85% of ChatGPT Categories Have No Owner: A 50,000-Brand Study Says Stop Chasing Domain Authority and Claim a Topic

The Opening: A 50,000-Brand Study Just Changed the Unit of Measurement for GEO

Semrush and independent researcher Kevin Indig have published an unusually large study of ChatGPT visibility: 1,094 US categories, more than 50,000 brands and 220,000 domains, observed from January through June 2026, with five prompt variants per category mapped to the real buying journey — definition, comparison, alternatives, use case and purchase decision. The conclusion is one sentence, and it dismantles the pricing basis of most GEO services sold over the past two years: only 15.2% of categories have a clear owner; the remaining 85% have no brand consistently holding the ground.

The market structure behind this is straightforward. Between 2025 and the first half of 2026, GEO went from fringe to a real market with dedicated tools, dedicated consultants and dedicated conferences — the US GEO market is projected at roughly $365 million in 2026, growing above 40% annually. But the whole market inherited SEO's scoring habit: sell the client a brand AI visibility score, one number per domain, compared month over month. The problem is that AI answers are not generated by ranking domains. They are generated by asking which topic a question belongs to and who gets mentioned repeatedly inside that topic. Wrong unit of measurement, wrong optimisation. This study is the first to quantify that at scale: domain-level metrics such as Authority Score and organic traffic predict category ownership only about half the time.

How does it compare to earlier research? Most AI-citation studies over the past year stopped at page-level hit rates — format affects citation (lists and tables beat plain prose), freshness affects citation, most citations point at someone else's domain rather than yours. This one moves up a level to ownership of the category itself: once a brand genuinely owns a topic, an owner holding a lead of five percentage points or more retains first place in 90.4% of month-over-month comparisons. AI visibility is not a die rolled fresh each day. It is sticky — hard to dislodge once someone is established, which also means the positions still vacant have an expiry date.

So what does this mean for a small business in Taiwan? Something quite direct. You assumed you were competing on domain authority against large brands, and that is a fight you lose. This study says the fight that matters is who gets mentioned consistently inside a narrow topic — and 85% of those fields are still empty. Below: what the study actually measured, what three types of readers can do today, and where the conclusion is being over-read.

The Study and the Full Numbers

Method first, because method determines how far these numbers travel:

• Sample: 1,094 US categories, 50,000-plus brands, 220,000-plus domains, drawn from Semrush's AI Visibility toolkit, January to June 2026.
• Prompt design: five prompt variants per category covering "what is X," "how does X differ from Y," "alternatives to X," "when should I use X" and "which should I buy" — deliberately mapped to a purchase journey rather than a single keyword.
• Definition of ownership: mentioned consistently across the category's prompts with a discernible lead over second place. Winning one prompt does not count.

Three core numbers:

• 15.2% — share of categories with a clear owner. Inverted: 85% of categories had no brand holding ground across a six-month observation window.
• ~50% — how often domain-level metrics (Authority Score, organic traffic) correctly predict category ownership. A coin flip.
• 90.4% — how often an owner with a five-point-plus lead keeps first place month over month.

Put together, the three yield a conclusion that is very friendly to small businesses and rather awkward for GEO tool vendors: AI visibility is a topic-level contest, most topics have not started, and once someone leads by enough, the door closes.

Immediate Actions for Three Types of Readers

Brand owners

• Retire "our AI visibility score" as a KPI. Replace it with "which three topics do we intend to own." A topic should be narrow enough that you can name the competitors and point to real work you have done.
• Self-test with five prompts. Ask your chosen topic once each as definition, comparison, alternatives, use case and purchase decision, and log which brands get named. Repeat weekly for four weeks and you have a baseline at zero cost.
• Check whether your topic already has an owner. If the same large brand appears every time, change topics. If the answer differs every time, that is an open field.

Marketing operators

• Stop planning content around domain authority. The study says it is right about half the time. Use coverage instead: can our content answer all five prompt types for this topic?
• Consolidate scattered content into topic clusters. Five to eight interlinked pieces with consistent terminology beat thirty unrelated articles.
• Track consistency, not a single hit. Ownership is defined as stable presence across prompt variants. One viral post does not give you a topic.

Developers and site owners

• Make topics visible in site structure: one hub page per topic, child pages beneath it, breadcrumbs and internal links pointing back to the hub. It is the cheapest semantic signal available.
• Complete FAQPage and Article JSON-LD, and make sure about and mentions describe the topic rather than the company name.
• Build an automated sampling script: hit the APIs weekly with a fixed set of five prompts and log the brand names that appear. That is free visibility tracking.

SaaS Comparison Table

What They Will Not Tell You

• This is US categories, English prompts, one platform. Category density, competitor count and content supply all differ in other markets. "85% vacant" is probably higher elsewhere — but so is the scarcity of prompts asked. Empty fields are not the same as traffic.
• The data comes from a company that sells AI visibility tooling. The method is published and the sample is large, but the conclusion happens to support "you need a category-level tracking tool." That does not make the numbers wrong; it means treat this as high-quality research with a position, not a neutral referee.
• "90.4% retention" has an ugly flip side: it also means displacing an incumbent is extremely expensive. For topics already owned, the correct strategy is to change topics, not fight — and no GEO agency will ever advise you to abandon a topic, because that is the same as declining the engagement.

The No-Subscription Alternative

You can build your own category visibility tracker in an afternoon at near-zero cost:

• ☐ Pick three topics. Write each as one sentence: who, in what situation, solving what. Cut anything too broad.
• ☐ Write five prompts each. Definition, comparison, alternatives, use case, purchase decision — mirroring the study design.
• ☐ Ask at a fixed time weekly. One round each on ChatGPT, Gemini and Perplexity; log every brand named into a spreadsheet, one row per run.
• ☐ After four weeks, read appearance rate, not rank. How many of 15 prompts named your brand? Going from 0/15 to 4/15 is real progress.
• ☐ Turn gaps into content. Whichever prompt never surfaces you is the piece to write next. More accurate than any keyword tool.

The only cost is 30 minutes a week. The sample is far smaller than Semrush's, but it measures the three topics you actually care about.

Frequently Asked Questions

What does "owning a topic" actually require?

The study defines it as stable presence across multiple prompts in the category, not an occasional hit. A workable operational threshold: run five prompts three times each, 15 runs total, and count appearing at least eight times as beginning to hold ground. Appearing only in the "alternatives" prompt usually means you are treated as a supporting act.

Is domain authority now useless?

Not useless — insufficient. Domain-level metrics predict category ownership only about half the time. High authority still helps with crawling and citation, but it is no longer sufficient. What decides the outcome is content consistency within the topic and the density of third-party mentions.

How many topics should a small business pursue at once?

Three at most. Ownership requires consistency across prompts, and consistency requires content density — five to eight interlinked pieces with unified terminology per topic. Running ten topics simultaneously usually means owning none of them.

Does this apply to non-English markets?

The method applies; the numbers do not transfer. Smaller markets have fewer competitors and less content supply, so the vacancy rate is probably higher, but total prompt volume is far lower. Run the same five-prompt design in your own language and build your own baseline rather than quoting 15.2%.

My Take

The mainstream reading will be "rush to claim the 85%." My judgement is different: the most important finding for a small business is not that seats are empty, but that a large brand's domain advantage is worth only half as much inside an AI answer — for the first time in a decade, specificity of content outweighs institutional scale for visibility. Small businesses historically lost SEO not because their content was worse but because their domain could not compete. When the unit of measurement becomes topic-level consistency, a thirty-person firm that has genuinely delivered twenty similar projects can write with a specificity no content farm can fake.

I do not think the window is long, though. A 90.4% retention rate says first movers hold hard, and non-English markets have far fewer topics to claim in the first place. The concrete implication for ScriptWalker: rather than selling clients an AI visibility report, sell a three-topic ownership programme — topic selection, a five-prompt baseline, content cluster planning and a four-week tracking sheet, delivered over 90 days with verifiable outcomes. A report is somebody else's number. Ownership is the client's asset.

Sources

• Primary: AI visibility is a topic-level game: A study of 50,000 brands in ChatGPT — Semrush
• Primary: How topical authority spreads (and where it doesn't) in ChatGPT — Semrush
• Primary: Does topical focus make your brand more visible? — Growth Memo (Kevin Indig)
• Third-party: ChatGPT topic ownership is rare, and SEO alone doesn't explain it — Search Engine Land
• Third-party: Does topical focus make your brand more visible? — Search Engine Land
• Third-party: Top 15 Generative Engine Optimization Platforms for 2026 — Evertune

Talk to Us About Topic Ownership

• Email: [email protected]
• Phone: 0916-224-047
• LINE: @ufv9089p