What AI Did to Stack Overflow, Visualized in One Graph

Last updated: July 18, 2026

A simple data visualization shared on Hacker News today tells a striking story: Stack Overflow's monthly question volume has dropped from around 300,000 questions per month at its peak to roughly 100,000 today - a 65% decline. The acceleration point? November 2022, when ChatGPT launched.

The Data#

The Stack Exchange Data Explorer query plots monthly question counts from 2008 to present. The pattern is clear:

• 2008-2017: Steady growth to peak volumes around 300k questions/month
• 2017-2022: Gradual decline of about 2% annually
• November 2022 onwards: Steep acceleration, dropping from ~180k to ~100k questions monthly

The COVID-19 pandemic shows up as a visible blip around 2020-2021, but the post-ChatGPT decline dwarfs it.

What HN Is Saying#

The Hacker News discussion drew 275+ comments, with the community divided on whether AI caused the decline or merely accelerated an existing trend.

"The decline was already there." Multiple commenters pointed to the pre-ChatGPT downward slope. One wrote: "Except for COVID, it seems the decline was already there." Another noted the pre-AI decline was about 2.2% per year - "hardly a death sentence."

Stack Overflow's moderation culture takes heat. A recurring theme: SO's aggressive moderation drove users away before AI arrived. Commenters described questions closed as duplicates pointing to outdated 2013 answers, hostile responses to beginners, and "karma farming" that prioritized easy questions over hard ones.

One developer shared a telling experience: "I wrote out in detail what I'd done, where I'd got stuck, what I'd read and tried to get unstuck. The very first comment was from some insufferable bellend saying, 'Oh, so you want us to do your work for you?'"

LLMs solve the wait time problem. Several commenters emphasized that response time, not just quality, matters. "On SO a good question might get answered in minutes if it was easy and someone was karma farming, but it could be days or weeks for general purpose stuff. Compare that to a few seconds for an LLM - it's a no brainer."

The duplicate problem compounded over time. Stack Overflow's strict anti-duplicate policy meant questions would get closed and pointed to years-old answers, even when the technology had changed. One commenter explained: "The world has changed since 2013, answers in 2026 would be different, but because the question would be the same, any contemporary attempt at asking would get marked as a duplicate."

Reddit faces similar problems. The thread took a turn toward comparing other platforms. "Reddit is on the same track. Moderators have become increasingly hostile. Reddit's AI moderation, which is designed to remove AI slop, has removed multiple top contributors I used to follow." One commenter claimed it's now possible to get almost any Reddit account banned through mass coordinated reporting.

The Structural Problem#

Stack Overflow was designed to be a canonical Q&A repository - ask once, answer definitively, point all future askers to that answer. This worked brilliantly for building a corpus of programming knowledge, but it created two fundamental issues:

1. Answers age poorly in a fast-moving field. A 2015 answer about React patterns is actively harmful in 2026.

2. The community optimized for the wrong things. Easy questions got fast answers (karma farming). Hard niche questions got ignored. "Working in something of a less common niche myself - embedded Linux - I never had questions get answered."

LLMs, trained partly on Stack Overflow's own corpus, can now serve as a more responsive oracle. The irony is not lost on the community.

Stack Overflow's Response#

Stack Overflow has made several pivots in response to the AI shift:

• Launched OverflowAI, their own AI assistant
• Licensed their data to OpenAI and others for training
• Cut staff significantly (28% layoff in October 2023, additional cuts since)
• Experimented with enterprise-focused products

Whether these moves will reverse the decline or simply manage it remains to be seen. The fundamental question is whether a Q&A site designed around human expertise can compete with AI systems trained on that same expertise.

What This Means for Developers#

For individual developers, the shift is largely positive. Getting unstuck is faster than ever. You no longer need to wait days for a niche question to get answered, craft the perfect SO-compliant question, or navigate hostile moderation.

For the ecosystem, the implications are murkier. LLMs were trained on Stack Overflow's corpus. If the site continues declining, where will future training data come from? Who will catch and correct the errors that LLMs confidently present?

One commenter captured the tension: "SO's downfall started long ago. The community was frankly horribly managed... It was ChatGPT which did it in, but it could've been anything. People were ready to abandon SO."

The Broader Pattern#

Stack Overflow may be a leading indicator for other knowledge platforms. Any site that accumulates expert knowledge into a queryable corpus is potentially vulnerable to AI systems trained on that corpus.

The question isn't whether AI will displace these platforms. The question is what comes next - how we'll generate, verify, and maintain the knowledge that future AI systems will need to stay accurate.

Continue Reading#

• Fable 5 Task Budgets: Capping Agent Spend Before It Happens
• Git Finally Gets a History Command Worth Using
• Three Ways to Ignore Files in Git (Beyond .gitignore)

Sources#

• Stack Exchange Data Explorer Query - Original data visualization
• Hacker News Discussion - Community discussion with 275+ comments