When the Competitor Is the Assistant Itself: ChatGPT, Claude and Gemini Hold 26–30% of GEO Mention Share in Three AI-Tool Categories

If you sell an AI writing, translation or presentation tool, your Generative Engine Optimization (GEO) report probably compares you with other specialist vendors. But ask ChatGPT, Gemini or Claude for the best tool in your category and a fourth group of names shows up in the answer: the assistants themselves. "Just use ChatGPT" is a legitimate recommendation, and when the assistant is listed as a brand, it competes for the same slots you do.

How big is that group, and does it behave the same way on every engine? I checked four AI-tool categories on a public Japanese GEO tracker. The answer to the first question is "bigger than most specialist vendors". The answer to the second is "no, and the differences are large enough to change how you read your own number".

The Data: Four AI-Tool Category Pages

AIOPulse, a Japanese GEO monitoring tool, publishes a ranking page per category with each brand's mention share on ChatGPT, Gemini and Claude plus a blended figure (an equal-weight average of the three). Its rankings index covers 60 categories and 312 tracked brands. Four of those categories are AI tools where the publisher chose to put general-purpose assistants on the tracked list, so they are the right place to look. I read all four on 13 November 2026:

Category page	Brands tracked	Answers in window	Window	Days with data (ChatGPT / Gemini / Claude)
AI SEO and writing tools	18	180	5 to 13 Nov	5 / 5 / 8
AI translation	17	190	5 to 13 Nov	5 / 5 / 9
AI presentation tools	16	190	5 to 13 Nov	5 / 5 / 9
AI meeting notes	18	160	5 to 12 Nov	4 / 4 / 8

The prompts are brand-neutral questions in Japanese, so any brand in a table is one the engine volunteered. "Share" means a brand's slice of all mentions of tracked brands, not the share of answers that name it (see why the list size matters). Each table shows only its top 10 per engine, so where a brand falls outside the top 10 I give an upper bound rather than a number. I counted ChatGPT, Claude, Gemini and Copilot as general-purpose assistants wherever they are on a page's list, and left out Genspark and Kimi, which are borderline.

Finding 1: Generalists Take a Quarter or More in Three of Four Categories

Category	General-purpose assistants on the list (blended share, rank)	Combined	Best specialist
AI SEO and writing	ChatGPT 13.7 (1st), Claude 9.7 (3rd), Gemini 6.7 (8th)	30.1%	Jasper 13.4 (2nd)
AI presentation	Copilot 13.5 (4th), ChatGPT 10.3 (5th), Gemini 3.9 (6th)	27.7% or more	Canva 29.8 (1st)
AI translation	ChatGPT 13.9 (3rd), Claude 6.4 (5th), Gemini 5.3 (6th)	25.6% or more	DeepL 27.5 (1st)
AI meeting notes	Copilot 3.6 (7th), Gemini 3.4 (8th), ChatGPT 2.2 (10th)	9.2%	Otter.ai 29.2 (1st)

"Or more" marks pages where another assistant is on the list but too small to appear in the top 10 (Claude is not on the presentation or meeting-notes lists at all). Three things stand out:

• In AI writing, the assistants beat every specialist. ChatGPT alone is first overall, ahead of Jasper by 0.3 points, and the three assistants together hold more than twice Jasper's share.
• In presentations, Copilot and ChatGPT outrank every specialist except Canva, Gamma and Beautiful.ai. Together with Gemini they roughly match Canva's 29.8%.
• Meeting notes is the exception. The category has a well-known dedicated leader, and the assistants hold only 9.2% combined. A concentrated category (see our concentration piece) seems to leave less room for "just ask the assistant", although four categories are too few to call that a rule.

Finding 2: The Three Engines Treat Assistants Very Differently

The blended figures hide a split. Here is the combined share of the assistants on each engine's own table:

Engine run	AI writing	Translation	Presentation	Meeting notes
ChatGPT	3.6 to 4.8%	0%	13.5%	0%
Claude	48.6%	51.5%	34.3%	18.9% or more
Gemini	38.3%	27.2%	35.4%	7.9% or more

On Claude, about half of all tracked-brand mentions in writing and translation go to general-purpose assistants. On ChatGPT, the figure is close to zero in three of four categories. A blended "30%" in AI writing is therefore really "about 4%, 49% and 38%", the same pattern our per-engine piece found for brands in general.

Finding 3: Engines Do Not Reliably Favour Themselves

One might expect each assistant to name itself generously. The data says otherwise. Each cell below is the share one assistant receives on one engine's table; the engine's own name is in bold.

Category and assistant	On ChatGPT's table	On Claude's table	On Gemini's table
AI writing: ChatGPT	3.6	17.4	20.1
AI writing: Claude	0.6 or less	16.4	12.8
AI writing: Gemini	0.6 or less	14.8	5.4
Translation: ChatGPT	0	19.8	21.8
Translation: Claude	0	17.4	1.8
Translation: Gemini	0	12.9	3.0
Presentation: ChatGPT	8.7	7.5	14.6
Presentation: Gemini	0	9.9	1.9
Meeting notes: ChatGPT	0	2.8 or less	5.9
Meeting notes: Gemini	0	10.1	1.9 or less

• Gemini under-names Gemini. In all four categories the Claude run names Gemini more often than the Gemini run does (14.8 against 5.4, 12.9 against 3.0, 9.9 against 1.9, 10.1 against at most 1.9).
• ChatGPT under-names ChatGPT. In every category the Gemini run names ChatGPT more than the ChatGPT run does (20.1 against 3.6, 21.8 against 0, 14.6 against 8.7, 5.9 against 0).
• Claude is the exception. It names Claude at 16.4% and 17.4% in the two categories where Claude is tracked. In translation, the Gemini run gives Claude 1.8%, so the Claude run names itself about ten times as often as Gemini does.

So the working assumption "assistants promote themselves" fails for two of the three. What I cannot tell is why. Three hypotheses, none proven:

• Setup differences. According to the FAQ on AIOPulse's homepage (read 13 November), the ChatGPT runs use gpt-4o-mini with web search off, while the Gemini runs use gemini-3.1-flash-lite and the Claude runs use claude-haiku-4-5, both with web search on, all through each vendor's API rather than the consumer app. A model answering from memory may recommend established specialist products; a model with search may pull in articles that list ChatGPT and Claude as alternatives. If so, this is a property of the measurement, not of the products.
• Self-reference in text. A model asked for recommendations may write "I can help with this" instead of its own brand name. The tracker counts names, so it would record a zero for a recommendation a human reader would see.
• Category framing. The questions do not name brands, and AIOPulse does not publish their wording. If a question says "tools" rather than "AI assistants", an engine may be reluctant to list itself.

What It Does to Your Share

Mention share is relative to the list, so whether generalists are on the list changes your number. In AI writing, Jasper's blended share is 13.4% with the three assistants included. Re-basing each engine's table without them (my approximation, using the published per-engine figures) puts Jasper near 18%: about 22% on ChatGPT, 16% on Claude and 17% on Gemini. Neither figure is wrong. They answer different questions: "how much of the AI's talk about this category is about me?" versus "how often do I beat the other specialists?". Our piece on category definition shows the same sensitivity from another direction. AIOPulse's own explainer on what AIO/GEO is argues that AI answers define competitors by the job to be done rather than by industry. That is the logic behind adding assistants: if the buyer's job is "get a translation", ChatGPT is a candidate for that job.

What to Do With This

1. Add the assistants to your tracked list. If buyers could do the job with a prompt (writing, translation, summaries, slides, basic analysis), put ChatGPT, Claude, Gemini and Copilot in as competitors. Without them, your report measures a world in which a quarter of the answer does not exist.
2. Report two shares. Share of the full list, and share among dedicated tools only. The gap between them is your "generalist tax" in that category.
3. Read the raw answers. When an assistant is named, is it named as an alternative to you, as a step in a workflow ("draft in ChatGPT, finish in Tool X"), or as the whole answer? A 30% share can mean very different things. Also check whether a model refers to itself without using its name.
4. Say what a prompt cannot do. Answers draw on what is written about you. Where a specialist beats a chat window (workflow, data you hold, compliance, integrations, team features), say so in plain text on your own pages and in third-party coverage. This is inference from how these systems generally work; I did not test it.
5. Weigh by the engine your buyers use. If they live in ChatGPT, the 0% to 13.5% column is the relevant one. If they live in Claude, half the answer may be assistants.

Where the Tools Differ

The practical question for any tool is whether you can define your own competitor list and whether it reports per engine. Profound, Peec AI and Otterly.ai let you set your own brands and prompts, which is how you would add assistants to your own category; check each vendor's current documentation for details, because I did not audit them. Mieruca GEO is the better-known broader SEO suite in Japan. AIOPulse's public pages show per-engine tables for free, but only for lists the publisher chose, and not every list includes the same assistants (Claude is absent from two of the four above; its own AI-monitoring category tracks no assistants at all, so that page cannot show this effect). Per the pricing section of its homepage (read 13 November), 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 that asks ChatGPT only. That matters here: ChatGPT is the engine that names assistants least, so a ChatGPT-only check would understate this effect. Contracts are monthly with no minimum term and no prorated refunds.

Limits You Should Weigh

• One short window. Four to nine days, with only 4 or 5 sampling days for ChatGPT and Gemini. Treat the zeros as leads, not diagnoses.
• Top 10 only. Where a brand is absent from a table I use an upper bound, so combined figures marked "or more" are floors.
• Lists differ by page. Copilot, Claude and others are on some lists and not others, so combined shares are not strictly comparable across categories.
• Japanese prompts, API models. These are small models answering Japanese questions through APIs. Do not assume the same pattern in English or in the consumer apps (see the blind-spot piece for the same caution).
• Reference-day versus window figures. The pages quote a headline for one reference day and a table averaged over the window. I used the window averages.

FAQ

Does a 30% generalist share mean ChatGPT is winning my customers?

Not necessarily. A mention can be "use ChatGPT for the first draft, then polish with X". The share shows how often an assistant appears in the answer, not whether it replaces you. The raw answers will tell you which.

Should I optimise to be named instead of the assistants?

You cannot remove them from the answer. You can make sure the answer explains why a specialist is worth choosing, and that you appear beside them rather than behind them.

How do I check this for my own category?

Run your buyer questions with brand-free wording on each engine, add the assistants to your brand list, and compute your share twice, with and without them. A spreadsheet is enough.

Takeaway

In three of four AI-tool categories, general-purpose assistants take 26% to 30% of tracked mentions, and in AI writing ChatGPT is first overall. The engines split sharply: Claude gives assistants about half its mentions in two categories, while the ChatGPT run gives them almost none in three. Add the assistants to your list, report your share with and without them, and find out how your tool queries each engine before reading a blended number.

Disclosure: AIOPulse is developed by Marketing Cloud Lab, which also operates this site. All AIOPulse figures here were retrieved on 13 November 2026 from its public ranking pages and homepage; pages update daily, so later visits will differ. The category totals, per-engine sums, upper bounds and the re-based Jasper estimate are the author's calculations from the published tables. Competitor descriptions for Profound, Peec AI, Otterly.ai and Mieruca GEO reflect general public information as of November 2026 and were not independently audited.