Carbuki Insights
AI Shopping Stalls at the Checkout, New Data Shows. At a Dealership, the Checkout Is the Phone.
Approximate share of AI usage by shopping-journey stage. Source: ResearchAndMarkets, AI Shopping Agents and Agentic Commerce 2026 (August 2026).
The auto industry has spent two years debating when an AI agent will buy a car on a customer's behalf. A new market study released this week suggests that debate is running ahead of the data. The report - AI Shopping Agents and Agentic Commerce 2026: Adoption Trends and Execution Limits, added to ResearchAndMarkets.com on August 17 - finds that AI usage in shopping is heavily concentrated at the front of the journey: roughly 62% of usage sits in product comparison, about 23% at checkout, and about 19% in post-purchase activity.
That shape - AI-saturated research, mostly human execution - should look familiar to anyone running a dealership in 2026. Shoppers arrive having done hours of AI-assisted homework, then complete the deal the way they always have: a phone call, an appointment, a desk. The new report puts numbers on the cliff between those two worlds. The numbers matter, because the cliff is exactly where a store either converts AI-era demand or hands it to the dealer down the road.
Myth vs. data: The myth is that agentic commerce means software quietly buying products end to end, with the retailer reduced to a fulfillment warehouse. The measured reality: AI usage drops from about 62% at product comparison to about 23% at checkout, and the report names payment authorization, digital identity, security and consumer trust as the binding constraints (ResearchAndMarkets, August 2026). The purchase still runs through channels people trust - and in auto retail, the most trusted execution channel is still a conversation.
What the report actually measured
The study tracks how consumers and platforms use AI across the full commerce journey - discovery, comparison, purchase, payment and post-purchase - across North America, Europe and APAC. Three findings stand out for automotive retail.
First, the funnel has gone AI-first at the top. Generative AI referral traffic to US retail websites grew by as much as 4,700% year over year in July 2025, while organic search traffic declined. Product visibility is increasingly decided inside AI-mediated environments rather than on page one of a search engine.
Second, execution has not followed. At checkout, AI usage runs at roughly a third of its research-stage level. The report attributes the gap to payment authorization frameworks that were not designed for autonomous agents, identity verification, fraud and security concerns, and plain consumer reluctance to delegate money decisions to software.
Third, the long-term stakes are large anyway. The report projects global agentic commerce revenue could reach 3 to 5 trillion US dollars by 2030 - a forecast that depends on the execution gap closing, not on research usage growing further.
None of this says AI shopping is overhyped. It says the growth so far has been in influence, not transaction. For a dealership, influence arrives as better-informed customers. Transaction still arrives as a ringing phone.
Cars make the checkout cliff steeper
In e-commerce, the 23% checkout figure describes a shopper hesitating to let an agent enter a card number. A vehicle purchase adds layers no cart has: financing approval, a trade-in that has to be seen and valued, state titling and registration, regulated F&I products, and franchise rules that keep the final sale with the dealer. Whatever timeline you believe for agentic checkout in general retail, the automotive version sits further out.
Which means the funnel cliff is not a distant strategy question - it is the current shape of demand. Cox Automotive's new AI in Auto Retail Tracker, launched August 11, found 63% of shoppers plan to use AI in their next vehicle purchase, and AI tools now rank at 36% as a vehicle research channel, nearly level with automotive-specific websites at 38%. On the other side of the counter, only 29% of dealers have begun adjusting to AI-powered search, and while 82% of dealers report using AI somewhere in the store, many are not measuring its impact. We looked at that tracker in detail in our breakdown of the shopper-dealer AI gap.
Put the two datasets together and the picture is consistent:
| Journey stage | Global AI usage share (2026) | What that stage looks like in auto retail |
|---|---|---|
| Product comparison | ~62% | Model comparisons, payment math and inventory questions asked of ChatGPT, Gemini and Claude |
| Checkout / transaction | ~23% | Financing, trade-in, F&I, scheduling - phone calls and showroom visits |
| Post-purchase | ~19% | Service booking, recalls, status calls, retention outreach |
Sources: ResearchAndMarkets (August 2026); stage descriptions reflect typical US dealership workflows.
The middle row is the one to stare at. Every stage where AI usage collapses is a stage that, in a dealership, happens on your phone lines and in your CRM.
AI-referred demand is worth more, not less
There is a tempting managerial read of the checkout cliff: if AI cannot close, AI traffic must be low quality. Retail data points the other way. Adobe Analytics, measuring over 1 trillion visits to US retail sites, reported that during the 2025 holiday season traffic arriving from generative AI tools grew 693% year over year - and that this traffic converted 31% better than other sources, produced 254% higher revenue per visit for the season to date, bounced 33% less often, and spent 45% more time on site.
Those are e-commerce figures, not dealership figures, and the caveat matters. But the mechanism travels: a shopper who spent an hour with an AI assistant narrowing to two trims and a payment range is not a cold lead. Cox's own buyer-journey research has linked heavier digital and AI tool usage with more efficient, more satisfied purchases. When that shopper finally calls - and in automotive, they call - they are past comparison and standing at the execution stage the AI did not do.
That is the practical meaning of the 62-to-23 drop for a GM: the highest-intent contact your store receives in 2026 may be sourced by an AI you cannot see, arriving on a channel you have run for fifty years.
Where the cliff shows up in the P&L
The report frames the execution gap in payments language - authorization, identity, trust. The dealership translation is simpler: the moment AI hands off, a human process takes over, and that process leaks. Vendor analyses of dealership call handling (Numa, 2025) put missed service-department calls at 300 to 500 per week at many stores; vendors have an interest in the number being large, but the direction is not seriously disputed, and our own look at what missed calls cost a dealership found the leak concentrated exactly where staffing is thinnest - peak hours, lunches and after close.
The irony of 2026 is that the same technology creating the research-stage demand can now also catch the handoff. The economics of speed to lead have not changed: contact rates decay within minutes, and the store that answers first tends to write the deal. What changed is that an AI-referred caller has already finished the comparison stage - so the cost of a missed or mishandled call is no longer a lost lead, it is a lost buyer.
What to do this quarter - no autonomy bet required
The measured response to this data is not to wait for agentic checkout, and not to buy every tool with AI in the name. Four moves follow directly from the numbers:
- Measure your answer rate before anything else. You cannot manage the execution stage if you do not know how many execution-stage contacts never reach a human. Pull connected-call rates by hour, including service.
- Treat AI-referred contacts as bottom-funnel. A caller who names a specific trim, payment range or in-stock unit has likely done AI-assisted comparison already. Route and script accordingly - answer the question they are actually at, not the top of a BDC script.
- Stay legible to the research layer. The 4,700% referral surge rewards stores whose inventory, pricing and hours are structured, current and machine-readable. We covered the mechanics in how dealers stay visible to AI shopping tools.
- Close the measurement gap Cox found. 82% of dealers use AI; far fewer measure it. Tag AI-sourced calls and leads in the CRM and track them to appointments and deals, so next quarter's decisions run on your data, not a vendor's.
The bottom line
The report's competitive conclusion is that the fight in agentic commerce is over the interfaces and infrastructure connecting merchants with consumers. In automotive retail, that connecting infrastructure is not a payments protocol yet. For now, it is the phone line, the appointment calendar and the person - or increasingly the AI agent - that answers within two rings. The research stage has already moved. The stores that win the next stage will be the ones standing at the bottom of the checkout cliff, catching what the shopper's AI hands off.
If you want to see what that handoff looks like when it is answered every time, Carbuki builds AI voice agents that pick up dealership calls 24/7, book service and sales appointments, and log every conversation to your CRM.
Sources
- AI Shopping Agents and Agentic Commerce 2026: Adoption Trends and Execution Limits - ResearchAndMarkets via GlobeNewswire, Aug 17, 2026
- New Cox Automotive AI in Auto Retail Tracker Finds Growing Gap Between Dealers and AI-Powered Car Shoppers - Cox Automotive, Aug 11, 2026
- Adobe: AI-driven traffic surges across industries with retail experiencing biggest gains - Adobe for Business, Jan 12, 2026
- Cox study finds 63% of car shoppers plan to use AI, but dealers are falling behind - Dealership Guy News, Aug 11, 2026
- AI for Dealership Communication - Numa (vendor analysis), 2025
Carbuki builds AI voice agents for retail automotive — answering sales and service calls, following up on leads, and booking appointments 24/7 in multiple languages.
See how it works →