A survey of 2,338 US adults, published by Semrush on September 18, found that nearly 81% of consumers who shop using online chatbots have decided against a purchase based on information the AI provided. The data reveals a two-way gatekeeper effect: 58% of AI users have bought products a chatbot recommended, and an equal 58% have abandoned a purchase over AI-supplied information. Among weekly heavy users, more than 80% have purchased AI-recommended products, but over 85% say their willingness to buy drops when a chatbot mentions mixed or negative reviews.
The finding reshapes the goal of generative engine optimization (GEO). Visibility in an AI answer is no longer enough. Constantine von Hoffman, the article's author, identified the shift in consumer behavior: what people ask AI has moved from "what should I buy" to "should I be buying from you." The channel picture shows that only 29% of consumers use chatbots for product research, but among weekly AI users, that share jumps to 55%. This group now reaches for chatbots more than recommendations from friends and family, social media, retail websites, or YouTube. Only search engines and review sites still rank ahead.
Two counterintuitive findings sit alongside the headline numbers. First, 65% of AI users say chatbots have replaced part of their Google product searches, yet nearly half made no change in the total amount of Google-based purchase research over the past year. Second, consumers bounce between Google, reviews, brand sites, and AI within a single purchase decision. The result is channel stacking rather than channel replacement. Brands have gained another shelf where they can win points, and another shelf where they can lose them.
Three pressure points for marketing teams
Content teams now own more than keyword rankings. They own the brand narrative as AI retells it. Negative reviews, controversies, and price doubts can all be relayed by the model to the next customer. Media teams face a math problem: paid placements buy exposure but not conclusions. In the survey, 42% of consumers dislike ads inside chatbots, and among those who do, 66% say the ads make them doubt the fairness of the AI's answers. Brand teams need daily monitoring of how AI talks about them. The tolerance window is narrowing: about three in ten consumers will churn upon discovering content was AI-generated, and 37% will leave when they expect a human and find an AI on the other side.
The operational fix sits in three steps. First, fold review governance into GEO by sweeping negative-review keywords and response times on major platforms. Reviews are high-weight material for AI to cite, and the 85% sensitivity figure is the budget justification. Second, rewrite content so AI can extract it cleanly: definitions up front, data in tables, sources credited. Third, stand up a weekly monitor of 50 purchase-intent prompts to watch how AI evaluates your brand and whether it is crossing you off the candidate list. When a negative narrative appears, fix the source content first.
One note on the data's provenance: Semrush is MarTech's parent company and also sells AI visibility tools, so the numbers carry a commercial interest. But the "talking customers away" side has now been laid on the table by data, and the heaviest word in the survey is "can't buy." Visibility can be bought with money; the AI's conclusion cannot. GEO's moat does not sit in the media budget. It sits in how fast you respond to bad reviews and how accurate your factual content is. For professionals working across AI for Marketing Courses and communications strategy, the implication is direct: if you get into AI answers and cannot hold the ground, every optimization before it was wasted.
Consumer trust gaps and compliance pressure
Adobe's 2026 annual report, based on surveys of 3,000 experience executives and 4,000 consumers, puts hard numbers on the disconnect between enterprise ambition and consumer comfort. While 49% of enterprises expect AI agents to handle 78% of customer-service interactions within 18 months, only 19% of consumers endorse making AI agents the primary brand touchpoint. Only 21% accept agents autonomously completing large-purchase decisions, a figure that falls to 16% for major purchases. On the compliance side, GDPR's seventh-anniversary retrospective, published by dotdigital, surfaces the penalties: a cap of €20 million or 4% of global turnover, with Meta fined €1.2 billion over cross-border data transfers. Processing personal data with AI now requires a lawful basis, explainable automated decisions, and a data protection impact assessment before launch.
An academic paper accepted in December 2024 maps the global privacy compliance landscape across GDPR, CCPA, and China's PIPL as three parallel tracks. Cross-border data transfer remains the biggest compliance risk exposure. The paper's implementation advice is to build a unified privacy baseline and make incremental adaptations per market. The frameworks of CCPA, UK GDPR, and Brazil's LGPD are all drifting toward GDPR, so building global compliance with it as the baseline is cheaper than constructing separate systems country by country.
Execution-layer standardization and process readiness
Two product signals point to the same trend. KNOREX released an ad API built for agentic AI in February, positioned as the execution layer for cross-channel campaigns across Google, Meta, TikTok, and other platforms. Unified.to published a blog post detailing the integration-cost math: connecting 13 ad platform APIs, each with different authentication methods, object models, reporting schemas, and rate-limit rules, drives per-platform maintenance labor costs linearly upward. Their answer is a standardized interface layer that supports multi-platform dashboards and AI-driven campaign management. The contest for the entry point of the AI-agent ecosystem has already started.
Process readiness lags behind tool adoption. Consultant Stacey Ackerman documented three client cases for MarTech where AI amplified existing workflow problems. An insurance content team bought AI tools expecting half the headcount and five times the output; content got produced, but everything jammed at approval because one reviewer's reading load quintupled. A homebuilder produced a full kit of go-to-market materials in one day, then spent three weeks in approval limbo because nobody had designated a single approver. A healthcare group produced new-hospital opening materials in days, then blew up over font colors because brand standards had never been written down. Ackerman's prescription is blunt: map the end-to-end process first, find the bottleneck, then decide what tools to buy.
Why this matters for creatives, marketers, and communications professionals
The gatekeeper is already on duty. AI is driving sales and turning customers away at equal rates, and 85% of AI users are highly sensitive to negative reviews as retold by AI. The preparation at the table has only just begun. Three actions cost nothing and set you up to absorb whatever comes next: sweep how AI talks about your own brand this week, add a data protection impact assessment for any AI email features already in use, and swap the content KPI from output volume to conversion contribution. The speed at which you write AI into your workflows must keep up with the speed at which you keep humans in the critical seats.
Your membership also unlocks:
