What do shoppers ask AI over the holidays? Notes on a 200,000-prompt study

What did shoppers ask?

Holiday shoppers mostly asked AI to help them decide, not to buy. Profound, which sells AI visibility software, sampled 200,000 prompts with commercial intent from 20 to 26 December 2025 and classified them. Product research made up 46.9% of the sample and product comparisons another 22%.¹ The study, published on 9 July, did not state the geography of the prompts.

A small note on the arithmetic: the article gives the combined share as 68.2% in one place and 68.9% in another, and the two parts add to 68.9%.¹ Either way, more than two-thirds of holiday prompts came before a purchase decision.

Category demand was lopsided. Electronics, tech and gaming took 31% of the prompts, more than four times any other category, with travel next at 6.9% and health, personal care and fitness at 6.4%.¹ Within electronics, prompts were spread thin: PC components were the largest subcategory at 2.1%.¹

Christmas week 2025

Shoppers used AI to research and compare

46.9%of commercial prompts were product researchProfound, 20–26 Dec 2025 ¹

22%were product comparisonsProfound, 20–26 Dec 2025 ¹

31%were about electronics, tech and gamingProfound, 20–26 Dec 2025 ¹

Why it matters for Q4 revenue

It matters because AI-attributed holiday spending is already large. Salesforce estimated that AI and agents drove 20% of global online holiday sales in November and December 2025, worth $262 billion, based on the activity of more than 1.5 billion shoppers worldwide.³ "Drove" is Salesforce's own attribution measure, not a count of sales completed inside an assistant.

AI visitors also arrive closer to buying. Adobe Digital Insights found AI-sourced traffic to US retail sites rose 393% year on year in the first quarter of 2026, and that in March 2026 those visits converted 42% better than other traffic.² These are US figures for retail sites Adobe measures.

Our view: put those together with Profound's prompts and the pattern is clear. The shortlist forms inside the answer, during research. By the time a shopper reaches your site from an AI answer, the comparison has often already happened.

What drove the answers shoppers saw

Quality mattered more than price in what shoppers asked. In Profound's sample, 24.4% of prompts were driven by quality and performance, 14.7% by convenience and speed and 13.2% by price or deals.¹ Shoppers wanted to know whether a product was good before they asked whether it was cheap.

Online shopping dominated where a channel was stated: 28.3% of prompts carried online signals against 1.6% for in-store, while most stated no channel at all.¹ Among retailers, Amazon had the largest share of voice at 9.1% of responses.¹ Marketplace listings are part of your AI shelf whether you manage them closely or not.

Profound also compared the product offers that engines recommended most with the rest. Top offers were 188% more likely to have an FAQ section, 106% more likely to have video and 35% more likely to have customer reviews.¹ Top offers in ChatGPT were 777% more likely to carry its "Best Price" tag.¹ These features correlate with recommendation; the study does not show they cause it.

How far can you trust the study?

Trust its direction more than its decimals. The prompts come from Profound's own prompt-volume product, which it describes as anonymised, aggregated data from real users of ChatGPT, Claude, Gemini, Perplexity and DeepSeek. How that panel is recruited, and which countries it covers, is not stated.

The study covers one week, the one before Christmas, when last-minute behaviour peaks. It tells you less about October and November, when most gift research starts. The categories and intents are Profound's own classification, applied to a sample of a larger set.

Profound also sells the tools that measure this kind of visibility, and its recommendations point towards them. That is worth knowing, and it does not change the core finding, which matches what other retail data shows: shoppers use AI to research.

What a D2C brand can prepare in July

Prepare the pages and feeds engines read, because they take time to be crawled and reflected in answers. Work done in December arrives too late for the week that matters. A practical order runs from data to pages to checks.

A July-to-December plan for the AI shelf

1. July

   Run a baseline: the 30 product questions shoppers ask, on each engine, on several days.

2. August

   Fix product data: specifications as text, accurate prices, clean feeds.

3. September

   Add FAQs from support tickets and reviews; publish comparison pages.

4. October

   Recheck the baseline; fix marketplace listings and stock data.

5. November

   Freeze major page changes; keep prices, stock and delivery cut-offs current.

6. December

   Watch availability and delivery messages daily through Christmas week.

Lead each product page with what shoppers asked about: what it does, how well, compared with what. Write FAQs from real support questions and reviews, not from a keyword tool. If engines cannot read your product data at all, start with our playbook for e-commerce brands, which covers feeds, checkout and the sources engines trust.

Update pages when their facts change, not on a calendar. Holiday pages tempt teams to bump dates for freshness, which our note on freshness studies argues against. A current price and a correct delivery cut-off are the freshness that counts in December.

How to measure the result

Measure on a fixed prompt set, before and after, with enough runs to trust the change. At Sigzen AI, our AI answer monitor covers six engines (ChatGPT, Gemini, Perplexity, Claude, Copilot and Google AI), through official APIs where offered and a licensed AI-answer data provider otherwise. Audits run each prompt on three separate days.

Our view: pick prompts by intent, not by keyword. Shoppers in the Profound sample asked research and comparison questions, so most of your prompt set should too. Track a few "where to buy" prompts as well, because that is where stock and delivery errors show up first.