ChatGPT or Gemini reports fewer users than a year earlier

Where things stand (13 Oct 2026)

ChatGPT (OpenAI, weekly active users). OpenAI's own figures have only gone up: 700M in Aug 2025, 800M in Oct 2025, 900M on 27 Feb 2026 (TechCrunch, citing OpenAI's funding post), about 1B in early Aug 2026, and "more than 1 billion weekly active users" on 31 Aug 2026 (https://openai.com/index/expanding-access-to-ai-with-chatgpt-ads/). At DevDay on 29 Sep 2026 OpenAI cited "our collective 1.2B weekly users" (https://openai.com/index/devday-2026-recap/), and its 7 Oct 2026 GPT-6 post repeated more than 1.2B weekly users. That is about 50% more than a year earlier.

Gemini app (Alphabet, monthly active users). The reported figures were 450M in Q2 2025, 650M in Q3 2025 and 750M in Q4 2025. Alphabet gave no figure with Q1 2026 results; Polymarket's market summary notes the omission. It then reported 900M at I/O, 950M in the Q2 2026 CEO remarks (https://blog.google/company-news/inside-google/message-ceo/alphabet-earnings-q2-2026/) and more than 1B on 11 Aug 2026 (https://blog.google/innovation-and-ai/products/gemini-app/one-billion-monthly-users/). That is roughly 55% growth over a year.

No qualifying decline has been published. The next scheduled releases are Alphabet's Q3 2026 results on 28 Oct and further OpenAI posts. Both will compare against much lower year-earlier figures.

Reference class and base rate

I made a rough list from memory of public consumer internet companies that report a headline user metric. About 40–50% reported a year-over-year decline within about five years: Twitter (2018), Snap (Q4 2018), Pinterest (2021–22), Zynga, King, Robinhood, Coinbase, Peloton, Bumble and others. Dominant products with network effects (Facebook, Instagram, WhatsApp, Spotify, Roblox) did not. ChatGPT is the category leader but has little network effect, and it faces bundled competitors with huge distribution: Gemini on Android and Chrome, Meta's Muse, which overtook ChatGPT on the free app charts in Sep 2026, Siri and Grok. Its app market share fell from 69% to 45% in a year (Fortune, 5 Feb 2026). The Wall Street Journal and Reuters reported in Apr 2026 that it missed internal user targets. Its web visits are flat year over year, while mobile daily users are up 69% (Bank of America and Sensor Tower data via citybiz, Oct 2026).

I put the chance of a genuine year-over-year decline in some quarter by 2031 at about 25–30%, concentrated in 2029–31 as growth saturates.

Disclosure is the bottleneck

• Gemini. Alphabet chooses when to disclose; it skipped the figure in Q1 2026. A decline would most likely just go unmentioned, so this path adds about 2%.
• ChatGPT while private. OpenAI announces milestones, often with "more than" wording, which does not count under the rules. A private OpenAI would probably stop citing numbers rather than print a lower one.
• ChatGPT after an IPO (main path). OpenAI filed a confidential draft registration in mid-2026. Its finance chief says it will be public in 2027 "or sooner", Altman has ruled out 2026, and Polymarket gives about 40% for a listing by 30 Jun 2027. A public OpenAI would very likely report weekly users every quarter, and companies usually report the first declines before changing the metric (Twitter, Snap, Netflix). I estimate about a 55% chance that a decline gets reported, after allowing for a switch to daily users or combined totals and for a delayed IPO.
• Technicalities. The rules compare rounded figures as printed and treat undated "now" figures as current, so a very stale lower number in a later post could count. I put this at about 1%.

Combining these gives about 19% by the end of 2031.

Hazard model

Computed in Python. Quarterly hazards:

• 0.2% in Q4 2026; both products are growing about 50%.
• 0.3–0.6% per quarter in 2027; a reversal of the Sep 2026 surge is possible, and reporting is unlikely.
• 0.8–1.1% per quarter in 2028, as OpenAI is likely public.
• 1.2–1.3% per quarter in 2029–30.
• 1.2% per quarter in 2031.

The cumulative probability is 2.0% at end-2027, 5.7% at end-2028, 10.3% at end-2029, 14.9% at end-2030 and 18.9% at end-2031.

Strongest case against my estimate

• Higher. Consumer AI is shifting fast toward agents built into operating systems and messaging apps, OpenAI is turning toward enterprise, AI backlash is growing (school bans, protests), and the government intervened in frontier AI in 2026. ChatGPT could peak by 2027–28 just as an IPO forces quarterly disclosure. That scenario would put this at 25–30%.
• Lower. OpenAI may redefine or drop the metric, stay private longer for safety reasons, or keep growing as models improve; Alphabet will not volunteer a decline. That scenario gives about 8–10%.

I kept a middle path.

Question. By the last day of each quarter, has OpenAI or Alphabet reported fewer weekly or monthly active users of ChatGPT or the Gemini app, respectively, than it reported for the same period a year earlier?

Definitions.

• Products and reporters: ChatGPT as reported by OpenAI, or the Gemini app as reported by Alphabet or Google. The app includes its web and native-app interfaces, not just mobile installations. A renamed continuation of the same product counts; an unrelated replacement or a combined total covering other products does not.
• Active users: the company's reported worldwide, product-wide weekly active users (WAU) or monthly active users (MAU), using its own activity definition. Either metric for either product can qualify independently. Compare WAU with WAU or MAU with MAU for the same product and coverage; do not compare across products or frequencies. Regional, subscription-tier, enterprise-only, feature-only, or device-only subsets do not count.
• Same period a year earlier: the same calendar month in the preceding year, the corresponding week 52 weeks earlier, or the corresponding fiscal/calendar quarter where the company reports quarterly averages or quarter-end WAU/MAU. Rolling windows count if the company identifies them as corresponding year-over-year periods. An express company year-over-year comparison establishes the period alignment. For an otherwise undated figure described as “now” or “currently,” use the publication month, or the reporting quarter if disclosed as part of quarterly results; compare only with a figure reported on the same basis a year earlier. Do not substitute the nearest available earlier milestone.
• Fewer: a strictly lower reported count, or an explicit statement of a year-over-year decline in the qualifying metric. There is no minimum size or persistence requirement. Rounded point estimates are compared as printed. Bounds and ranges establish a decline only if they mathematically establish that the newer count is lower; a smaller “more than” milestone alone does not qualify.

Resolves YES if an eligible official publication reports a qualifying decline, with a prior official report available for the corresponding year-earlier period. The earlier report may predate the question's opening date. The newer observation must satisfy the Observation period convention. An explicit year-over-year decline statement need not disclose both numerical counts, but the metric must have been officially reported for the earlier period.

Comparisons must use unchanged coverage and methodology, or an official like-for-like reconciliation. A revision may qualify only if it concerns an eligible newer observation and establishes a decline on a comparable basis against the previously reported year-earlier observation. A revision solely to pre-opening observations does not qualify. A later correction does not reverse an already qualifying YES under the Resolution window convention.

Resolves NO / Exclusions. Resolve NO for a horizon if no qualifying report has appeared by its end. Slower growth, unchanged counts, declining growth rates, sequential weekly/monthly/quarterly declines without a year-over-year decline, and falling visits, downloads, paid subscribers, revenue, messages, or time spent do not count. Neither do model/API usage, Gemini usage in Search, AI Overviews, Workspace or other Google products, or aggregate OpenAI usage outside ChatGPT. Changes in reporting scope alone do not establish a decline. Discontinuing disclosure or shutting down a service without reporting a qualifying usage decline does not count.

Resolving source. Official OpenAI publications on openai.com; Alphabet investor releases, filings and earnings-call recordings/transcripts on abc.xyz/investor or sec.gov; and Google-authored announcements on blog.google or other official Google domains. For example, Alphabet publishes app MAU in its Q2 2026 CEO earnings-call remarks. Official earnings calls count, but executive interviews, personal social-media posts, leaked figures and third-party traffic estimates do not. This is a narrower company-authored source requirement than Public confirmation (A2); third-party confirmation alone cannot resolve it. Apply Resolution date (A3), the Resolution window convention and the Observation period convention.