What 113 Prompts Revealed About One Brand's AI Visibility

Why we ran this

Most "AI visibility" claims are assertions. We wanted a measured one.

So we took a single brand, a mid-market inventory-management platform, and asked a simple question: across the AI tools its buyers actually use, how often does it get named, on which questions, and why. Not a survey. Not an estimate. 113 prompts, run and recorded across six platforms, scored against eight competitors.

This is the first published experiment from the Slate AI Lab. The brand and competitors are anonymized. Everything else is exactly what we found.

Method

• 113 prompts, written to mirror how real buyers ask, spread across five intent categories: discovery ("best inventory software for a small business"), comparison ("X vs Y"), technical ("how to track inventory with barcodes"), trust ("is X reliable"), and pricing.
• Six platforms: ChatGPT, Perplexity, Gemini, Claude, Google AI Overviews and Google's standard results.
• Eight competitors tracked in parallel, so share of voice was measured against the real field, not in isolation.
• For each answer we recorded whether the brand appeared, how prominently, which competitors appeared, and which sources the model cited.
• We also scored the brand's own pages, and its strongest competitor's, on four factors: AI visibility, extractability, entity clarity, and schema.

A methodology note, stated plainly: this is one brand, in one category, at one point in time, scored with our own rubric. It is a deep case study, not an industry benchmark. Treat the lessons as directional.

Finding 1: Leading the category is not the same as winning

The brand had the highest share of voice in its category, at 40%. The nearest competitor sat at 24%, the next at 17%. On the headline number, it was winning.

Then we looked at the distribution. Of the 113 prompts, 52 showed healthy visibility, 21 needed work, and 40 were critical, meaning the brand was rarely or never mentioned. That is 35% of the questions its buyers ask, on which it was effectively invisible, sitting underneath a category-leading average.

This is why we score prominence, not just presence. A strong average can hide the fact that you are absent from a third of the conversation. The average flatters. The gaps are where deals are lost.

Finding 2: The brand was missing on exactly the questions that convert

The 40 critical prompts were not random. They clustered in comparison, technical and trust intent, and they were among the highest-purchase-intent questions in the set. Some concrete examples of where the brand scored near zero:

• "Inventory software that integrates with QuickBooks" — 0%
• "How to prevent inventory loss" — 0%
• "How to track inventory using barcodes" — 17%
• "How to track company assets across locations" — 17%

By contrast, the brand performed strongly on broad discovery questions like "best inventory software". In other words, it showed up when someone was browsing, and disappeared when someone was deciding. Competitors filled the gap because they had published the comparison and workflow content that answers those specific questions, and the brand had not.

Finding 3: 94% of its citations came from off its own site

We logged 2,373 individual citations across the answers, drawn from more than 1,200 unique domains. The breakdown of where those citations came from:

• Third-party and neutral sources: 66%
• Competitor websites: 12%
• Social and community platforms: 8%
• The brand's own website: 6%
• Editorial publications: 5%
• User-generated content: 3%

Read that again. When AI described this brand, it was quoting the brand's own site 6% of the time. The other 94% came from places the brand does not control, including, at 12%, its competitors' own websites.

The single most-cited domain in the whole study was not the brand or any competitor. It was YouTube. Product walkthroughs and comparison videos were referenced repeatedly when models formed recommendations, and the brand was largely absent from them. This is generative engine optimization in one statistic: the material that decides your reputation mostly is not on your website.

Finding 4: The same brand scored very differently on each platform

Visibility was not uniform. The brand appeared most often on Gemini and Google's AI surfaces, and least often on ChatGPT and Claude.

The reason was structural. Google's surfaces lean on their own index and reward the structured product information the brand did have. ChatGPT and Claude lean more heavily on third-party comparison articles, community discussion and editorial coverage, which is exactly the off-site material the brand was missing. One brand, one set of facts, and a large swing in visibility depending on where each platform looks. A single-platform strategy would have completely misread this brand's position.

Finding 5: The competitor won on structure, not on being better

We scored the brand's pages and its strongest competitor's on the same rubric. The brand's site averaged 43 out of 100. The competitor's averaged 86, almost exactly double.

The gap was not quality of product or even quality of writing. It was structure and coverage:

• The brand's blog content scored lowest of any page type on extractability. The answers were there, buried in prose no model could lift cleanly.
• The brand had no comparison pages. The competitor had several, and they scored highest of all.
• The competitor had real documentation and consistent schema; the brand's was thin.

This is the answer engine optimization lesson stated as data. The brand did not have a content problem. It had an extraction and coverage problem. The competitor was not being recommended because it was better. It was being recommended because it had built the structured, comparison-shaped content that models can quote, and the brand had left that space empty.

What this means for any brand

Strip out the specifics and five things generalize:

• Your category-average visibility can be strong while you are invisible on the questions that matter. Measure prominence on high-intent prompts, not just an overall score.
• Most of what AI says about you is sourced off your own site. If you only work on your website, you are working on a minority of the problem.
• Comparison and workflow content is where purchases are decided, and where most brands are weakest. It is also the most buildable gap.
• Extractability beats volume. Content that is not structured to be quoted does not get quoted, however good it is.
• Platforms disagree. Being strong on Google's surfaces tells you almost nothing about ChatGPT or Claude.

None of these gaps were expensive to identify. All of them were invisible to the brand's existing analytics, because rankings and traffic do not measure any of them.

Limitations

One brand, one category, one moment in time, one scoring rubric. AI answers shift week to week, so a re-run would not reproduce these numbers exactly. This is a case study designed to show what the questions and the method reveal, not a statistical claim about all brands. We are publishing the approach as much as the result, and we will run it again.