Brand Tier Decides How Often AI Names You — 73%, 44%, 11% (2026)

Key takeaway: A new study tracked which brands AI engines name when the question contains no brand name, across more than 100,000 responses. As reported, large brands appeared in 73% of answers, mid-market brands in 44%, and small brands in 11%. The cohort is convenience-sampled by an author affiliated with a commercial measurement tool, so these are not market-wide rates. This article sets out that caveat first, then places our own Korean B2B SaaS measurement — where the gap between engines was as wide as the gap between tiers — alongside it to identify what smaller brands can actually act on.

---

Three-line summary

• Tier shapes how often you appear. On the first run of unbranded category questions, Tier 1 brands appeared in 73% of answers, Tier 2 in 44%, and Tier 3 in 11% (as reported by the study).
• Read the sample before the number. The author is affiliated with a commercial measurement tool and states plainly that the cohort is convenience-sampled with no claim to category representativeness. Read the direction of the gap, not the absolute rates.
• Tier is not the only variable. In our 12-company Korean B2B SaaS sample, ChatGPT mentioned brands 33% of the time while Perplexity and Gemini each did so 83% of the time. The same brand can land very differently depending on the engine.

What the study measured

arXiv 2606.20065 (Pratyush Kumar, submitted 2026-06-18) is a large-scale tracking study of brand visibility. It is a pre-peer-review preprint, so every figure below is as reported.

Item	Detail
Sample	102 brands · 3,508 completed tracking runs · 102,025 prompt responses
Engines	ChatGPT · Gemini · Perplexity · Claude · Grok
Period	March–May 2026
Citations	149,912
Verticals	SaaS · retail execution · fintech · Indian DTC

What matters most is the prompt type: these are unbranded category questions. Not "what do you think of our company?" but closer to "what should I use in this category?" Why that distinction changes the result is covered in why brand-name prompts can't measure AI visibility.

Three brand tiers and their appearance rates

The study sorts brands into three tiers, and publishes the criteria — which means you can locate your own company on the scale.

Tier	Criteria	Appearance rate in category answers
Tier 1	Wikipedia article over 5,000 words + 3 or more major press mentions in 12 months + public company, Series C or later, or $100M+ annual revenue	73%
Tier 2	Wikipedia article + Series B or later, or top 3 in category	44%
Tier 3	Real web presence but below the above thresholds	11%

"Tier 1 brands appear in 73% of unbranded category answers on the first run, Tier 2 in 44%, and Tier 3 in just 11%" — Kumar, arXiv 2606.20065

Two things travel with that sentence.

First, it is a first-run figure. AI answers vary from run to run, so a single result is one point in a distribution — our own measurement found run-to-run variance larger than the month-over-month change.

Second, the tier criteria are themselves the work items. Wikipedia depth, press mentions, funding stage — none of these live inside your content. However well a page is optimized, the ceiling on appearance rate appears to be set outside it.

Why these numbers should not be read as market-wide

The author is affiliated with a commercial measurement tool, and the data is that tool's tracking. The limitations section says so directly.

"Convenience-sampled cohort. The brands skew toward SaaS, retail-execution, fintech, and Indian DTC, so we do not claim category representativeness" — Kumar, arXiv 2606.20065

A convenience sample means the 102 tracked brands were not drawn at random. They may well be companies already paying attention to AI visibility, which skews the cohort toward brands that already invest in it. The verticals cluster in four areas.

So do not carry 73/44/11 into your own vertical. The safe reading is that the gap between tiers is large, and that the variables driving it sit outside content optimization.

If 89% of demand is unowned, why do small brands appear only 11% of the time?

Placed next to another study, this creates an apparent contradiction. A separate analysis tracking 1,094 US categories found that an estimated 89.3% of AI search demand sat in categories with no clear owning brand (Growth Memo, 2026-07; our write-up: 89% of AI search demand has no clear owner).

Most seats are open, yet small brands appear only 11% of the time. The two figures measure different things.

• 89.3% means no fixed leader in the category. It describes rankings that have not settled.
• 11% is how often a small brand shows up at all in an answer.

Put together: the leading seat is usually vacant, but small brands reach the candidate list infrequently in the first place. An open contest is not the same as an easy entry.

The variable our own measurement adds — the engine

The study pooled five engines to produce its per-tier rates. When we measured 12 Korean B2B SaaS companies in July 2026, the spread between engines was as wide as the spread between tiers.

Engine	Brand mention rate (12-company Korean B2B SaaS sample)
Perplexity	83%
Gemini	83%
ChatGPT	33%

(RanketAI July 2026 benchmark; our write-up: Korean B2B SaaS AI visibility measured)

Twelve companies is a small sample, used here for direction only, and the two studies differ in sample, prompts and period, so we do not compare the numbers directly. The implication is still clear: the same brand can differ by a factor of 2.5 depending on which engine measures it. Moving up a tier takes years. Identifying your weakest engine and reinforcing the sources that engine draws on is a quarterly job.

Why engines construct answers differently is covered in measuring brand visibility in Gemini, the mention-first engine.

Where the citations go — 78%

In the same study, about 78% of the 149,912 citations went to corporate websites, counting both brand-owned pages and third-party pages.

"about 78% of those citations go to corporate websites, including the brand's own pages and third-party pages" — Kumar, arXiv 2606.20065

That figure argues against treating your own site as irrelevant. It does not, however, support narrowing the work to "fix our own pages," because third-party corporate pages sit inside the same denominator. On the content-format side, our earlier analysis put ranked listicles at 43.8% of 26,283 ChatGPT citation URLs (self-promotion and listicle citations), while this study reports roughly 21% for the same category. Different samples and engine mixes make the two values non-comparable; what they share is the direction — format accounts for a large share of citations.

Sentiment flips 6.7 times more often than mention

The study reports that how a brand is described is far less stable than whether it is described at all.

"Sentiment is the unstable part...flips about 6.7 times more often than whether the brand is mentioned at all" — Kumar, arXiv 2606.20065

The practical consequence lands in measurement design. Mention rate stabilizes over relatively few runs, but positive-versus-negative framing read from one or two runs will turn noise into a signal. To use sentiment as a metric, collect more runs and read the trend.

What a smaller brand can start on now

Inverting the tier criteria produces an order of operations.

Priority	Action	Why
1	Identify your weakest engine	The engine gap can rival the tier gap. Look at the lowest engine, not the average across engines
2	Earn third-party pages	78% of citations go to corporate sites, third-party pages included. Owned pages alone have a ceiling
3	Build tier signals — press, encyclopedic coverage	The tier criteria are the work items, but treat them as a multi-quarter track
4	Collect enough runs	First-run figures are one point in a distribution, and sentiment especially needs volume

Whether your own pages are in a state AI can read and extract is checkable with Page Structure Diagnostics, and per-engine mention and citation status can be compared in AI Brand Visibility Analysis.

What item 2 actually asks you to do

"Earn third-party pages" is the kind of phrase that never turns into work. The list is not what drives action — the classification is. Collect the sources that cite your competitors but never you, then sort them by whether you can realistically get in. Here is how 15 such sources broke down in one comparison run we conducted in the Korean AI marketing SaaS category in July 2026.

Class	Share	Examples	What you do
Directly publishable (UGC)	3 of 15 (20%)	Naver Blog, Brunch, WikiDocs	Open an account and publish a version rewritten for that channel. Zero cost, startable this week
Reachable via press or contributed articles	5 of 15 (33%)	Business press, startup media, industry blogs	Package a launch or a research finding as a press release or byline. One placement often pulls portal syndication with it, closing two gaps at once
Structurally closed	7 of 15 (47%)	Other companies' blogs, government and research bodies	Drop them from the target list

A single measurement in a single category, so the proportions do not generalize. The direction does: roughly half the gap list is somewhere you cannot get into, and working from an unsorted list spends effort in the wrong place. The four-step procedure — identify, classify, act, re-measure — is laid out in source gap analysis: finding the sources that cite only your competitors.

Frequently asked questions

We're Tier 3. Is it hopeless?

11% is an appearance frequency, not a ceiling. It also comes from a convenience sample measured on first runs. Broken out by engine the spread is wide — our own measurement found a 2.5x difference between engines. Rather than reading the pooled average as a verdict, identify the engine where you rank lowest and reinforce the sources that engine cites.

Should we work on tier signals or on content first?

They run on different clocks. Press coverage, encyclopedic entries and funding stage take quarters or longer and are not fully under marketing's control. Structural work on your own pages and earning third-party coverage can start within a week. Begin with what you can start, and keep tier signals as a parallel track.

With a small budget, where do we look first?

In order of cost: identify per-engine mention rates and find the weakest engine; check which source types that engine actually cites; then fix what you control on your own pages. Tier signals come after that.

Can we apply 73/44/11 to our own vertical?

Not advisable. The author flags the convenience sample, and the verticals cluster in SaaS, retail execution, fintech and Indian DTC. Your own numbers have to be measured. Use this study to confirm the structure — that tier gaps are large and driven by variables outside content — rather than to predict your rate.

Should we ignore sentiment as a metric entirely?

No. It works once you have the runs. Given the report that it flips about 6.7 times more often than mention, avoid judging from a single result and read a series collected under the same conditions.

In closing

The contribution here is not the absolute rates. It is the structure: brand tier stratifies how often AI names you, and most of what determines that tier sits outside content optimization. The convenience-sample caveat holds, and in our own measurement the engine turned out to be a variable as large as tier.

For a smaller brand, the actionable conclusion is this. Treat moving up a tier as the long track, and first measure where you stand engine by engine — then reinforce the sources that engine draws on.

Further reading

• What is GEO? Definition, how it works, and 7 verified methods — the field's definition and evidence base.
• 89% of AI search demand has no clear owner — category ownership structure.
• Korean B2B SaaS AI visibility measured — the basis for the per-engine gap.
• Why brand-name prompts can't measure AI visibility — category versus brand queries.
• The illusion of month-over-month gains — run variance and measurement design.
• Source gap analysis — the four-step procedure for earning third-party citations.