The Rise and Death of Stack Overflow: 9 Moments That Took It From 200,000 Questions a Month to Almost Zero

I still remember my first accepted answer on Stack Overflow. It was something trivial about a null reference in C#, worth maybe fifteen reputation points, and I refreshed the page four times to make sure the green checkmark was real. That was 2013. For most of my career, Stack Overflow was not a website I visited. It was infrastructure, the same way DNS is infrastructure.

Then it stopped.

In July 2026, developers across the entire planet posted 1,442 new questions on Stack Overflow. In March 2014, that number was 207,204 in a single month. That is a decline of 99.3%, and the figures are not modeled, scraped, or estimated. They are raw row counts pulled from the Stack Exchange Data Explorer, the platform's own public SQL query interface, as reported by PPC Land. Anyone can re-run the query. The counting method is not in dispute.

Developer Daniel Lockyer, who posted the chart on X in August 2026, summed it up in five words: "The end of an era."

Below is my reconstruction of how we got here, told through the nine moments that mattered most. I have ordered them chronologically rather than by importance, because the sequencing is the whole story. Stack Overflow did not die when ChatGPT launched. It was already bleeding out eight years earlier.

 

---

 

1. September 2008: A launch that solved a real, expensive problem

Jeff Atwood and Joel Spolsky launched Stack Overflow on 15 September 2008 as a deliberate alternative to Experts-Exchange, a site notorious for dangling answers behind a paywall. The founding insight was not technical. It was economic. Programming knowledge was abundant, but the market for distributing it was broken.

Their fix combined three mechanisms that were unusual in 2008:

• Reputation as currency. Upvotes granted points, points granted moderation privileges. Contribution was gamified before "gamification" was a boardroom word.

• Canonical answers over threads. Instead of a forum thread where the correct fix is buried on page seven under signature GIFs, one question mapped to one best answer, voted to the top.

• A permissive content license. Everything was published under Creative Commons BY-SA, which meant the corpus was legally reusable. Hold that thought. It becomes the single most important detail in this entire story.

2. 2010 to 2013: The golden age and the network effect flywheel

This was the period where Stack Overflow achieved something rare: a self-reinforcing content flywheel. More questions attracted more answerers, which improved search engine ranking, which drove more traffic, which produced more questions.

By the mid-2010s the site had accumulated what is now roughly 24 million questions and 36 million answers, the largest structured programming Q&A corpus ever assembled. For a working developer, a Stack Overflow profile carried more professional weight than a LinkedIn page. I know engineers who put their reputation score on their resume, and I know hiring managers who actually looked at it.

3. March 2014: Peak questions, and the moderation turn

Here is the moment almost everyone gets wrong. Stack Overflow peaked in March 2014 with 207,204 monthly questions, and the decline started immediately after. That is eight years before ChatGPT existed.

What changed in 2014 was moderation tooling. Stack Overflow significantly improved moderator efficiency, which meant questions were closed faster, closed more often, and flagged as duplicates more aggressively. The stated goal was preserving signal-to-noise ratio. The actual outcome was a hostile onboarding funnel for new users.

The numbers back this up. A 2022 academic study sampled 968 posts from new users and found that 49% ran into at least one of the following: closure, zero responses, or unexplained downvotes. Roughly half of all newcomers had a bad first experience. That is not a moderation policy. That is a churn engine.

Gergely Orosz of The Pragmatic Engineer, who documented the collapse in Stack Overflow is almost dead, described his own exit around this period bluntly: the site, he wrote, "felt unwelcome." I stopped asking questions around the same time, for the same reason. I kept reading. I just stopped contributing.

4. March 2020: The pandemic bump that disguised the trend

Lockdowns pushed millions of developers into remote work overnight. You could no longer swivel your chair and ask the senior engineer next to you, so you Googled instead, and Google sent you to Stack Overflow. Question volume spiked.

This bump matters mostly because of what it hid. It temporarily masked a structural decline that had been running for six years, and it gave leadership a reason to read the trend line optimistically.

5. June 2020: The decline resumes, and accelerates

Within three months the spike faded and questions started falling faster than before. This is the detail I find most damning for the "AI killed Stack Overflow" narrative. The second-derivative acceleration began two and a half years before ChatGPT shipped.

6. June 2021: The $1.8 billion exit

On 2 June 2021, Prosus acquired Stack Overflow for $1.8 billion, the Dutch consumer-internet conglomerate best known for its enormous early stake in Tencent. Spolsky's announcement promised continuity, writing that "Stack Overflow will continue to operate independently."

In hindsight, the founders sold with close to perfect timing. Atwood and Spolsky exited before the terminal phase of the decline was visible in any public dashboard. That is not a criticism. It is an observation about how illegible platform decay is from the inside, even to the people who built the platform.

7. November 2022 to mid-2023: ChatGPT, the ban, and the moderator strike

ChatGPT launched on 30 November 2022, and the question curve broke.

The competitive dynamic was brutal and obvious. ChatGPT was trained substantially on Stack Overflow's own CC-licensed corpus, so answer quality was comparable. It was faster. It never closed your question as a duplicate. It never told you to read the manual. Stack Overflow's response was to ban AI-generated answers in December 2022, which triggered a strike by a portion of its volunteer moderators in mid-2023 over policy changes they viewed as reactive and dismissive.

A developer quoted by DevClass captured the emotional core of the migration better than any analyst has: people were relieved to have "a tool that didn't tell them their questions were stupid."

The annual numbers from this point read like a controlled demolition:

December 2025 alone saw just 3,862 questions, a 78% year-over-year drop. The 12,322 questions posted across the first seven months of 2026 represent roughly two days of traffic at 2016 volumes.

8. October 2023: Layoffs, and the pivot from community to corpus

In October 2023, Stack Overflow cut 28% of its staff. CEO Prashanth Chandrasekar framed it as a path to profitability under macroeconomic pressure.

Then came the strategic reversal. Having banned generative AI output in December 2022, Stack Overflow announced an OverflowAPI partnership with OpenAI in May 2024, licensing its vetted Q&A data for model training and grounding in exchange for attribution inside ChatGPT. The company that banned the technology became a supplier to it in roughly eighteen months.

9. 2024 to 2026: The forum dies, the business survives

This is the twist that makes Stack Overflow genuinely interesting rather than merely sad.

According to Sherwood News, Stack Overflow's annual revenue has roughly doubled to $115 million, while losses narrowed from $84 million in FY2023 to $22 million in the most recent fiscal year. The growth engine is no longer advertising against public forum traffic. It is Stack Internal, an enterprise knowledge platform with a retrieval-augmented generative AI layer built on the accumulated corpus, now deployed at 25,000 companies.

Read that carefully. The public community collapsed by 99%, and the company got healthier. The asset was never the daily question flow. The asset was seventeen years of human-verified, structured, deduplicated ground truth, and that asset appreciated exactly as its production stopped.

 

---

 

What the data actually says about why developers left

I want to push back on the simple version of this story, because Stack Overflow's own research complicates it.

The 2025 Developer Survey collected 49,000+ responses from 177 countries. Yes, 84% of developers use or plan to use AI tools, up from 76% in 2024. But trust moved the opposite direction: 46% actively distrust AI accuracy versus 33% who trust it, with only 3% reporting high trust. Favorability fell from over 70% in 2023 and 2024 to 60% in 2025.

The friction points are specific. 66% of developers cite solutions that are almost right but not quite as their top frustration, and 45% say debugging AI-generated code takes longer than writing it themselves. Chandrasekar's own framing was measured: "AI is a powerful tool, but it has significant risks of misinformation."

Most revealing of all: 35% of developers still turn to Stack Overflow specifically after AI-generated code fails, and when asked why they would still ask a human in an AI-saturated future, the top answer at 75% was simply that they do not trust the AI's answer.

So developers did not leave because AI is better. They left because AI is sufficient, instant, and never rude, and because Stack Overflow spent a decade optimizing for archive quality at the direct expense of participant experience.

Three lessons I am taking from this

Community moats are consumption moats, not production moats. Stack Overflow's defensibility looked like network effects. It was actually search-engine distribution plus a permissive license, both of which transferred cleanly to LLM providers. When your content is CC-licensed and your traffic comes from Google, you do not own your funnel.

Onboarding hostility compounds silently. Closing a duplicate question costs a moderator ten seconds and costs the platform one potential lifetime contributor. Half of new users hitting friction in 2014 meant a thinner answerer cohort in 2018, which meant slower response times in 2022, which meant no defense when ChatGPT arrived.

The corpus outlived the community, and that should worry us. Pieter Levels raised the sharpest version of this point when the July 2026 numbers circulated: if nobody asks questions publicly anymore, what does the next generation of models train on? Stack Overflow's archive is a snapshot of human programming knowledge frozen somewhere around 2023. New frameworks, new runtimes, new failure modes, none of it is being documented in public, verified, structured form at anything close to the old rate.

I do not think Stack Overflow will formally shut down soon. The enterprise business works, and 24 million questions remain genuinely useful reference material. But the thing I loved, a public square where a stranger in another timezone fixed my broken build for free and got fifteen internet points for it, is gone. The replacements are Discord servers, private Slack workspaces, and chat windows, and none of them are indexed, permanent, or shared.

That is the part I actually mourn. Not the site. The commons.

 

---

 

Sources: The Pragmatic Engineer · Stack Exchange Data Explorer · PPC Land · DevClass · Sherwood News · TechCrunch · 2025 Stack Overflow Developer Survey · Wikipedia