All I initially wanted was something exceedingly simple: a newer, well-performing smartwatch that would complement my Android phone. I had already narrowed my choices to a few models—the Google Pixel Watch 4, celebrated for its seamless integration with the Android ecosystem, and the Garmin Vivoactive 6, known for its robust fitness tracking features and durable design. Yet, despite my apparent clarity, my AI-powered shopping assistants seemed unwaveringly convinced that opting for older smartwatch models from previous years would somehow be a wiser decision. Their insistence was both puzzling and amusing, revealing that artificial intelligence can be somewhat inflexible when interpreting user intent.
During the last month, a number of major technology companies—including OpenAI, Google, Perplexity, and Microsoft—have competed to enhance the consumer shopping experience by embedding new purchasing-focused functionalities into their AI interfaces, just in time for the holiday buying frenzy. Each platform has taken a slightly different approach. OpenAI’s ChatGPT, for instance, can now produce individualized buying guides tailored precisely to one’s needs and preferences. Microsoft’s Copilot integrates tools to track historical and real-time price fluctuations, while Google’s Gemini expands its functionality beyond digital research by actually contacting local stores on behalf of the user. The landscape of digital shopping assistance, therefore, has evolved into something highly experimental, an intersection of automation, personalization, and seemingly intelligent recommendation.
When I enlisted the help of four distinct AI programs to guide my smartwatch search, the experience was simultaneously enlightening and exasperating. On the one hand, I was genuinely surprised by how interactive these tools have become; on the other, I quickly learned that their shortcomings could be equally significant. Each AI displayed pockets of real competence—an accurate understanding of my priorities or a creative product comparison—but every one of them stumbled in ways that were both humorous and, at times, deeply limiting.
To begin my exploration fairly, I posed an identical question to ChatGPT, Gemini, Perplexity, and Copilot: “Can you help me find a good Android smartwatch for my Nothing CMF Phone 1?” From that single prompt, each assistant proceeded to steer me through its own research and shopping workflow, translating my vague curiosity into a concrete selection process.
Among all four, ChatGPT proved the most conversational and immersive. Its recently introduced Shopping Research feature demonstrated a remarkable degree of structure and depth. The AI began by clarifying what mattered most to me—design aesthetics and battery longevity—then carefully curated a lineup of roughly a dozen watches that it asked me to rate using a simple “more like this” or “less like this” scale. From my responses, it synthesized a shortlist of recommendations accompanied by a comparative specifications table, highlighting dimensions, power efficiency, price ranges, and feature sets. The culmination of this digital dialogue was a neat compilation of buying links, directing me to discounts on what it deemed the optimal choice: the Garmin Vivoactive 5. The full process was surprisingly swift, lasting about ten minutes, and left me impressed by its polished interaction.
In addition to that top suggestion, ChatGPT recommended several other viable options, including the Fitbit Versa 4, the Google Pixel Watch 3, and the Ticwatch Pro 5. Each came with nuanced evaluations comparing different metrics such as endurance, compatibility with Android apps, pricing flexibility, and intended usage categories—ranging from “ideal for daily life” to “budget-friendly” or “optimized for Google ecosystem integration.” This level of tailored segmentation was useful and gave me a sense of curation reminiscent of a conversation with a knowledgeable sales consultant.
Ordinarily, I would not think to consult an AI engine for guidance on high-tech shopping, assuming the process would lack the nuance and contextual awareness that human reviewers offer. Yet I can now appreciate how an AI-driven recommendation system might genuinely assist a less experienced buyer—someone uncertain about starting points or overwhelmed by endless product choices. However, my cautious optimism waned when I realized that following ChatGPT’s lead could easily have resulted in purchasing a last-generation device rather than the newer Garmin Vivoactive 6. The second iteration offers several meaningful advancements—expanded storage, more accurate GPS capabilities, and modernized functions such as Smart Alarm—that could significantly influence a buyer’s satisfaction. To be fair, ChatGPT only delivered results based on the parameters I provided, and I never explicitly asked for “the latest” model. Moreover, the Vivoactive 5 remains widely stocked by major retailers like Amazon and Best Buy. Still, awareness of generational differences is crucial when deciding between saving money and embracing the most up-to-date technology.
Moving on to Google’s domain, my experience with its “Call for me” functionality—accessed through the Google app rather than directly through Gemini—was more awkward than empowering. After appending “near me” to my query and confirming location details, I triggered a series of automated phone calls to nearby retailers. Fifteen minutes later, I received an email politely informing me that none of the contacted stores carried Garmin watches. While technically efficient, the outcome underscored the clumsiness of automating real-world logistics through an interface that still struggles to interpret market availability.
Similar complications appeared across the remaining AI systems. Gemini correctly identified some watches as the “latest” in name yet paired its recommendations with data tables featuring models from prior product cycles—many dating to 2023 or even 2022. For example, it highlighted the Google Pixel Watch 2 as a top contender, even though that model had long since been superseded by newer entries like the Pixel Watch 4. While older variants can often be found at discounted rates, they are typically saddled with practical trade-offs: diminished battery performance, thicker bezels, older charging systems, slower processors, and reduced sizing flexibility.
Perplexity.ai performed slightly better when it correctly pointed me toward the contemporary Pixel Watch 4, but it confusingly supplemented that suggestion with the Samsung Galaxy Watch 4—an artifact from 2021. Its integrated shopping tab, on the other hand, impressed me with how quickly it surfaced product listings and direct purchasing links, though the “More Products” section was a chaotic mix of questionable no-name devices and even an unrelated smartphone. It was as though the AI’s eagerness to be helpful had overwhelmed its sense of relevance. Nonetheless, if affordability were my sole criterion, I could have walked away with a $7 “Smart Watch with Bluetooth Call,” a humorous reminder that precision still eludes even the most advanced recommendation engines.
Microsoft’s Copilot offered an experience somewhere between polish and imperfection. It began by suggesting the CMF Watch Pro 2—an entirely logical recommendation given my Nothing CMF Phone 1—yet it overlooked the more recent CMF Watch Pro 3. Still, Copilot’s user interface stood out as intuitively organized. The shopping sidebar provided a detailed view including historical pricing trends, aggregated consumer sentiments, multiple merchant links, and an optional alert system for price drops. Adjusting my query to specify “the best current smartwatches” finally prompted all four AIs to acknowledge the CMF Watch Pro 3 as the state-of-the-art iteration; however, relics like the Watch Pro 2 persisted among their lists of recommended options. The likely explanation is that AI models tend to favor data with abundant user reviews, which older product releases naturally possess.
This reveals a subtle but significant flaw: unless a user explicitly requests the “latest” or “current” models, the AI tools appear blind to generational advancements. A human reviewer, by contrast, would instinctively highlight those differences, contextualizing whether an upgrade is worth the added cost or not. While machine learning systems are making strides toward that kind of contextual awareness, their grasp of evolving product timelines remains inconsistent at best.
After all these encounters, the overall impression is clear: though these AI shopping assistants sound revolutionary in principle, their real-world execution remains imperfect. Their recommendations draw upon product databases that seem frozen in time, confidently endorsing devices that are, in fact, behind the curve. For an inexperienced shopper, this temporal mismatch could easily result in buying an obsolete model or simply overlooking fresher, more efficient alternatives that the algorithm fails to recognize.
This disconnect between technological ambition and practical utility makes it difficult to wholeheartedly endorse AI shopping aids—at least in their current form. A few, like ChatGPT and Copilot, are clearly evolving in the right direction, exhibiting structural sophistication and consumer-oriented design. Nevertheless, until these systems consistently surface up-to-the-minute information comparable to what can be found in expert human reviews and detailed comparison videos, the safest course remains traditional self-directed research. For now, it appears that in the realm of AI-assisted purchases, a bit of human curiosity and discernment still go farther than any algorithmic shortcut could.
Sourse: https://www.theverge.com/ai-artificial-intelligence/830450/ai-shopping-assistants-are-stuck-in-the-past
