← AP News
NO. 25 · Insight

ChatGPT's Web Share Is Rising Again — What 56.4% Tells Us, and What This Number Does Not Count

Translated from the Korean original by AI; native-speaker review pending.

In Similarweb's September count, ChatGPT's web share rose to 56.4% and Gemini's fell to 25.0%. We separated what the number actually counts, why it differs from the user figures companies announce, and what we have not yet been able to confirm.

This article was written with AI (Claude) and reviewed by an editor. Claude is among the services compared here, and Anthropic, the company that makes it, is the maker of the Claude we used to write this article, so we have a conflict of interest. We therefore make no judgment about which service is better, and we give the source and the measurement method alongside each figure. User counts that companies stated are marked as "what the company announced," and figures from research firms are marked as "estimates." This is not investment advice on any particular company.

In mid-October, several outlets reported the September count from the research firm Similarweb: of worldwide visits to AI chatbot websites, ChatGPT had 56.4%, Gemini 25.0%, and Claude 9.6%. In the previous article (NO.16) we covered why the numbers differ from survey to survey. Now that a new month's figures are out, this time we start by reading what exactly this number counts.

What the numbers say (research-firm estimates)

Below are Similarweb's worldwide website-visit shares as quoted by outlets. The order is ChatGPT, Gemini, Claude.

  • October 2025: 71.8%, 13.6%, 2.3% (as quoted by Visual Capitalist)
  • April 2026: 53.7%, 26.7%, 8.0% (as quoted by HuggingNews and OfficeChai)
  • May 2026: 52.7%, 27.3%, 8.9% (PPC Land, update as of May 26)
  • July 2026: 54.3%, 26.6%, 9.6% (as quoted by OfficeChai)
  • August 2026: 55.6%, 25.6%, 9.3% (as quoted by OfficeChai. The Decoder reported ChatGPT for the same period as 55.5%)
  • September 2026: 56.4%, 25.0%, 9.6%

HuggingNews calculated that compared with April, ChatGPT is up 2.7 percentage points and Gemini is down 1.7 percentage points, and that compared with October 2025, ChatGPT is 15.4 percentage points lower. ChatGPT's September web visits were about 5.9 billion, and Visual Capitalist wrote that only this value was published directly by Similarweb. Forbes Australia reported that the September figure is marked as a "projection," that it is up from about 5.6 billion visits in August, and that it is the highest since October 2025 (about 6.2 billion visits). The visit counts for Gemini and Claude (about 2.6 billion and about 1.0 billion) are estimates back-calculated from the shares.

In sum, ChatGPT, which had fallen to 52.7% in May (we could not confirm the June figure), is rising again but is still far below a year ago, while Gemini and Claude remain at much higher positions than a year ago.

What this number does not count (interpretation)

1. It counts website visits only. App use, API use, and AI built into other products are left out (Visual Capitalist). The Decoder pointed out that use where a notification on Android opens the Gemini app directly is likely not captured as a web visit, and that use of the ChatGPT mobile app and desktop app is not in these statistics either.

2. App metrics paint a different picture. The download share in the second-quarter 2026 count (up to the time of the count) that Forbes Australia (May 29) reported from Sensor Tower data was ChatGPT 47%, Gemini 22%, Claude 14%. Compared with Claude's web share of around 8% in the same period, its weight is larger in apps. However, a download is an install, not usage, and because the period and the subject differ, the two numbers cannot be compared directly. Similarweb's own report also shows website visits and app metrics (monthly and daily users and so on) separately.

3. "A year ago" differs by outlet. For ChatGPT's share a year earlier, PPC Land wrote 76.4% (June 2025), The Decoder 73.3%, and Visual Capitalist 71.8% (October 2025). That is why the size of the drop measured against September's 56.4% looks different, from about 15 percentage points to 20 percentage points. We have to compare against the same baseline.

4. Why it rose is a guess. Forbes Australia included analysts' views (user loyalty, model evaluations) and a Similarweb representative's remark on Gemini's decline that "Google may have integrated Gemini features into regular search." Both are interpretations, not confirmed causes.

User counts that companies announce use yet another yardstick

  • Google (August 11, what the company announced): It said Gemini app monthly active users passed 1 billion. According to SiliconANGLE, Google did not explain how it defined and measured monthly active users, and it did not disclose the number of paid subscribers or revenue. The same article estimated that because Gemini is built into Android, Chrome, Search, and Workspace, inflow through default settings could be substantial, but that is the outlet's estimate, not something Google confirmed.
  • OpenAI (reported as announced by the company): TNND reported that it announced on August 6 that ChatGPT's weekly active users passed 1 billion. However, an AFP report (carried by Aletihad) said OpenAI stated on July 31 that it reaches "more than one billion active users" without saying weekly or monthly, and Crypto Briefing reported that OpenAI confirmed 1 billion weekly active users on September 8, so the announcement date and the "weekly" basis differ by outlet. We could not confirm the exact date and basis, and we could not open the original OpenAI announcement directly.
  • Third-party estimate: ChatGPT app monthly users of 1 billion is a Sensor Tower estimate (as of May 2026), reported by TNND citing Reuters.

Weekly users and monthly users are different metrics, and neither company has detailed its calculation basis, so the numbers cannot be put side by side and compared. ChatGPT leading in web visits and the Gemini app having 1 billion monthly users do not contradict each other; they are the results of counting different things.

What we could not confirm in this news

  • We could not open Similarweb's raw data (its own platform, its X post) directly, and all the shares above are quoted through outlets. Similarweb's own blog post covered web traffic data up to May 2026 and did not carry percentage figures.
  • We could not confirm the June figure. OfficeChai wrote that "Gemini's share fell for four straight months." The May, July, August, and September figures we confirmed point downward, but "four straight months" is that outlet's wording.
  • We could not confirm why the August figure splits into 55.6% and 55.5% (an update or rounding difference).
  • We could not confirm Similarweb's counting method, sample composition, or which services fall into "Other."
  • We could not open Google's official blog and OpenAI's original announcement directly, and read them through SiliconANGLE, TNND, and other reporting.

Our view

Fact: In Similarweb's September count, ChatGPT's web-visit share rose to 56.4%, higher than in April (53.7%), and Gemini's fell to 25.0%, lower than in April (26.7%). This number counts website visits only. The user counts reported for Google and OpenAI are on a monthly and a weekly basis respectively, so they differ.

Interpretation: A share number reads as if it shows "who is ahead," but the picture changes depending on what is counted. We cannot say that use has fallen because web visits have fallen, and we cannot say that people use something a lot because its app downloads are high.

What we would do: We do not switch the tools we use because a share goes up or down. We keep the criterion set out in NO.17: the results of running our three weekly tasks ourselves. And when we look at the numbers for our own apps and channels, we ask the same questions. We write next to each number whether it is visits, installs, or users, which period it covers, and who measured it.

Articles we read together

We opened and checked sources of different kinds directly. All shares are Similarweb estimates, and anything we read through an outlet is as given by that outlet.

Research-firm material:

Reporting and analysis:

Reporting on company announcements:

← The Inspectors Came Inside the…Skills became a "shared standa… →