Data Intelligence Report — 2026
Stack Overflow · 2008 – 2022

We turned 14 years of developer data into business insights for less than a cup of coffee

23 million questions, 3 data layers, $0.040 — built on Google BigQuery using a Medallion architecture. Here is what the data actually says.

Questions analyzed
23Mrecords
Time span
14years
Total processing cost
$0.04USD
Architecture
Bronze / Silver / Gold
Avg monthly questions
134k
Across all 171 months
Avg answer rate
53%
14-year overall average
Avg score per question
5.2
Down from 65 in 2008
Avg engagement ratio
1.8%
Answers + comments / views
01 / 05
Monthly question volume — growth and decline
Stack Overflow grew explosively from 2008 to 2014, then declined consistently. March 2014 was the peak with 209,224 questions in a single month — a record never broken again.
Insight

The platform grew 55x in 6 years (2008 to 2014), then lost nearly 40% of monthly volume by 2022. This mirrors the maturation of the developer ecosystem — foundational questions had already been asked and answered.

02 / 05
Answer rate — from 85% to 23% in 14 years
In 2008, nearly every question received an accepted answer. By 2022, only 1 in 4 did. This is one of the most striking signals in the entire dataset.
Insight

The answer rate drop reflects a shift in question complexity and community behavior. Questions became more niche, more specific, and harder to answer with a single accepted solution. The community's knowledge base was no longer growing faster than the questions being asked.

03 / 05
Average score — the value signal collapsed
The average score dropped from 65 in August 2008 to nearly 0 by 2022. Early questions were universal — they got upvoted every time someone googled the same problem for years.
Insight

High scores equal high reuse. A question scored 65 means thousands of people had the same problem and found that answer useful. A score near 0 means the question was too specific to benefit anyone else. The internet ran out of universal programming questions.

04 / 05
High impact vs low quality questions per year
High impact questions (10,000+ views) peaked in 2012–2014 and nearly disappeared by 2022. Low quality questions (score below -5) grew as a share of total volume.
High impact (10k+ views)
Low quality (score < -5)
Insight

In 2012 there were 87 high impact questions for every 1 low quality one. By 2022 that ratio had nearly inverted. The gold mine of universal knowledge questions had been exhausted — what remained was increasingly niche or low value.

05 / 05
Engagement ratio — the paradox of decline
As overall volume fell, engagement per view increased consistently. Fewer questions were asked, but the ones that were asked generated proportionally more discussion relative to their views.
Insight

Higher engagement per view is a double-edged signal. It could mean the remaining community is more dedicated — or that questions are so specific they require back-and-forth discussion rather than a clean accepted answer. The data suggests both are happening simultaneously.

Before you go

Every data point in this dashboard — every question, every vote, every accepted answer — happened before the explosion of generative AI. ChatGPT launched in November 2022. The last month in this dataset is September 2022. What you are seeing is the full arc of Stack Overflow as the world's primary source of developer knowledge, before developers had somewhere else to ask their questions. That context changes everything about how you read these trends.