Questions about this study

The study tested whether 47 Utah AI companies from the AI Utah 100 list appear in answers generated by four major AI systems: ChatGPT, Claude, Gemini, and Google AI Overview. Each company received six non-branded, buyer-intent prompts (the kind a prospective customer might type when searching for a solution), producing 1107 usable answers from 1296 total queries. The goal was to measure AI visibility, meaning how often and where these companies show up when AI tools answer category-level questions rather than direct brand searches.

Zero AI visibility means the company did not appear in any answer to any of its six prompts on any of the four engines tested. In this study, 12 companies (25.5 percent of the cohort) fell into this category. These companies include Airin, Artifact, Bunked, Cinco.AI, Cynch.ai, Drumdata.ai, Harbor Health, Jaunt, Kahoa.ai, Phi, Inc., Soar.com, and energy.work. When a buyer asks an AI assistant for solutions in these companies' categories, the AI recommends competitors instead.

Several factors likely contribute. First, AI models draw from training data and web retrieval, so companies with thin online footprints or limited third-party coverage may not surface. Second, the models tend to recommend well-documented national or global competitors when they lack strong signals about a smaller player. For instance, when prompts relevant to Harbor Health were tested, models recommended University of Utah Health, Intermountain Health, and national insurers instead. This suggests that being on a regional list does not automatically translate into the kind of structured, widely cited web presence that AI systems rely on.

Six prompts is a modest sample, and we do not claim it captures every possible buyer query. What it does capture is a structured, consistent test across all 47 companies, producing 1107 usable answers. The prompts were designed to reflect realistic buyer-intent searches, not brand queries. A company invisible across all six prompts on all four engines has a visibility problem, though we cannot say exactly how large. A company visible on some but not others has uneven coverage. The method is sufficient to detect presence or absence, even if a larger prompt set would refine the percentages.

Utah-qualified prompts lifted appearance rates from 29.6 percent to 44.7 percent. The likely mechanism is that when a prompt includes a geographic qualifier, the AI system gives more weight to regional authority signals: local news coverage, Utah-specific directories, mentions in state business publications like utahbusiness.com, and so on. This suggests companies can improve visibility by building a stronger regional content footprint, though it also means their visibility may remain weak for generic, non-geographic searches where national competitors dominate.

ChatGPT recommended the highest share of the cohort at 68.1 percent (32 of 47 companies appeared at least once). Claude recommended the lowest share at 53.2 percent (25 companies). Google AI Overview and Gemini fell in between at 57.4 percent and 55.3 percent respectively. This variation matters because 13 companies (27.7 percent of the cohort) appear on at least one engine but not all four.

Mention position is where in a list of recommendations a company appears. Position one means it was the first vendor mentioned, position three means it appeared third, and so on. The average mention position across all 382 appearances in this study was 2.44. Google AI Overview placed companies earliest on average (position 1.92), while ChatGPT placed them latest (position 2.71). Earlier positions likely receive more attention from users, just as higher rankings do in traditional search results, though we did not measure click-through rates.

The five most visible companies were Limble CMMS, Recursion, Spiff, Weave, and Entrata. These companies appeared consistently across multiple engines and prompt types. The common thread is substantial third-party coverage: press mentions, analyst reports, review site profiles, and other external sources that AI models can retrieve and cite. Visibility correlates with having a web presence that extends well beyond the company's own website.

The most frequently cited sources across all engines include youtube.com, gartner.com, linkedin.com, clutch.co, en.wikipedia.org, reddit.com, g2.com, and builtin.com. Different engines have different preferences: ChatGPT leans on gartner.com and linkedin.com, Claude cites gartner.com and en.wikipedia.org, and Google AI Overview draws heavily from youtube.com and reddit.com. This means companies seeking AI visibility should consider their presence on these specific platforms, not just their own websites.

For each of the 12 invisible companies, the AI models consistently recommended a small set of national or global competitors. For example, prompts relevant to Soar.com surfaced Otter.ai, Fireflies.ai, Sonix, Granola, and Fathom. Prompts relevant to Cynch.ai surfaced QuickBooks Online, Xero, FreshBooks, Zoho Books, and Docyt. Each invisible company has its own list of 5 competitors that tend to appear in its place. These are the alternatives buyers encounter when AI systems do not know about or do not recommend the Utah company.

Yes, AI outputs are not perfectly deterministic. The models can return different answers on repeated runs, and they are updated periodically. This study ran all queries within a single window on 31 July 2026 to minimize variation, but we cannot rule out that some borderline companies might appear or disappear on a different day. The core finding, that 12 companies never appeared across 1107 answers, is robust enough that day-to-day variation is unlikely to flip it entirely, but specific percentages should be understood as a snapshot rather than a fixed value.

The study does not test interventions, but the data points toward several hypotheses worth testing. First, build presence on the third-party platforms these models cite, particularly gartner.com, linkedin.com, clutch.co, g2.com, and Wikipedia. Second, strengthen regional authority signals through coverage in Utah-specific publications, since Utah-qualified prompts lifted appearance rates significantly. Third, ensure structured data and clear category descriptors on your own site so models can correctly classify what you do. None of these are guaranteed fixes, but they address the mechanisms the data suggests are at work.

The AI Utah 100 list includes some very large organizations and academic institutions. We excluded these because their visibility dynamics differ fundamentally from growth-stage companies. A company like Adobe or a university research lab has decades of accumulated web presence and citations. Including them would skew averages and make the findings less useful for the mid-market and growth companies that make up most of the list. The final cohort of 47 companies represents those for whom AI visibility is an open, actionable question.

The prompts were non-branded and buyer-intent in nature, meaning they reflected how a prospective customer would search (for example, 'best CMMS software for manufacturing' rather than 'tell me about Limble'). This is the relevant test because brand queries already assume awareness. The same six-prompt structure applied to every company. You can argue that different prompts might yield different results, and that is likely true at the margins, but a company invisible across all six prompts on all four engines has a genuine visibility gap. The method is consistent, even if it is not exhaustive.