Made in USA and Country-of-Origin Claims: Testing What Consumers Actually Infer

An unqualified "Made in USA" claim in the United States means something specific and demanding: final assembly or processing occurs in the US, all significant processing occurs in the US, and all or virtually all ingredients or components are made and sourced in the US. That is a supply chain question your operations team answers with documentation.

The question your operations team cannot answer is the one that generates most of the risk: what do customers infer about origin from everything else on the package? A flag motif, a brand name containing "American," a "Designed in California" line, or a qualified claim like "Assembled in USA" all create origin impressions that no sourcing document addresses. Those impressions are measurable, and measuring them is a research job.

The rule, precisely

The Made in USA Labeling Rule, 16 CFR Part 323, took effect in 2021 under authority of 15 U.S.C. 45a. Its structure is worth knowing exactly, because the definitions are broader than most teams assume.

Section 323.1(a) defines "Made in the United States" as any unqualified representation, express or implied, that a product or service - or a specified component of it - is of US origin. The rule lists "made," "manufactured," "built," "produced," "created," or "crafted" in the United States or in America, and then adds "or any other unqualified U.S.-origin claim."

Two words in that definition do the heavy lifting:

• Implied. The rule does not require the words "Made in USA" to appear. An implied representation counts, which is precisely what a flag, a state outline, or a heritage story can create.
• Unqualified. A qualified claim - one that discloses the extent of US content - falls outside the rule's prohibition, but only if the qualification actually communicates the limitation. Whether it does is an empirical question.

Section 323.2 makes it an unfair or deceptive act to label a product Made in the United States unless three conditions hold cumulatively: final assembly or processing occurs in the US, all significant processing occurs in the US, and all or virtually all ingredients or components are made and sourced in the US. There is no partial credit. Section 323.3 extends the requirement to seals, marks, tags, or stamps in mail order catalogs and promotional material, which the rule defines broadly enough to cover material disseminated by electronic means.

Section 323.5 preserves state law that gives consumers greater protection, so state origin statutes stack on top rather than being displaced.

Why the stakes rose

Violations are treated as violations of a rule under Section 18 of the FTC Act, which makes them subject to civil penalties. In February 2025 the Commission's maximum civil penalty for such violations was adjusted to $53,088 per violation. The largest civil penalty in the history of the Labeling Rule remains the $3.175 million paid by Williams-Sonoma in April 2024 in connection with a prior Commission order.

Enforcement attention has increased since. The FTC sent warning letters on Made in USA compliance in July 2025, and in March 2026 an Executive Order directed the FTC to prioritise enforcement against unlawful Made in USA claims and to consider a rulemaking addressing online retail platforms listing third-party items that violate the standard. For teams selling through marketplaces, that last clause is the one to watch.

The gap research fills

Sourcing documentation answers whether your product qualifies. It says nothing about whether your packaging conveys a claim you did not intend to make. That gap is where the two evidence files diverge:

Cases are lost on the second one. A company can hold impeccable sourcing records for a qualified claim and still face a problem because the qualification was invisible next to a flag occupying a quarter of the front panel.

The origin inference ladder

Origin impressions form through several channels, ranked roughly by how often teams forget them:

1. Explicit claims. "Made in USA," "Product of USA." Everyone reviews these.
2. Qualified claims. "Assembled in USA from imported parts," "Made in USA of US and imported materials." Reviewed, but rarely tested for whether the qualification registers.
3. Process claims. "Designed in California," "Engineered in Detroit," "Crafted in Vermont." These describe one activity, and are frequently heard as describing all of them.
4. Symbolic cues. Flags, stars-and-stripes colourways, eagles, state silhouettes, "Est. 1947" heritage marks.
5. Nominal cues. Brand names containing American, USA, a state, or a city; domain names; a US address in the corner of the pack.
6. Contextual cues. Retailer shelf placement in a "Made in America" section, marketplace badges, or category conventions.

Rungs 3 through 6 are where unintended unqualified claims come from, and none of them appear in a copy review because none of them are copy.

Designing the origin perception study

The design principle is the same one that governs all claim comprehension work: never mention origin until the participant does. The moment you ask "where do you think this is made?" you have created the salience you were trying to measure.

Stage 1 - unaided impression. Show the pack or product page as it appears. Ask an open question about what kind of product this is and what they can tell about the company. Let the AI moderator probe naturally. Count how many participants raise origin spontaneously, and record exactly what cue they attribute it to. That spontaneous rate is your best measure of how loudly the pack speaks about origin.

Stage 2 - unaided attribution. Now ask directly: where is this product made? Take a free-text answer before offering options. Then ask what on the pack tells them that. This second question is the one that produces the actionable finding, because it names the element to change.

Stage 3 - qualified-claim comprehension. If you use a qualified claim, test whether it lands. Present the claim and ask what it means in their own words. The failure mode to look for is participants treating a qualification as decoration - reading "Assembled in USA from imported components" as simply confirming American origin, with the second half doing no work at all.

Stage 4 - cue isolation. Where you suspect a specific element, test variants with and without it. Removing a flag and re-running Stage 1 will tell you what that flag was worth in inference terms, and that number is what lets you argue the design change internally.

Mapping to structured questions

All six of Koji's structured question types have a role here, and mixing them is what turns impressions into evidence:

The open-ended stages are the ones that historically made this study expensive. Getting a usable spontaneous-mention rate requires enough participants to count, and getting the attribution requires a follow-up question tailored to whatever each person just said. A static form cannot do the second; a moderated study cannot afford the first. Koji's AI moderator does both at once - probing every participant's reasoning conversationally while the sample runs in parallel - which is what makes a properly sequenced origin study a two-day exercise rather than a two-month one.

Reading the output

The spontaneous mention rate is your headline. If a meaningful share of participants raise US origin without being asked, your pack is making an origin claim whether or not your copy does.

Report the misattribution rate against the truth. For a product that is assembled domestically from imported components, the number that matters is the proportion who believe it is wholly US-made. That is the population who would feel misled.

Treat "unsure" as a good outcome. For a genuinely mixed-origin product, participants who decline to guess are evidence the pack is not overclaiming. Teams sometimes read this as a branding failure; in origin research it is the target.

Test the qualification separately from the claim. A qualified claim that tests identically to the unqualified version has failed, and that is the specific finding that justifies redesigning the panel.

For the mechanics of reporting qualitative counts honestly, see qualitative research validity; if you need to attach a margin of error to the attribution stage, see the survey sample size guide.

Where this fits

Origin perception belongs in the same evidence programme as your other claims work. It shares its structure with green claims research - both are cases where a true statement can produce a false takeaway - and with advertising claim substantiation generally. If you are testing packaging anyway, fold the origin questions into packaging concept testing rather than fielding a separate study; the incremental cost is a few questions and the coverage gain is substantial.

Keep the results, the asset versions, and the dates in a research repository. A documented perception study showing you tested the pack and acted on the result is a materially better position than sourcing records alone, and it is the artifact nobody has when they need it.

Frequently asked questions

Does the Made in USA rule apply if we never write "Made in USA"?

Yes. Section 323.1(a) defines the covered claim as any unqualified representation, express or implied, that a product is of US origin, and expressly extends to "any other unqualified U.S.-origin claim." Flags, state imagery, brand names, and heritage cues can create an implied representation without the phrase appearing anywhere, which is exactly why perception testing is the only reliable way to know what your pack claims.

What does "all or virtually all" actually require?

Under Section 323.2, three conditions must all hold: final assembly or processing occurs in the United States, all significant processing occurs in the United States, and all or virtually all ingredients or components are made and sourced in the United States. The conditions are cumulative - meeting two is not partial compliance.

Can consumer research prove our product qualifies for the claim?

No, and this boundary matters. Whether the product meets the all-or-virtually-all standard is answered with sourcing and manufacturing documentation. Research answers the separate question of what origin customers infer from your packaging and marketing, which is where unintended implied claims arise.

Are qualified claims like "Assembled in USA" safe?

They fall outside the rule's prohibition on unqualified claims, but only if the qualification genuinely communicates the limitation to the people reading it. The common failure is a qualification that participants treat as decoration - if your qualified claim tests the same as an unqualified one, it is not doing its job.

How many participants do we need for an origin study?

For surfacing which cues drive inference and why, 30 to 50 probed conversations per pack variant is generally enough. If you plan to report a misattribution rate as a headline number or compare variants statistically, size the attribution stage as a proportion estimate and report a margin of error.

What changed in enforcement recently?

Civil penalties for Labeling Rule violations were adjusted to $53,088 per violation in February 2025, the FTC issued Made in USA warning letters in July 2025, and a March 2026 Executive Order directed the Commission to prioritise enforcement and consider a rulemaking covering online retail platforms listing non-compliant third-party items. Marketplace sellers should treat platform listings as in scope.

Related Resources

• Structured Questions Guide - the six question types and when to use each
• Green Claims Research - the closest analogue, where true facts create false takeaways
• Advertising Claim Substantiation - designing survey research that backs a marketing claim
• Packaging Concept Testing - folding origin questions into pack testing you already run
• AI Research for CPG Brands - the wider consumer goods research programme
• Brand Perception Surveys - what customers really think about the brand
• Qualitative Research Validity - reporting counts honestly

---

Find out what your packaging claims. New Koji accounts include 10 credits - enough to run an unaided origin impression study with AI follow-up probing and learn which pack element is doing the talking.