TITLE: Microsoft Copilot Ad Shows AI Tasks It Can't Perform

Microsoft Releases Holiday-Themed Copilot Advertisement

Microsoft has launched a new television advertisement promoting its Copilot AI assistant, centered on holiday preparations. The 30-second spot, aired during the festive season, depicts users interacting with Copilot on Windows devices to handle tasks like synchronizing smart lights with music, scaling recipes for larger gatherings, assembling furniture, and reviewing homeowners association guidelines for decorations. The ad features actors in holiday settings, including a brief appearance by a Santa Claus figure, and poses the question, "Are you ready for the holidays?"

The promotion highlights Copilot's Voice and Vision modes, where the AI analyzes on-screen content and provides step-by-step guidance. However, examinations of the advertised prompts reveal that Copilot often fails to replicate the seamless assistance shown, frequently producing inaccurate or incomplete responses.

Company Response to Simulation Claims

Microsoft employs fictional elements in the ad, such as the "Relecloud" smart lighting interface, which does not exist as a real product. Similar placeholders appear in Microsoft's case studies, like the oft-used "Contoso" corporation. A Microsoft spokesperson, Nicci Trovinger, general manager of Windows marketing, stated that all Copilot responses in the ad are derived from actual AI outputs to the depicted scenarios, though shortened for the commercial's duration in line with industry standards.

The ad's visuals, including an image of a large inflatable reindeer decoration and a fabricated HOA guidelines document, were created specifically for the spot and modeled after real references, according to Trovinger. Despite these assurances, real-world testing of the prompts underscores limitations in Copilot's capabilities.

Testing Smart Light Synchronization

One key scene shows a user prompting Copilot: "Show me how to sync my holiday lights to my music." The ad cuts to lights pulsing in rhythm with "A-Punk" by Vampire Weekend, navigated via the Relecloud website.

In practice, using a screenshot of the ad's interface, Copilot attempted to guide navigation by highlighting a "Sync Mode" drop-down with its on-screen cursor. However, it frequently claimed to have highlighted elements that were not selected and invented a green "Apply" button, mistaking it for a color preset. When tested with the actual Philips Hue Sync app, Copilot correctly identified the interface initially and suggested clicking the Music tab and "Start light sync" button. But it soon hallucinated absent buttons, referenced pre-configured Entertainment Zones unnecessarily, and failed to highlight elements despite assertions to the contrary.

Copilot's cursor feature, intended to assist visually, activates only on specific request and responds slowly, often persisting after the AI shifts topics. These issues render the process frustrating rather than efficient.

Challenges with Recipe Scaling and Assembly

The ad briefly shows prompts like "Help me figure out these instructions" and "Convert this recipe on my screen so it feeds 12," with an interjection of "14!" from another character. No on-screen details are provided, prompting tests with common holiday scenarios.

For assembly, displaying IKEA's Kallax shelf manual led Copilot to confuse dowels with screws or nails and misread page numbers as steps, complicating guidance. In recipe scaling, using a stuffed mushrooms recipe from Sip and Feast, Copilot recognized the need to multiply ingredients by approximately 2.3 for 14 servings. Yet it performed only partial calculations, deferred to the user, or pivoted to unrelated questions.

When the site offered scaling buttons for 2x or 3x, Copilot erroneously described them as adjustable plus/minus controls for exact servings. A direct request to compute and document ingredients resulted in no output, despite Copilot's affirmation.

Reviewing HOA Guidelines

The final prompt involves: "Can you read the HOA guidelines and make sure I'm not breaking any rules?" alongside an image of an oversized reindeer encroaching on a neighbor's property. Copilot in the ad advises adjusting the decoration to avoid crossing property lines.

Testing with the ad's screenshot, Copilot identified the relevant rule prohibiting large inflatables from crossing boundaries. However, it provided equivocal assessments, speculating on the reindeer's positioning without clear conclusions and often yielding to user judgment.

Broader Implications for AI Advertising

This advertisement follows previous Microsoft promotions for Copilot that faced similar scrutiny. Earlier tests of Copilot Vision and Voice modes highlighted inconsistencies, such as the AI making a device feel "incompetent" during interactions. The pattern raises questions about the realism in AI marketing, where demonstrations may prioritize appeal over accuracy.

Microsoft continues to refine Copilot, integrated into Windows and other products, amid growing scrutiny of AI reliability. Issues like hallucinations—where the AI generates false information—persist across large language models, including those from competitors like OpenAI.

As holiday demands peak, users relying on such tools for practical tasks may encounter the gaps between advertised promise and real performance. Microsoft has not announced immediate updates addressing these specific shortcomings.