AI Utah 100 AI-Visibility Study

Why AI Visibility Matters for Utah's Tech Sector

The AI Utah 100 is a public signal. Being named to the list says the market, or at least its curators, views you as a leader. But that recognition lives in one PDF, one press release, one LinkedIn announcement. When a buyer types "best CMMS maintenance software" into ChatGPT or asks Gemini for "AI-powered commission tracking tools," they do not see a list of honorees. They see whatever the model retrieves, summarises, and ranks.

This study asks a narrow question: if a Utah AI company has been publicly recognised as a leader, does that recognition translate into visibility when someone shops the category without using the company's name?

The practical stakes are straightforward. If a B2B buyer uses an AI assistant to shortlist vendors and your company never surfaces, you are excluded from consideration before any sales conversation can happen. You may have the best product, the strongest references, and the most relevant case studies. None of that matters if the model never suggests you.

Who should care

• Founders and executives whose companies made the AI Utah 100 but who have not measured whether AI engines recommend them.
• Marketing leaders deciding where to invest in visibility, especially those who suspect that traditional SEO may not translate into AI-era discoverability.
• Investors and board members evaluating portfolio companies' go-to-market risk in a changing search landscape.
• Utah economic development stakeholders interested in whether the state's tech ecosystem is discoverable to buyers outside the region.

What this study does not measure

It does not measure brand awareness, purchase intent, or actual revenue. It does not measure whether these companies should be recommended, only whether they are. It cannot tell you whether a recommendation leads to a demo request or a closed deal. Those questions require different data. This study measures one thing well: when a buyer asks a category question, does the model name you?

How the Measurement Works

The study began with the AI Utah 100 list published by AI Utah and Codebase. From that list, we removed mega-cap companies and universities whose brands would appear in AI answers regardless of any optimisation effort. The exclusion list is published as a visible column in the dataset, not applied as a silent filter. After exclusions, 47 companies remained.

Prompt design

For each company, we wrote six buyer-intent prompts based on the company's own website description of what it sells. Four prompts were generic (e.g., "best CMMS maintenance management software"), and two were Utah-qualified (e.g., "best CMMS maintenance management software in Utah"). No prompt ever contained a company name. The test is whether AI engines recommend you when someone shops your category, not whether the model has heard of you.

This produced 282 distinct prompts across the cohort.

Engines tested

Each prompt was sent to four engines:

All queries used web search enabled because an ungrounded model tests memory rather than the product a buyer actually uses.

Extraction and matching

Many companies in the cohort have names that are common English words: Grain, Seer, Pattern, Artifact, Latitude, Domo, Phi, Jaunt. To avoid false positives, a fixed extraction model converted every answer into an ordered list of the vendors it recommended. Matching ran against that structured list rather than raw text. Two lookalike false positives were caught in manual review.

The run window was 31 July 2026, producing 1296 total queries. After removing cases where no AI Overview was shown (18) and extraction failures (5), 1107 usable answers remained.

The Headline Result: One in Four Are Invisible

Of the 47 companies tested, 12 never appeared in any AI answer to any prompt on any engine. That is 25.5 percent of Utah's designated AI leaders who are completely absent from AI-driven buyer discovery.

The invisible twelve are:

• Airin
• Artifact
• Bunked
• Cinco.AI
• Cynch.ai
• Drumdata.ai
• energy.work
• Harbor Health
• Jaunt
• Kahoa.ai
• Phi, Inc.
• Soar.com

What invisibility means

When a buyer asks for the category these companies serve, models do not include them. The slot does not go unfilled. Instead, models recommend competitors. For each of the twelve invisible companies, we recorded which vendors appeared in their place. The pattern is consistent: a small set of national or global alternatives captures the attention that local specialists do not.

For example, when prompts target Bunked's deepfake detection category, models recommend Reality Defender, Sensity AI, Hive Moderation, Pindrop, and Attestiv. When prompts target Cynch.ai's accounting automation category, models recommend QuickBooks Online, Xero, FreshBooks, Zoho Books, and Docyt.

What invisibility does not mean

Invisibility in this study does not mean a company is bad, obscure, or unsuccessful. Some of these companies may have strong inbound pipelines, excellent customer retention, and no need for AI discoverability. Some may sell through channels where buyer self-service is rare. The study measures one variable. Companies should weigh that variable against their actual go-to-market motion.

The binary nature of the finding

This result is not gradual. You either appear or you do not. For the 12 companies that never appear, the problem is not that they rank low. The problem is that they do not exist in the answer at all. This makes the optimisation challenge different from traditional search, where you might rank on page two and work toward page one. Here, the first step is simply getting into the response.

Engine Disagreement and Why a Single Score Misleads

If all four engines recommended the same companies, you could measure AI visibility with a single number. They do not.

Visibility by engine

ChatGPT recommends the highest share of the cohort. Claude recommends the lowest. The gap is nearly fifteen percentage points.

Partial visibility is common

13 companies (27.7 percent of the cohort) appear on at least one engine but not all four. This means more than a quarter of the companies have visibility that depends entirely on which AI assistant the buyer happens to use.

Why engines disagree

Engines differ in their retrieval sources, their summarisation logic, and their priors about which brands are authoritative. ChatGPT with web search pulls from a different index than Gemini. Claude cites gartner.com, en.wikipedia.org, and clutch.co heavily; ChatGPT cites gartner.com, linkedin.com, and learn.microsoft.com. Google AI Overview draws on youtube.com, reddit.com, and linkedin.com.

These retrieval differences mean that a company with strong presence on Clutch may appear in Claude but not ChatGPT. A company featured in a well-ranked YouTube video may appear in Google AI Overview but nowhere else.

Position also varies

When companies do appear, their position in the recommendation list differs by engine:

Google AI Overview places companies earliest, ChatGPT latest. If your company appears fourth in a list of five, that is very different from appearing first. Position matters for the same reason it matters in traditional search: attention decays as you scroll.

Utah-Qualified Prompts Lift Visibility Substantially

The study used two prompt types: generic category prompts and Utah-qualified prompts. The difference in results was pronounced.

Appearance rates by prompt type

When the prompt included "in Utah," the appearance rate rose from 29.6 percent to 44.7 percent. That is more than a fifty percent relative increase.

What this suggests

Models respond to geographic signals. When a buyer specifies Utah, models appear to weight Utah-headquartered or Utah-focused companies more heavily. This means geographic authority signals are a lever companies can pull.

The asymmetry is notable

12 companies appeared only on Utah-qualified prompts and never on generic prompts. Only 1 company appeared only on generic prompts. The asymmetry suggests that for most of the cohort, regional identity is a prerequisite for visibility, not a bonus.

Practical implications

If your company appears on Utah-qualified prompts but not generic ones, you have a narrow foothold. Buyers who add a geographic qualifier will find you; buyers who do not will not. The question becomes: what share of your actual buyers add that qualifier? If you sell primarily to Utah businesses, the foothold may be sufficient. If you sell nationally, you are invisible to most of your addressable market.

Companies in this position might consider whether their web presence, third-party mentions, and content strategy emphasise their Utah roots appropriately. For some, leaning into regional identity could be strategic. For others, it may indicate a need to build broader national authority signals.

What the Models Are Reading When They Decide Who to Recommend

When an AI engine with web search recommends a vendor, it is not drawing solely from parametric memory. It retrieves pages, reads them, and synthesises. The pages it retrieves determine who gets recommended.

Top cited sources across all engines

The 30 most-cited sources included:

• youtube.com
• gartner.com
• linkedin.com
• clutch.co
• en.wikipedia.org
• reddit.com
• g2.com
• builtin.com

Citation patterns differ by engine

Gemini's citation pattern is notably narrow, with google.com as the dominant source. This may reflect how Gemini handles retrieval attribution differently from other engines.

Why this matters for companies

If your company is not mentioned on the sources that models read, you are unlikely to be recommended. This is not about your own website. Models do not trust vendor claims about vendors. They look for third-party validation: analyst reports, review sites, Wikipedia entries, discussion forums, professional networks.

The practical question

For a company seeking visibility, the question becomes: are you mentioned on gartner.com, clutch.co, g2.com, or builtin.com? Do you have a Wikipedia page? Have authoritative LinkedIn posts or articles mentioned you? Is there a YouTube video from a credible source that reviews or features your product?

These are the surfaces that matter. A company with excellent SEO on its own domain but no presence on these third-party surfaces may rank well in traditional search but remain invisible in AI answers.

What a Company on the Wrong Side of This Should Do

If your company is among the 12 that never appeared, or among the 13 that appear inconsistently, the path forward involves understanding what models are looking for and building presence on the surfaces they read.

Step one: audit your third-party footprint

List every third-party source where your company is mentioned. Check whether you appear on:

• Analyst reports (Gartner, Forrester, IDC)
• Review aggregators (G2, Clutch, Capterra, TrustRadius)
• Wikipedia (do you have a page? is it stub or substantial?)
• Professional content (LinkedIn articles, Medium posts from credible authors)
• Discussion forums (Reddit threads where your category is discussed)
• Video content (YouTube reviews, demos, or conference talks)

If you are absent from these surfaces, models have no third-party signal to work with.

Step two: understand your category language

The prompts in this study were derived from how each company describes itself. If your website says "AI-powered revenue operations platform" but buyers search for "commission tracking software," models may never connect you to the query. Audit the language buyers actually use and ensure your third-party mentions use that language too.

Step three: build for the sources that matter

This is not a quick fix. Getting mentioned in a Gartner report requires being on Gartner's radar. Getting a Wikipedia page requires meeting notability guidelines. Getting G2 reviews requires customers willing to write them. But these are the surfaces that drive AI recommendations.

Step four: monitor over time

AI visibility is not static. Models update. Retrieval sources shift. A company invisible today may become visible if its third-party footprint grows. Conversely, a company visible today may fade if competitors build stronger signals. Periodic measurement, using the same methodology, shows whether your efforts are working.

What probably will not work

Trying to game models with keyword stuffing or manufactured content is unlikely to succeed and may backfire. Models are trained to synthesise authoritative sources, not to echo whoever repeats a phrase most often. The path to visibility runs through genuine authority, not SEO tricks adapted to a new medium.

How to Read the Published Dataset

The full dataset is published alongside this study. Here is how to use it.

Structure

Each row represents one query: one company, one prompt, one engine. Key columns include:

• Company: The AI Utah 100 company being tested
• Prompt: The buyer-intent question sent to the engine
• Prompt type: Generic or Utah-qualified
• Engine: ChatGPT, Claude, Gemini, or Google AI Overview
• Appeared: Whether the company appeared in the response
• Position: If the company appeared, its rank in the recommendation list
• Competitors mentioned: Other vendors that appeared in the same response
• Sources cited: URLs the engine cited in its response
• Excluded: Whether the company was excluded from analysis (mega-caps, universities)

Filtering for your company

Filter to your company name to see all 4 engine responses to all six of your prompts. Check which engines recommended you, which did not, and who appeared instead.

Comparing across engines

Pivot by engine to see which models recommend you most consistently. If you appear on ChatGPT but not Claude, examine what sources ChatGPT cited and consider whether your presence on those sources explains the difference.

Examining displacement

For companies with zero visibility, the dataset shows which competitors appeared in their place. This competitor list is a roadmap: these are the companies that have solved the visibility problem you have not.

Limitations to keep in mind

The dataset captures one moment in time: 31 July 2026. AI models update frequently. A company invisible in July may be visible in September. Use this data as a baseline, not a permanent verdict.

The prompts were written from company website descriptions. If a company's website does not accurately describe what buyers search for, the prompts may not reflect real buyer behaviour. The dataset includes the prompts so you can judge their relevance yourself.