Beyond the Metric Illusion: Why High AI Visibility Scores Can Hide a Brand’s Biggest Blind Spot

In the rapidly evolving digital landscape, modern marketing analytics have found a new North Star: AI visibility. For CMOs and digital strategists, watching a brand’s mention frequency climb across generative search and conversational platforms like ChatGPT, Claude, and Gemini has become the ultimate validation of digital dominance.

However, a groundbreaking study from Foundation Labs released in September 2026 suggests that this metric may be dangerously misleading. Analyzing project management giant ClickUp, researchers uncovered a stark paradox: a brand can achieve near-ubiquitous presence in AI-driven discovery while simultaneously failing to secure a single default recommendation when real buyers ask generative engines to make a choice.

This phenomenon—dubbed "Model Drift" and the illusion of visibility—exposes a critical vulnerability in modern brand strategy. As enterprise buyers increasingly turn to artificial intelligence as a primary decision-making engine, tracking how often a brand is mentioned is no longer enough. The real battleground is why and when AI recommends a product to close a deal.

Main Facts: The ClickUp Case Study

The Foundation Labs study challenges foundational assumptions about how artificial intelligence evaluates B2B software.

In September 2026, researchers ran a comprehensive audit across six major AI surfaces: ChatGPT, Claude, Gemini, Perplexity, Google AI Overview, and Google AI Mode. Using 21 distinct prompts executed twice, they tested how these engines handled unaided software discovery, direct comparisons, and specific feature queries.

The results exposed a dramatic disconnect:

• The Presence Rate: Across 120 discovery answers where the brand was not explicitly named in the prompt, ClickUp appeared 117 times, registering a staggering 97.5% presence rate.
• The Recommendation Rate: When researchers prompted each platform to choose between ClickUp, Asana, and monday.com for a hypothetical 100-person company, ClickUp received zero out of 12 default recommendations.
• The Sentiment Paradox: Despite missing out on recommendations, ClickUp’s average sentiment remained above neutral on five of the six surfaces, with models regularly praising its feature breadth, customization options, and automation capabilities.

The core takeaway is clear: high visibility ensures a brand is part of the conversation, but it offers zero guarantees about who ultimately wins the recommendation.

Chronology and Methodology: Uncovering Model Drift

To understand how a platform could be universally recognized yet systematically bypassed, Foundation Labs mapped out a rigorous, multi-phased testing methodology throughout September 2026.

Phase 1: Unaided Discovery Mapping

The researchers initiated the study by feeding the AI platforms broad, category-level prompts without naming specific vendors. Across OpenAI’s ChatGPT, Anthropic’s Claude, Google’s Gemini, Perplexity, and Google’s search-integrated features (AI Overview and AI Mode), ClickUp dominated the results. It appeared in 117 of 120 discovery responses, establishing that the brand’s digital footprint was massive enough to consistently trigger inclusion algorithms.

Phase 2: The Default Recommendation Challenge

Next, the research team introduced friction into the queries. They asked the models to act as advisors recommending a specific tool for a 100-person enterprise weighing options among ClickUp, Asana, and monday.com.

This is where the paradigm shifted. While the AI engines acknowledged ClickUp’s robust feature set, they consistently penalized it for perceived adoption risks. Across all 12 direct evaluations, competitors like Asana and monday.com swept the default recommendations.

Phase 3: Contextual Query Shifting

Recognizing that generalized prompts yield generalized penalties, the researchers tested how the models responded when the prompt parameters changed. When the questions pivoted toward specific use cases—such as consolidating tech stacks, reducing software sprawl, or leveraging embedded AI agents for project execution—ClickUp’s fortunes reversed dramatically. When asked about consolidating tools, ClickUp placed first in 11 of 12 answers. When narrowed to embedded AI execution, Claude specifically lauded ClickUp as "the most aggressive on agents," catapulting it to the top of the evaluation stack.

Supporting Data: The Anatomy of AI Divergence

The study’s extensive dataset highlights several critical mechanics driving AI outputs: source mix variation, category language fragmentation, and unpredictable temporal shifts.

The Source Mix Split

Different AI surfaces rely on vastly different data ecosystems, leading to fragmented brand narratives.

• ChatGPT favored product-owned pages and official documentation.
• Claude leaned heavily on editorial and third-party review sources.
• Google AI Overview and AI Mode relied heavily on community-driven content, displaying a massive reliance on Reddit and YouTube.

For instance, Google AI Overview cited 24 Reddit links and 23 YouTube links in answers concerning ClickUp and its competitors, while AI Mode cited 24 Reddit links and 36 YouTube links. Conversely, Claude cited zero community video or forum links, illustrating how drastically a brand’s narrative shifts depending on the underlying retrieval engine.

Semantic Drift and Category Language

ClickUp officially positions itself as a "Converged AI Workspace." However, the AI models frequently ignored this self-assigned positioning. Only ChatGPT and Claude used the term "Converged AI Workspace" consistently; the remaining four platforms defaulted to traditional industry labels like "project management," "work management," or "work OS."

This variance underscores a vital rule of generative optimization: AI surfaces category language from retrieved sources. If a proprietary term appears almost exclusively on a brand’s own website while the rest of the internet uses traditional nomenclature, the AI will default to the language of the broader web.

Temporal Instability

Results even fluctuated within identical surfaces across consecutive days. On September 12, Google AI Overview omitted ClickUp from its "best AI productivity platforms for teams" response, only to include ClickUp Brain in the exact same query on September 13. This volatility proves that a single audit on a single day is insufficient for accurate brand tracking.

Official Responses and Industry Context

The implications of the Foundation Labs findings extend far beyond project management software, touching on a massive impending shift in global commerce.

According to research cited from McKinsey & Company, only 16% of Fortune 500 brands systematically track their performance in AI search. This oversight persists despite nearly half of consumers already relying on AI search tools, with McKinsey projecting that $750 billion in US revenue will flow through generative search channels by 2028.

Industry analysts note that B2B software marketing has been slow to adapt from traditional search engine optimization (SEO) to generative engine optimization (GEO). While traditional SEO prioritizes keyword density and backlink volume to capture a spot on a static search engine results page (SERP), generative AI synthesizes arguments, weighs risk factors (such as onboarding friction and administrative burden), and delivers synthesized recommendations.

In ClickUp’s case, the AI models calculated that a 100-person company faced a steeper learning curve and higher administrative overhead with ClickUp than with Asana. Consequently, even though the models knew what ClickUp was (driving the 97.5% visibility score), they actively advised against choosing it for a mid-market deployment.

Implications: How to Run a Model Drift Audit

For modern marketing leaders, the lesson is clear: vanity metrics hiding behind high visibility scores can mask active losses at the decision stage of the buyer’s journey. Chasing an overall presence score is a fool’s errand if the AI engines are actively warning prospective buyers away from your product due to unaddressed friction points.

To combat Model Drift and capture genuine pipeline value from AI search, marketing teams must adopt a proactive, four-step audit framework:

1. Map Your Decision Prompts: Stop testing generic brand keywords. Instead, input the exact scenario-based prompts your buyers use when asking AI to compare solutions for specific company sizes, use cases, or pain points.
2. Analyze the Recommendation Gap: Measure the delta between how often your brand appears in discovery versus how often it wins the default recommendation. Identify why the AI hesitates (e.g., perceived complexity, onboarding risk, or lack of governance features).
3. Audit Your Source Footprint: Determine which data ecosystems (Reddit, YouTube, editorial review sites, or official documentation) feed the specific AI surfaces your buyers use. If community sentiment is driving negative adoption narratives, invest in community-focused digital PR and advocacy.
4. Align Positioning with Web Consensus: Evaluate whether generative engines are adopting your proprietary marketing terminology or defaulting to industry-standard phrasing. Build external validation across trusted third-party sites to reinforce your preferred category language.

As generative search fundamentally rewrites the front door to the internet, brands must evolve past the superficial comfort of high visibility metrics. As the data shows, being seen is only half the battle—the real challenge is giving artificial intelligence an undeniable reason to pick you.