: QuestionPunk, via questionpunk.com.

Appendix A — Complete instrument, verbatim

Items are shown in respondent order, with the platform's item type. All choice items were required; options were presented in the fixed order listed. Item 1 in the platform's internal numbering was deleted before fielding; numbering below follows respondent order.

Item 0 · message

Thank you for participating in this short survey. We're interested in your thoughts on where products are made and how that influences your shopping decisions. Your responses will help us better understand how consumers feel about products labeled "Made in the USA" today.

This survey will take about 5 minutes. Your answers are completely anonymous and confidential.

Item 1 · single choice

When you see a product labeled "Made in the USA," what is your immediate reaction?
Very negative · Somewhat negative · Neutral · Somewhat positive · Very positive

Item 2 · matrix (5-point agree/disagree)

How strongly do you agree or disagree with the following statements? (Strongly Disagree · Disagree · Neutral · Agree · Strongly Agree)

• I try to support American-made products whenever possible
• "Made in the USA" is a sign of higher quality
• I trust American-made products more than those made elsewhere
• The label "Made in the USA" is mostly just marketing
• Buying American made products is a good way to support the economy
• American-made products are usually too expensive
• Buying imported products contributes to global economic development
• It's not practical or necessary for all products to be made in the U.S.
• It is important to support American jobs whenever possible

Item 3 · multiple selection (up to 3) with write-in

In which of the following product categories does "Made in the USA" matter most to you? (Select up to 3)
Clothing · Food and groceries · Cars & trucks · Electronics · Furniture · Tools & home improvement · Health & beauty · [write-in]

Item 4 · single choice

Have you bought a product specifically because it was labeled "Made in the USA"?
Yes · No · Not sure

Item 5 · single choice

How much more would you be willing to pay for a product made in the USA vs. a similar imported product?
I wouldn't pay more · Up to 5% more · 6-10% more · 11-20% more · More than 20% · Not sure

Item 6 · single choice

Compared to five years ago, how important is "Made in the USA" to you today?
Much less important · Somewhat less important · About the same · Somewhat more important · Much more important

Item 7 · single choice

In your opinion, how authentic are most companies that promote "Made in the USA"?
Very inauthentic · Somewhat inauthentic · Neither authentic nor inauthentic · Somewhat authentic · Very authentic

Item 8 · single choice

How often do you notice or check where products are made before buying them?
Never · Rarely · Sometimes · Often · Always

Item 9 · AI-moderated follow-up interview (3 questions)

Personalized follow-up questions generated per respondent; the moderator's standing instruction is reproduced verbatim in §11.b.

Items 10–14 · demographics (single choice)

Age: 18-24 · 25-34 · 35-44 · 45-54 · 55-64 · 65 or older · Prefer not to say
Gender: Male · Female · Non-binary/third gender · Prefer not to say · [write-in]
Religious affiliation: Christian (any denomination) · Jewish · Muslim · Hindu · Buddhist · Atheist · Agnostic · Spiritual but not religious · Prefer not to say · [write-in]
Household income: Less than $25,000 · $25,000-$49,999 · $50,000-$74,999 · $75,000-$99,999 · $100,000-$149,999 · $150,000-$199,999 · $200,000 or more · Prefer not to say
Political views: Very liberal · Somewhat liberal · Moderate/centrist · Somewhat conservative · Very conservative · Prefer not to say · [write-in]

Item 15 · message

Thanks for completing the survey! Please return to Prolific and enter this code to confirm your participation: [code]