A Hacker News thread this week claimed Stack Overflow now gets fewer than 1,000 questions a month. That number sounded like a typo — the site I learned to program on used to get that many before lunch. So I counted it myself, straight from the Stack Exchange API, instead of trusting a chart.
How I counted
No scraping, no dashboard screenshot. I called the Stack Exchange API's /questions endpoint with a fromdate/todate window for each month, pagesize=1, and read the total field. That counts questions created in the window, as the API sees them. The one caveat worth stating up front: October 2026 is a partial month in my data (measured October 29), so September — a complete month — is the number that actually settles the claim.
import requests, datetime
def count_month(y, m):
fromdate = int(datetime.datetime(y, m, 1, tzinfo=datetime.timezone.utc).timestamp())
todate = int(datetime.datetime(y, m + 1, 1, tzinfo=datetime.timezone.utc).timestamp()) if m < 12 else \
int(datetime.datetime(y + 1, 1, 1, tzinfo=datetime.timezone.utc).timestamp())
r = requests.get("https://api.stackexchange.com/2.3/questions",
params={"site": "stackoverflow", "fromdate": fromdate,
"todate": todate, "pagesize": 1, "filter": "total"}, timeout=30)
return r.json().get("total")
The numbers
Month Questions created
2026-08 1,019
2026-09 969
2026-10 (to 10/29) 977
2025-10 5,263
2024-10 23,162
2020-10 141,749
September 2026, a full month, came in at 969 — under 1,000, with no partial-month asterisk needed. August was 1,019, so the crossing happened right around now. October is already at 977 with days left, so it will probably finish just over the line — but the trend is not ambiguous about where this is going.
The declines, computed from those API totals:
September 2026 vs October 2025: 969 vs 5,263 = -81.6%
September 2026 vs October 2024: 969 vs 23,162 = -95.8%
September 2026 vs October 2020: 969 vs 141,749 = -99.3%
A 99.3% drop from October 2020. Read that again. The site that defined "Google the error message" now produces, in a month, roughly what it once produced in a few hours.
What this does and doesn't prove
It proves the volume collapse is real and measurable through Stack Exchange's own API, not an artifact of one chart's query. It does not prove why. The usual suspects — AI assistants answering the question before anyone posts it, the long decline in new-user onboarding, a decade of moderation friction — are all plausible, and this count can't separate them. What it does rule out is the comforting version: that the numbers are fine and only the discourse is gloomy. They are not fine.
There is also a second-order effect worth noticing. Stack Overflow's value was never just answers to your question; it was answers to the next person's question, found by search. At under 1,000 new questions a month, the corpus is still enormous, but it is effectively frozen in time for fast-moving topics. The site becomes an archive with a shrinking intake valve — still useful, increasingly historical.
I learned to code by reading that archive. I did not expect to watch its intake drop to three digits a month. But the API says what it says: 969.
Sources: Stack Exchange API v2.3, /questions endpoint (site=stackoverflow), queried October 29, 2026; Hacker News front page, October 29, 2026 (Stack Overflow question-volume thread).
Top comments (1)
Love how you highlight the problem with slop code by some overly complex python code with external dependency for query which could have been a curl
October 2026:
Past 30 days: