The Engine That Never Names You: 55 of 248 Tracked Brands Have a GEO Blind Spot on One AI Engine, and ChatGPT Is the Usual Culprit

Your Generative Engine Optimization (GEO) dashboard says your brand holds 18% of AI mentions in its category. That sounds respectable, and the blended number may be accurate. But 18% can mean "named about equally often by ChatGPT, Gemini and Claude" or it can mean "named often by two engines and almost never by the third". The second case calls for a different fix, and a blended score cannot tell you which one you are in.

We have looked at engine splits at the category level before (see per-engine divergence). This piece goes down to the brand level and asks a narrower, more practical question: how common is a single-engine blind spot, which engine causes most of them, and what do they cost in blended points?

The Data: 49 Category Pages, 248 Brands

AIOPulse, a Japanese GEO monitoring tool, publishes a ranking page for each category it tracks, showing every brand's mention share for ChatGPT, Gemini and Claude separately, plus a blended figure. Its rankings index links 60 pages. I read all of them on 10 November 2026 (reference date 9 November; the window is 8 or 9 days, although the pages label it "last 7 days"). The prompts are brand-neutral questions in Japanese, so every brand named is one the engine volunteered. "Share" is a brand's slice of all mentions of tracked brands, not the share of answers that name it.

I kept 49 pages and dropped 11:

• Ten older pages built on only 3 questions and 48 to 51 answers each, too thin to split three ways.
• The AI-monitoring-tools page. On ChatGPT and Gemini the engines named almost none of the tracked brands (ChatGPT's column sums to 25%, Gemini's is a single brand at 100%), so a zero there means "empty answer", not "rejected brand". I come back to this in the disclosure, because AIOPulse itself sits on that page.

The 49 pages hold 248 brand entries on lists of two to ten brands, with 85 to 272 answers per page. On a typical page ChatGPT and Gemini each contribute 4 sampling days and Claude 8 or 9, so a 10-question page rests on roughly 40 ChatGPT answers and 40 Gemini answers. The blended figure is the plain average of the three engines.

The Rule: What Counts as a Blind Spot

I flagged a brand on an engine when two conditions held: the other two engines averaged at least 10% for that brand, and the engine in question gave it less than one third of that average. A brand at 12% and 15% on two engines and 3% on the third is flagged; a brand at 4%, 5% and 1% is not, because there is not enough to lose. 191 of the 248 brands pass the 10% test on at least one engine.

What the Counts Show

Engine	Brand-engine pairs tested	Blind spots	Rate	Of which exactly 0%
ChatGPT	172	38	22.1%	13
Gemini	169	16	9.5%	3
Claude	168	5	3.0%	1
Total		59		17

The 59 cases involve 55 distinct brands, 22% of everything tracked and 29% of the 191 that could qualify, and they appear on 35 of the 49 pages. ChatGPT accounts for about two thirds of them. In 33 of its 38 cases both Claude and Gemini give the brand at least 10%, so ChatGPT is the outlier against two engines that agree with each other. Claude rarely leaves a brand out in this data set.

Examples (shares in %, blended = the average of the three):

Category (brands tracked)	Brand	ChatGPT	Claude	Gemini	Blended	Blended rank
Career-change support (3)	Hataractive	0.0	36.3	44.7	27.0	2 of 3
Flights (4)	Trip.com	0.0	38.5	36.0	24.8	2 of 4
Diet apps (3)	FiNC	0.0	26.7	34.9	20.5	2 of 3
Project management (6)	Backlog	0.0	18.2	18.0	12.1	5 of 6
CRM (4)	kintone	5.2	24.6	23.7	17.8	4 of 4
Home purchase (4)	Mitsui no Rehouse	10.3	40.5	36.5	29.1	2 of 4
Life insurance (5)	Lifenet Insurance	4.8	32.9	25.5	21.1	3 of 5
Mobile plans (8)	ahamo	2.7	14.7	17.2	11.5	4 of 8
Mobile plans (8)	LINEMO	1.6	15.5	16.6	11.2	5 of 8
Mobile plans (8)	povo	2.2	14.9	10.5	9.2	6 of 8
Home-loan banks (5)	Mizuho Bank	27.1	9.0	1.8	12.6	5 of 5
AI translation (10)	Microsoft Translator	28.9	3.2	3.1	11.7	3 of 10
AI meeting notes (10)	Otter.ai	75.7	9.1	2.0	28.9	1 of 10

The last rows show the mirror image. Otter.ai leads its page on the blended number, yet 75.7% of ChatGPT's tracked mentions go to it while Gemini gives it 2.0%. A brand can be first on the blended table and invisible on an engine at the same time. The three mobile plans (ahamo, LINEMO, povo) are all online-only sub-brands of big carriers, and all three are close to zero on ChatGPT while Claude and Gemini give each of them 10% to 17%.

What a Blind Spot Costs in Blended Points

If the blind engine matched the average of the other two, the blended score would rise by one third of the gap. Across the 59 cases the median is 5.2 points (mean 6.2, maximum 13.5), and 16 of the 59 are worth 8 points or more. Holding every other brand's number fixed, which is a rough approximation because shares on an engine sum to about 100%, the brand would climb at least one place in the blended table in 24 of the 59 cases. In a category where the leader holds 25% to 30%, five points is a large amount of ground, and it is the cheapest ground to reclaim because two engines already show that the brand is recommendable.

Why ChatGPT? Three Hypotheses, None Proven

• What is actually being queried. According to the FAQ on AIOPulse's homepage (read 10 November), the ChatGPT runs use gpt-4o-mini without web search, while the Gemini and Claude runs use gemini-3.1-flash-lite and claude-haiku-4-5, both with web search on. An engine answering from memory alone will miss brands that its training data covers thinly, which includes many domestic Japanese brands. If that is the mechanism, the blind spot is a property of this measurement setup and may not describe the consumer ChatGPT app. That is the single most important thing to find out about any tracker you use.
• Brand-family confusion. On the mobile-plan page ChatGPT's top three are Rakuten Mobile, IIJmio and mineo (30.1%, 26.9%, 22.0%), while the three online-only carrier sub-brands barely register. That fits a story in which the model recommends the parent brand or independent operators instead, but the data cannot confirm it.
• Name matching. A tracker counts a brand when it finds a name or alias in the answer text. If a model writes a brand in a form the tracker does not recognise, the result is an engine-specific zero that has nothing to do with visibility.

Three Checks Before You Act on a Zero

1. Is it sampling noise? Over about 40 answers, a brand that a model names in 10% of its answers is missed completely 1.5% of the time (0.9 to the power of 40); at 5% it is missed 13% of the time. A zero on a brand that other engines name at 10% or more is unlikely to be luck. A single-digit share on a 3-question page is no evidence at all. Our notes on reporting windows and list size explain why small samples mislead.
2. Read the raw answers on that engine. Ask the ten buyer questions yourself, or open your tool's raw responses. Does the model name your parent company, a product under a different spelling, or a competitor? This separates a matching problem from a knowledge problem in an afternoon.
3. Check rate as well as share. A brand can be named in many answers and still have a small share if the engine lists many brands per answer. Our piece on share versus mention rate shows how to tell the two apart.

What to Do If the Blind Spot Is Real

• Report the weakest engine next to the blended number. One extra line in the weekly report, such as "lowest engine: ChatGPT, 2.7% versus 15.9% elsewhere", turns a hidden problem into a tracked one.
• Prioritise by engine type. If an engine answers from memory, changes to your site matter only when the next model version learns them, so effort is better spent on being described consistently across third-party sources. If it uses web search, pages that engine can retrieve may move the number sooner. This follows from how these systems generally work and was not tested here.
• Make the entity unambiguous. For sub-brands, state in plain text which company runs the plan, what it is called in both scripts, and what category it competes in. The mobile-plan case suggests parent-brand confusion is worth ruling out.
• Weigh by where your buyers ask. If your customers mostly use one engine, that engine's number is your number, and a blended score can mislead you in either direction.

Where the Tools Differ

Whatever you use, ask three questions: can I see each engine separately, does it query APIs or the consumer apps, and is web search on or off? Profound, Peec AI and Otterly.ai let you set your own prompts and brands, which is how you would run this analysis on your own category; check each vendor's current documentation for how its engines are sampled, because I did not audit them. Mieruca GEO is the better-known broader SEO suite in Japan. AIOPulse publishes per-engine tables free on its public pages (read its explainer for how it defines AIO) but only for lists the publisher chose; brands outside them are "unmeasured". Per the pricing section of its homepage, the site tracking plan is JPY 9,800 per site per month excluding tax with five keywords included and a first payment of JPY 980 excluding tax. There is no free trial, though there is a free diagnosis (a quick check that asks only ChatGPT, delivered by email) and a sample report. Contracts are monthly with no minimum term and no prorated refunds. The page says each keyword sends 5 questions to 3 AIs daily, which means about 35 answers per engine per keyword per week; the public ranking pages are sampled less often (every other day for ChatGPT and Gemini), which is why those engines have fewer days behind them here.

Limits You Should Weigh

• One week, one snapshot. The trend pages show blended values only, so I could not test whether these blind spots persist over 30 days. Some will not.
• Thin engine samples. ChatGPT and Gemini rest on 4 sampling days each. Treat the 13 exact zeros on ChatGPT as leads to investigate, not diagnoses.
• Arbitrary thresholds. One third and 10% are my choices. Moving them changes the counts; the ranking of engines (ChatGPT most, Claude least) is the part I would trust.
• Small lists. Shares are relative to a list of two to ten brands the publisher picked. On a three-brand page a brand's share can move a lot because a competitor was added or removed.
• Japanese only, small models. These are Japanese-language prompts to specific model versions. Do not assume the same engine ranking in English or in the consumer apps; see our piece on cross-language visibility.

FAQ

Is a blended GEO score useless, then?

No. It is a good summary when the engines agree. It becomes misleading when they do not, so look at the spread between the highest and lowest engine before trusting it.

Should I optimise for ChatGPT first because it shows most blind spots?

Only if your buyers use it and the check against raw answers confirms the gap. In this data the ChatGPT runs also differ in method (no web search), so the pattern may say more about the setup than about ChatGPT.

How do I run this on my own category?

Export per-engine shares for every brand in your list, compute the average of the other two engines for each, and flag any brand under one third of it with an average of at least 10%. Do it in a spreadsheet in ten minutes.

Takeaway

In 49 Japanese-language category pages, 55 of 248 tracked brands were nearly absent from one engine while two others named them, and ChatGPT was that engine in 38 of 59 cases. The median cost is about five blended points. Before you spend on content, put the lowest-engine number next to the blended one, rule out sampling and name-matching, and find out how your tool queries each engine.

Disclosure: AIOPulse is developed by Marketing Cloud Lab, which also operates this site. All AIOPulse figures here were retrieved on 10 November 2026 from its public ranking pages and homepage; pages update daily, so later visits will differ. The blind-spot rule, counts, gaps and rank effects are the author's calculations from the published per-engine tables. AIOPulse itself appears on the page I excluded: there it shows 0% on ChatGPT and Gemini (where almost no tracked brand was named) and 25.0% and first place on Claude, which is exactly the kind of split this article describes. Competitor descriptions for Profound, Peec AI, Otterly.ai and Mieruca GEO reflect general public information as of November 2026 and were not independently audited.