Who's Letting the Bots Shop? AI Shopping Trust by the Numbers

Agentic commerce is no longer a demo reel. Shoppers are already letting AI research products, compare prices, and in some cases buy on their behalf, even as the stores they shop at have not decided whether to let the bots in. To map both sides of that gap, we surveyed 1,000 Americans about what they would hand to an AI shopping agent and audited the robots.txt files of 500 major retailers plus 55 consumer goods brand sites to see who blocks the crawlers behind tools like ChatGPT and Claude. For operations leaders at consumer product companies, the findings are an early read on where agentic demand is heading and how ready your storefront is to meet it.

Key Takeaways

• 77% of shoppers are willing to let an AI agent shop in some capacity (researching, comparing products, or actually buying something).
• Gen Z is most likely to let AI complete a purchase, at 11%.
• Nearly 1 in 5 consumers (18%) have switched brand loyalty due to AI recommendations.
• Among shoppers who'd let AI spend without approval, more than 1 in 10 (11%) would let it spend $100 or more in a single purchase.
• 1 in 10 CPG brands block AI shopping crawlers.

Letting an AI agent help you shop and letting it spend your money turned out to be two very different levels of trust.

Most shoppers were open to some version of AI-assisted buying. More than three-quarters (77%) said they would let an AI agent shop for them in some capacity, whether that meant researching, comparing products, or actually buying something. Full autonomy was the holdout: only 6% said they would let an agent complete purchases entirely on its own.

What people would hand over without being asked clustered around low-risk tasks. Monitoring for price drops led at 63%, with comparing products close behind at 62%, while completing the purchase and paying sat at the bottom of the list. Shoppers wanted AI watching the deal, but not holding the wallet.

Trust also tracked with how routine and replaceable a product felt. Groceries and household goods topped the list of categories shoppers would let an agent buy in at 34%, ahead of personal care at 29% and pet supplies at 23%. Commodity restocks earned the most rope, and identity-driven purchases earned the least.

Age and dollar limits drew the clearest boundaries. Gen Z was the most willing generation to let AI complete a purchase, at 11%. And among shoppers who would let an agent spend without approval at all, 11% would let it go to $100 or more in a single purchase.

AI recommendations are already nudging shoppers away from the brands they used to grab on autopilot. That openness came with conditions, and a short list of worries that kept a human firmly in the loop.

Brand loyalty proved more negotiable than many consumer goods teams would like. Most shoppers (79%) were open to letting an AI agent swap them to a cheaper brand to save money, and that figure rose to 81% among Gen Z. The habit is already forming, with 18% saying they had switched brands because of an AI recommendation, led by Gen Z at 26%.

The hesitation centered on money and control. More than half of shoppers worried that an agent would quietly spend more than they would (54%) or be secretly steered by brands that pay for placement (51%). Those concerns help explain why so few were ready to hand over checkout completely.

The upside was that a rocky start did not end the relationship. Among shoppers who had tried an AI agent, 83% said they would use one again, and 81% were satisfied with its picks. Get the basics right, and most people come back.

As AI shopping assistants move from novelty to mainstream, a key question is playing out in a file most shoppers never see: retailers' robots.txt. We analyzed the public robots.txt directives of 500 major retailers and e-commerce platforms to find out how many are blocking the AI crawlers that power tools like ChatGPT, Claude, and Perplexity.

The front door was open at most stores, but not all. Roughly one-tenth of major retailers blocked the large language model crawlers behind today's shopping assistants, and the resistance concentrated among marketplaces. The companies with the most products to lose were often the most cautious about letting agents index them.

Some of the biggest names drew the hardest lines. Amazon was the most aggressive, fully banning the largest set of AI crawlers in the study, including GPTBot, ClaudeBot, PerplexityBot, and Google-Extended. Two categories pushed the rate well past average: marketplaces blocked at 23% and apparel at 21%, more than double the overall retail rate of about 10%.

The manufacturers behind those products were not far behind. Ten percent of consumer goods brands blocked AI shopping crawlers on their own sites. For brands betting on direct-to-consumer growth, that is a setting worth checking before an agent checks it for you.

The demand side is moving faster than the supply side is ready for. Shoppers, and Gen Z in particular, are already comfortable handing routine, commodity purchases to an agent, yet many storefronts and even some brand sites still treat AI crawlers as something to keep out.

For operations leaders at consumer product companies, that gap is the opportunity: clean product data, agent-readable storefronts, and clear restock logic are about to matter as much as shelf placement once did. The systems behind the storefront, from your consumer goods ERP to your product feeds, will decide whether an agent can actually find, compare, and reorder your products, and the brands that get there first will win the next cart.

Methodology

We surveyed 1,000 Americans to explore their online shopping habits and their attitudes toward AI shopping agents. All survey participants were adults aged 18 or older. Percentages may not total 100% due to rounding and multi-select questions.

Retailer sample

We analyzed the robots.txt files of 500 leading retailers and e-commerce platforms spanning multiple categories. Retail rates are calculated against the 364 sites that returned a readable robots.txt. For each site, we recorded whether it named and disallowed any of the major generative-AI crawlers, including:

• OpenAI's GPTBot, ChatGPT-User, and OAI-SearchBot
• Anthropic's ClaudeBot
• Google-Extended
• PerplexityBot
• Applebot-Extended

CPG sample

To address the consumer packaged goods (CPG) sector specifically, we also analysed 55 CPG brand sites. These were manufacturer-owned domains only (for example, coca-cola.com, tide.com), spanning food, beverage, household, personal care, and pet, and were kept distinct from the retailer sample because CPG brands are product manufacturers rather than stores. CPG figures are reported against the 41 CPG sites that returned a readable robots.txt. Category groups range from 16 to 60 readable sites.

Crawler classification

We distinguished these from general-purpose or non-LLM crawlers such as Huawei's PetalBot, Amazon's Amazonbot, and ByteDance's Bytespider, which sites commonly block for reasons unrelated to AI assistants; sites blocking only those were not counted as AI-blockers. A site was classified as blocking AI if it applied a full Disallow rule to a generative-AI crawler, and as restricting AI if it disallowed only specific sections.

Data collection

Data was collected on June 17, 2026. Since new AI crawlers emerge frequently, figures reflect the named agents active as of that date. Sites that did not return a readable robots.txt were excluded from rate calculations rather than counted as permissive.

Limitations

This study measured site-level crawler policy (robots.txt) and the homepage response only. A few limitations are worth noting:

• Scope of testing: We did not test cart, checkout, or product-detail pages, so the findings capture whether a retailer blocks AI crawlers at the front door, not whether an agent could complete a purchase.
• Checkout friction: Checkout friction, such as forced account creation, CAPTCHA at the cart, or payment-step bot walls, is a real form of agent-hostility we did not measure, and it could shift the picture.
• Stated policy versus enforcement: Robots.txt also states intent rather than enforcement: a site can publish an open robots.txt and still block AI crawlers through server-side bot detection, and the reverse is also true. We measured the stated policy, not what happens when a crawler reaches the server.

About DOSS

DOSS is the operations platform for consumer goods brands scaling $10M to $200M across DTC, retail, and wholesale. We unify inventory, orders, procurement, and finance data into one system of action, so your team stops reporting what happened in the past and starts driving the outcomes founders and VPs of Ops want next. Learn more at www.doss.com .

Fair Use Statement

The data and findings in this article may be used for noncommercial purposes only. When sharing or republishing, please include a link with proper attribution to DOSS.