New survey data shows AI shopping chatbots talk a lot of people out of buying. Nearly 81% of consumers who've used a chatbot while shopping say they've decided against a purchase because of something the AI told them, according to a Semrush survey of 2,338 U.S. adults reported by MarTech on 18 September 2026. If you run a chatbot that recommends products, this changes how you should be building it.
Do AI Shopping Chatbots Actually Stop People From Buying?#
Yes, often. MarTech senior editor Constantine von Hoffman reported the numbers on 18 September: nearly 81% of consumers who use chatbots for online shopping have decided against a purchase based on AI guidance, drawing on a Semrush survey of 2,338 U.S. adults.
Fifty-eight percent of AI users say they've bought something a chatbot recommended. The same 58% say they've talked themselves out of a purchase based on what the bot told them. Among people who use AI multiple times a week, more than 80% have bought an AI-recommended product, but more than 85% of that same group back off the moment a chatbot surfaces mixed or negative reviews. Sixty-five percent say chatbots have already replaced some of their product-related Google searches.
Does This Actually Affect Your Business?#
Short version: it depends on what your chatbot does.
If you run a services business and your chatbot only books calls or answers hours-and-location questions, none of this touches you today. This is about chatbots doing product comparison and recommendation. Skip to the next section.
If you sell physical or digital products through a WhatsApp bot, a website assistant, or anything that answers "which one should I get," this is worth pausing on. The chatbot has become a comparison engine your customer trusts more than your product page. Seventy-one percent of consumers now believe chatbots recommend the "best" brands and products.
It cuts both ways, though. 59% say they've discovered a new brand through a chatbot recommendation, so a well-built bot is an acquisition channel too, not only a risk.
The Honest Trade-Offs#
The upside is real. A bot that can answer "does this fit a size 10" or "is this in stock" at 11pm converts leads a static FAQ page never will, and that 58% purchase-assist number backs it up.
The part nobody puts in the pitch deck: a chatbot answering freely will occasionally out-honest you. Ask it to compare your product against a named competitor and it might tell the truth about something the competitor does better. That's not a bug you patch with a cleverer prompt. It's what happens when you hand a stranger a microphone into your sales conversation.
What we don't know yet: whether this 81% figure holds outside the U.S., or whether it shifts as people start treating chatbot answers with the same skepticism they already have for chatbot ads. 42% already say they dislike chatbot ads, and 66% of that group say the ads make them question the AI's honesty generally. My read is that skepticism is coming for organic chatbot answers too, not just the paid ones.
A few signs your bot might already be doing this to you:
- It answers open comparison questions ("is X better than Y") instead of routing them
- It can surface a competitor by name without a human in the loop
- Nobody on your team has actually read a week of its transcripts
How We're Building Chatbots Around This#
We reviewed this the week it came out, and it didn't change our build approach so much as confirm it. The chatbot and WhatsApp lead-response flows we ship for D2C brands and service businesses answer product questions from a locked knowledge base and a fixed set of product cards, not from the model improvising a comparison. If a shopper asks something the knowledge base can't answer honestly, the bot hands off to a human instead of guessing.
That's the difference between a bot that occasionally talks someone out of a sale because it made something up, and one that occasionally talks someone out of a sale because the honest answer was genuinely no. We'd rather lose the sale the second way.
At WebEpex, we build these chatbot and WhatsApp flows in Voiceflow for D2C brands, high-ticket service businesses, and SaaS founders across the GCC, Europe, the US, Canada, and Indian SMBs. The carousel-plus-knowledge-base pattern is our starting point on any bot that touches a buying decision, not an add-on we upsell later. We also keep the lead-response layer separate from the sales-assistant layer, because a slow reply and a bad recommendation are different failure modes.
Meta and Salesforce are racing to put their own AI agents on WhatsApp for this kind of conversation, and Google is doing the equivalent for product discovery inside Search and Shopping. None of that changes the build principle. More of your customer's buying decision now happens inside a chat window you don't fully control, so the parts you do control need to be right.
What I'd Tell a Client This Week#
Check whether your chatbot can freely compare you against a named competitor, or answer "is this worth it" with anything beyond your own catalog. If it can, decide on purpose whether that's a feature or a leak. For a considered, high-ticket purchase, an honest bot builds trust. For a low-margin item, a bot that improvises can talk your buyer into a cheaper alternative down the road.
If you don't run a shopping bot at all, do nothing. This survey is about people who already use one, and building one just to chase this data point is the wrong reason to build one.
If you're running a WhatsApp or web bot and want a second pair of eyes on what it's actually telling people, send me the flow and I'll tell you straight. Two minutes, no pitch. cal.com/webepex/growth-review
