For the past eighteen months, search marketers have argued endlessly in Slack channels over whether topical authority actually carries over into LLM answers. Does ChatGPT reward brands that go deep on a niche? Or is generative AI such a blank slate that any generalist site with a massive crawl footprint can walk in cold and grab top recommendations?
Kevin Indig just settled that argument with numbers. Analyzing six months of ChatGPT answer data collected by Semrush between January and June 2026, Indig evaluated 1,094 U.S. categories, testing five distinct prompt intent types per category. The dataset spans more than 50,000 brands and over 600,000 citations.
The headline finding? Topical authority is real. Once a brand earns an outsized share of answers, it tends to hold onto it. But the far more actionable finding is tucked deeper in the data: almost nobody has actually locked down their market yet.
An astounding 89.3% of total AI search volume sits in categories where no brand holds a decisive lead. If you thought you missed the boat on generative engine visibility, the math says otherwise. The window is wide open right now.
The Myth of the Occupied AI Search Market
Every time a new platform arrives, marketers panic into believing all prime real estate gets snapped up instantly. We saw it when mobile app stores launched, we saw it in early programmatic display, and we are watching the exact same panic play out in answer engines.
Indig's data proves that category control in generative AI remains rare. Across all 1,094 categories tested, only 15.2% featured a clear category owner. Over half—53.7%—were open fields populated by multiple contenders trading mentions back and forth without a single player pulling away.
The misconception stems from watching high-profile consumer categories where household names dominate output answers. Across the broader economy, ChatGPT is still figuring out who the actual authorities are.
If your team hesitated to build out a structured approach to Generative Engine Optimization (GEO) because you assumed legacy competitors already own the answer box, you are misreading the room. Most markets are completely unsettled.
How ChatGPT Classifies Category Owners and Contenders
To bring discipline to the research, Indig classified market positions using strict thresholds across five prompt variations per category: definition, comparison, alternatives, use case, and buying decision.
A brand qualified as a Clear Owner if it appeared in at least four out of five prompt types and maintained at least a 5 percentage point lead over the second-place runner-up.
An Emerging Leader was defined as a brand that led at least three prompt types but failed to build that 5-point cushion.
Everything else fell into the Unsettled bucket—categories where no single brand could command even three out of five basic prompt variations.
Across the whole landscape, only one in six categories meets the standard of true ownership. The remaining 84.8% of categories are either undergoing early skirmishes or sitting completely unclaimed.
Why High-Volume Categories Stay Wide Open
You might expect high-value, high-volume categories to be the first ones locked down by enterprise marketing budgets. The data shows the exact opposite.
Indig ranked all 1,094 categories by estimated AI search volume and split them into two equal halves. The higher-volume half—which accounts for 98% of all AI search demand in the sample—had an ownership rate of just 11.3%. By contrast, the lower-volume half saw a 19% ownership rate.
Weighted by overall search volume, 89.3% of total AI search demand sits inside unsettled categories.
Why are the highest-value categories the least settled? High-volume topics draw a broader range of user intents, more nuanced prompt variations, and a far larger pool of competing publishers. ChatGPT synthesizes answers from thousands of indexed pages, making it harder for a single brand to dominate every query permutation without intentional topical coverage.
This volume dynamic creates massive commercial upside. The search categories worth the most money to your business are precisely the ones where incumbents are most vulnerable.
Lead Size Matters Far More Than Growth Velocity
While most categories remain unclaimed, earning true ownership pays off because AI search authority displays remarkable durability.
In month-over-month comparisons across the six-month window, established clear owners retained their top ranking in 90.4% of cases. Once ChatGPT establishes high confidence in a primary entity for a topic, it rarely rotates that entity out of its primary answer set.
However, Indig found that lead size dictates survival far more than monthly growth trajectory:
- Categories where the leader was dethroned had a median lead of just 1.3 percentage points.
- Categories where the leader held their position maintained a median lead of 2.9 percentage points.
Momentum alone proved surprisingly unreliable. The data revealed multiple cases where a leading brand grew its mention share month-over-month yet still lost first place because an agile competitor expanded faster across specific prompt types.
If your brand leads a category by only one or two percentage points, you do not have a durable moat. You are in an active dogfight.
The Citation Trap: Why Links Aren't Brand Mentions
One of the most eye-opening findings in Indig's analysis dismantles a common assumption in GEO pitch decks: the belief that getting more domain citations automatically makes you the recommended brand.
The study measured a weak negative correlation (-0.229) between citation frequency and overall brand recommendations. The most frequently cited domain in a category matched the most frequently recommended brand only 20.8% of the time.
Citations and recommendations serve distinct functions in LLM retrieval:
- Citations get you into the retrieval set. Strong technical authority scores and frequent domain links help LLM web crawlers index your site as reliable reference material.
- Brand mentions reflect topical authority. Being cited as a background reference does not mean ChatGPT will name your product when a buyer asks for vendor recommendations.
Interestingly, while top-cited domains rarely won top brand mention status, the inverse held true: the most mentioned brand was cited as a source at least once in 69.9% of queries.
Citations get your content into the model's reference window, but clear brand positioning and comprehensive prompt coverage win the customer. Teams focusing solely on link-style citations while ignoring direct brand recommendation signals—a distinction explained in our guide to winning AI search recommendations—are tracking the wrong metric.
Five Practical Steps to Claim Your Category
Building durable AI search authority requires systematic coverage across every way a buyer queries an answer engine. Drawing from Indig's framework and standard methodologies for proving AI search impact through split testing, here is how to execute:
1. Audit Share of Voice Across All Five Prompt Types
Never judge your AI visibility by testing a single brand query. For every category you target, track your share of voice across five distinct prompt structures:
- Definition: "What is [Category/Concept]?"
- Comparison: "[Brand A] vs [Brand B] for [Use Case]"
- Alternatives: "Best alternatives to [Market Leader]"
- Use Case: "How to solve [Specific Workflow Problem]"
- Buying Decision: "What is the best [Category] platform for [Company Size]?"
2. Defend Thin Leads Before Competitors Notice
If your mention tracking shows a lead under 3 percentage points, treat that category as vulnerable. Identify which prompt types your competitors win and publish dedicated comparison tables, integration guides, and structured entity data to close the gaps.
3. Shift Focus From Pure Link Volume to Entity Clarity
Citations get your site read by the model; brand clarity gets your name written in the response. Ensure your product is associated with specific attributes, key workflows, and verified customer outcomes across authoritative industry sources and review platforms.
The window highlighted in Indig's 1,094-category dataset will not remain open forever. With nearly 90% of AI search demand sitting unowned, brands that systematically build topical authority across all five prompt intent types right now will lock in multi-year leads that competitors will struggle to break.
For a deeper look at the underlying dataset, check out Search Engine Journal's analysis of Kevin Indig's AI Search Authority study.