5 Surprises in the Sources Cited by AI’s Halloween Answers

1. Walmart could be a shopping option without a Walmart citation

One question asked where a US shopper should compare Halloween costumes if the complete outfit needed to cost under $40. We saved eight Perplexity answers to that same question.

All eight offered Walmart as a purchase or comparison option. None included a Walmart-family source link. Target was also offered in all eight answers, and Target links appeared in all eight. [H2-02]

The difference was not limited to that question’s source lists: across all 556 saved Perplexity answers to the study’s 30 questions, there were no Walmart-family citations.

For a retailer, being offered and being linked are distinct outcomes. A report of cited websites alone would miss Walmart’s shopping role in the budget answers. The evidence does not establish which external page supplied the Walmart advice.

2. Morphsuits was cited in a comparison that chose no retailer

Another question asked how to compare the long-term value of a cheap costume with a more expensive reusable one. We saved 16 Perplexity answers to that one question.

All 16 linked to Morphsuits, a costume brand. None named it in the answer body or offered a named retailer choice. The answers discussed value rather than selecting a store. [H2-03]

One answer explained that the cheaper costume could be financially preferable for a single use unless comfort, appearance or reliability justified paying more. Its sources included a Morphsuits page about cheap versus expensive costumes.

The 16 answer texts differed, but all carried the same ten-URL source set, including five Morphsuits links. That is repeated source presence, not 16 independent discoveries of the brand’s pages or proof that a page caused the advice.

3. Target appeared across several seller-comparison questions

The panel included seven questions specifically about choosing sellers—for example, which online stores to compare for adult costumes at different budgets. These are seven different questions, not seven repeated answers.

For each system, Target links appeared in answers to six of those seven questions. That establishes breadth across the seller questions, rather than how often Target was named as a preferred store. [H2-01]

Target links also appeared beyond seller selection. ChatGPT cited Target in answers to five of the six questions about acquisition and value, which covered decisions such as whether to make, buy, rent or reuse an outfit.

The useful finding is that Target’s source presence accompanied several kinds of decision. The more detailed calculation across the full question panel is available in the methods below; it should not be mistaken for shopper share or a share of all citations.

4. Public agencies accompanied the costume-value question

ChatGPT answered the same cheap-versus-reusable costume question discussed in section two. We saved 48 ChatGPT answers to it.

All 48 cited both the Federal Trade Commission and the Environmental Protection Agency. None selected a named retailer. [H2-04]

The answers discussed future use, care and conditions under which paying more might not make sense. In one example, adding cleaning costs reversed an initially attractive cost-per-wear calculation. Those numbers were assumptions inside an AI answer, not measured product economics.

The agencies belong in the source picture, but they are not retail competitors. Their links also do not independently validate every calculation in the answer.

5. Retailer links accompanied display planning without always producing a shopping list

For a first outdoor display, we asked how to compare inflatables, static yard props and animatronics. We saved 28 ChatGPT answers to that one question.

Home Depot and Lowe’s source families each appeared in 23 of the 28 answers. Only seven of the 28 explicitly offered named retailer comparisons. [H2-05]

Much of the advice concerned the display’s configuration and practical constraints. Retailer links accompanied that explanation even when the answer did not provide a buying destination.

“Source family” matters to the count: it includes the main domain and the subdomains grouped with it by the study. Counting only each retailer’s principal host gives 21 of 28 answers instead of 23. Neither rule proves ownership of every linked page or that a particular page shaped the answer.

Read the shopping advice alongside the sources

Walmart’s shopping presence exceeded its source links. Morphsuits had links without a named buying role. Target appeared across different questions, while agencies and home-improvement retailers accompanied explanatory advice.

Those relationships are useful to a brand team precisely because a source list and a shopping list answer different questions.