Everyone assumes AI is killing Stack Overflow. The reasoning seems obvious: if developers can ask ChatGPT instead of searching Google, fewer people click through to SO. The data tells a stranger story.
Stack Overflow's US organic traffic is roughly where it was in early 2022. What did change is what sits underneath. Google removed millions of keywords from SO's index, deindexed hundreds of thousands of pages, and pushed the site to position 2 on thousands of SERPs. None of this moved the traffic needle. Because almost all of SO's traffic was already concentrated in a tiny fraction of pages.
TL;DR
- 700,797 Stack Overflow pages have organic rankings. Just 6,988 (1%) carry 59% of all US organic traffic.
- SO's total keyword rankings dropped from 8.2M to 1.1M since early 2022. An 87% decline. Traffic stayed flat.
- The keyword loss is almost entirely in positions 11+, which collapsed from 7M to 195K. Top 3 rankings barely moved (606K to 480K).
- When SERP features appear, SO drops from position 1 to position 2 on 51% of affected queries.
- The pages that survived are almost all git commands and Python basics. The pattern: questions too dangerous to let an LLM hallucinate.
- Tech media lost 58% of its traffic over the same period. Developer Q&A held. The difference is structural.
Methodology
All traffic, keyword, and page-count data is from Ahrefs, pulled for the US market between February 2022 and February 2026. Page-level analysis uses Ahrefs' "Top Pages" report for stackoverflow.com (subdomains mode), filtered to pages with ≥50 monthly organic visits unless otherwise noted. SERP feature analysis uses Ahrefs' SERP feature filter on the /questions/ subdirectory. Ahrefs estimates organic traffic based on ranking positions and modeled click-through rates. The numbers are directional, not exact.
The Concentration
Stack Overflow has roughly 59 million questions. How many of those actually receive organic search traffic from Google in the US?
Read that bottom row again. 6,988 pages. Out of 700,797 with any organic ranking at all. Out of 59 million total questions. That 1% carries 2.6 million of the site's 4.4 million US organic visits.
The distribution within that 1% is even more extreme. Only 28 pages get 5,000+ monthly visits. Only 414 get over 1,000. The majority of "surviving" pages pull between 100 and 500 visits a month.
Four hundred and fourteen pages. That's the real Stack Overflow, as far as Google is concerned. Everything else is rounding error.
The Keyword Collapse That Didn't Collapse Traffic
Here's the number that should be alarming but isn't: Stack Overflow's total keyword rankings in the US dropped from 8.25 million to 1.11 million since February 2022. That's an 87% decline. On any other site, that would be a death sentence.
On Stack Overflow, traffic barely moved.
The loss is almost entirely in positions 11 and beyond. These are pages that ranked on page 2, page 3, page 15 of Google results. Pages that drove close to zero clicks. Google removed Stack Overflow from millions of long-tail SERPs where it wasn't useful, while keeping it in the top spots where people actually click.
Top 3 rankings dropped only 21%. The core inventory is intact. Google didn't downrank Stack Overflow. It trimmed the dead weight.
Most sites would collapse if they lost 87% of their keywords. Stack Overflow didn't because its traffic was never distributed across those keywords in the first place. The 7 million keywords in positions 11+ contributed almost nothing. They were the SEO equivalent of empty calories: they showed up in dashboards and ranking reports but drove no visits, no engagement, no value.
What Survived: Git Commands and Python Basics
The pages still pulling real traffic from Google aren't random. They follow a pattern.
| Top Keyword | Monthly Traffic | Position | Keyword Volume |
|---|---|---|---|
| no keyword (MySQL) | 19,349 | 1 | 59,000 |
| git delete branch | 15,276 | 2 | 8,400 |
| git undo commit | 14,871 | 2 | 4,600 |
| python string contains | 11,368 | 2 | 6,100 |
| pip update | 9,128 | 2 | 3,800 |
| python switch | 8,843 | 2 | 3,800 |
| python substring | 7,650 | 2 | 6,200 |
| git checkout remote branch | 7,646 | 2 | 5,700 |
| python multiline comment | 6,634 | 2 | 3,300 |
| conda create environment | 6,484 | 2 | 3,800 |
US, Feb 2026. Sorted by estimated traffic.
Git operations. Python syntax. Package manager commands. Error messages. Docker troubleshooting. These aren't conceptual questions or opinion-seeking queries. They're operational commands where a wrong answer breaks something.
By topic, the traffic breakdown across the top 13,884 pages (those with 50+ monthly visits) clusters heavily:
Python and Git alone account for a third of all surviving traffic. These topics have something in common: the answers are deterministic. git branch -d feature-x either deletes the branch or it doesn't. There's no subjective nuance. There's no "it depends." The community-verified answer on Stack Overflow is still the safest place for Google to send users.
Notice the position column in that table. Almost every top page ranks at position 2, not position 1. Google's AI Overviews and other SERP features have taken the #1 slot. Stack Overflow has been moved to the first "real" result, right below Google's own answer. This is a controlled demotion, not a removal.
SERP Features: The Position 2 Problem
We filtered Stack Overflow's /questions/ pages to only those where SERP features (AI Overviews, featured snippets, discussion boxes) appear on the results page. The shift is clear.
With SERP Features Present
5,283 pages face SERP features
Only 6% hold position 1
51% sit at position 2
Google answers the query itself, then shows Stack Overflow.
All Pages (No SERP Filter)
13,884 pages with 50+ traffic
45% hold position 1
27% sit at position 2
Without SERP competition, SO still dominates.
When SERP features show up, Stack Overflow's position 1 rate drops from 45% to 6%. Google takes the answer, formats it, and gives it to the user without a click. Stack Overflow becomes the backup. The page users scroll past the AI-generated answer to reach, if they reach it at all.
This is a controlled arrangement. Google isn't removing SO from these SERPs. It's keeping it as the verification layer. The source the AI answer was probably derived from. But the traffic split has changed. Google gets the easy clicks. Stack Overflow gets the skeptical ones.
The Deindexing
The number of Stack Overflow pages appearing in Google's index with organic rankings started falling in mid-2025 and hasn't stopped.
The Top Pages charts tell the story through the gap between organic pages (how many pages rank) and organic traffic (how many visits those pages generate).
At traffic ≥ 1: 261,822 pages remain in the index. That means roughly 440,000 pages that previously ranked now show zero estimated organic traffic. Google either deindexed them or pushed them so far down they no longer register.
At traffic ≥ 10: only 54,850 pages. The organic pages line on this chart shows a sharp decline starting around August 2025, while the organic traffic line holds steady or even ticks up slightly.
At traffic ≥ 100: 6,988 pages. The page count drops but the traffic line is flat. Google is removing pages from the index without losing the traffic those pages weren't generating anyway.
Google is pruning Stack Overflow the way a gardener prunes a tree. Cut the dead branches. Leave the ones bearing fruit. The tree's yield doesn't change because the dead branches weren't producing anything. But the tree looks very different in a keyword tracking dashboard.
Why SO Survived When Tech Media Didn't
In a separate analysis, we found that ten major tech publications lost 58% of their combined US organic traffic since 2024. Digital Trends dropped 97%. ZDNet dropped 90%. These are catastrophic declines.
Stack Overflow, over the same period, stayed essentially flat. So did other developer reference sites.
| Site | Type | Peak Traffic | Jan 2026 | Change |
|---|---|---|---|---|
| Stack Overflow | Developer Q&A | 8.3M | 8.6M | +4% |
| GeeksForGeeks | Developer tutorials | 19.2M | 21.4M | +11% |
| W3Schools | Reference docs | 20.5M | 13.4M | -35% |
| MDN Web Docs | Reference docs | 7.5M | 6.1M | -19% |
| Digital Trends | Tech editorial | 8.6M | 269K | -97% |
| ZDNet | Tech editorial | 7.6M | 769K | -90% |
| Wired | Tech editorial | 4.5M | 1.7M | -62% |
US organic traffic, monthly estimates. Peak = highest month from Feb 2022 to date.
The pattern: developer Q&A and reference documentation held. Tech editorial collapsed. Why?
Three Structural Differences
Verifiability. Stack Overflow answers can be tested. Copy the code, run it, see if it works. Tech editorial is opinion dressed as analysis. AI can replicate opinion. It can't replicate "this code compiles and produces the correct output," backed by community upvotes from people who actually ran it.
Canonical authority. For "git delete branch," there is one correct procedure. Stack Overflow's answer has been tested by millions of developers. Google has no reason to replace this with a generated answer because the generated answer would be derived from this page anyway. For "best laptop 2026," there is no canonical answer. AI can synthesize that just as well as a human reviewer.
Risk profile. If an AI Overview gives you a slightly wrong laptop recommendation, you buy a slightly wrong laptop. If it gives you a slightly wrong git command, you might delete your production branch. Google keeps SO visible for queries where hallucination has real consequences.
The dividing line isn't "developer content" vs "non-developer content." It's deterministic answers vs. subjective analysis. W3Schools is down 35% because some of its reference content can be absorbed into AI Overviews without risk. GeeksForGeeks grew because its tutorial-style content drives engagement Google values. The format matters as much as the topic.
What This Means
For Developers
Stack Overflow isn't dying. It's consolidating. The site's value to Google is concentrated in a small number of canonical answers to operational questions. If you're searching for something where the answer is a specific command or code snippet, SO will keep showing up. If you're searching for architectural advice, framework comparisons, or "should I use X or Y" questions, AI is already eating that. Those are the pages Google deindexed.
For Content Creators
The traffic concentration data is a warning. Stack Overflow had 59 million pages of content. Google cares about 7,000 of them. The lesson isn't "create less content." It's "understand that most content will eventually get zero organic traffic, and plan accordingly." Build around the canonical answers in your space. Don't build around the long tail.
For SEOs Watching Their Keyword Counts
An 87% keyword decline looks apocalyptic in a dashboard. Stack Overflow shows that keyword count and traffic can decouple completely. If your keywords are in positions 11 or beyond, they're already dead. Losing them from your ranking report is Google catching up to reality, not a new penalty. Tools that track keyword counts without weighting by traffic are missing the real picture.
Stack Overflow isn't a cautionary tale about AI killing developer communities. It's a case study in extreme concentration. The site's traffic was always held up by a tiny percentage of canonical answers. What changed is that Google stopped pretending the other 99% mattered. The pruning is mechanical, not malicious. But it reveals something most site owners haven't internalized: the vast majority of your content might already be invisible. You just don't know it yet because the keywords still show up in your dashboard.
Raw Data: Stack Overflow US Organic Traffic (Monthly)
| Month | Organic Traffic | Top 3 Keywords | Pos 4-10 Keywords | Pos 11+ Keywords | Total Keywords |
|---|---|---|---|---|---|
| Feb 2022 | 8,289,836 | 605,777 | 586,677 | 7,062,436 | 8,254,890 |
| Jun 2022 | 7,137,917 | 356,474 | 391,711 | 5,597,682 | 6,345,867 |
| Nov 2022 | 7,480,644 | 359,289 | 384,951 | 5,574,266 | 6,318,506 |
| Jan 2023 | 5,976,949 | 309,754 | 386,319 | 5,661,126 | 6,357,199 |
| Jun 2023 | 5,862,889 | 500,894 | 570,167 | 7,483,797 | 8,554,858 |
| Jan 2024 | 7,086,938 | 503,402 | 716,648 | 7,940,024 | 9,160,074 |
| Jun 2024 | 16,063,647 | 584,136 | 853,551 | 8,061,259 | 9,498,946 |
| Jan 2025 | 11,088,089 | 495,692 | 626,319 | 5,140,707 | 6,262,718 |
| Jun 2025 | 7,160,376 | 446,927 | 476,599 | 2,748,811 | 3,672,337 |
| Oct 2025 | 8,670,336 | 497,391 | 460,178 | 1,205,299 | 2,162,868 |
| Jan 2026 | 8,614,048 | 498,511 | 462,563 | 248,144 | 1,209,218 |
| Feb 2026* | 6,563,794 | 480,023 | 437,950 | 195,264 | 1,113,237 |
*February 2026 data is partial (month not complete at time of pull). All figures are estimates for the US market.
Raw Data: Comparison Sites (US Organic Traffic, Monthly)
| Site | Feb 2022 | Nov 2022 | Jan 2024 | Jan 2025 | Jan 2026 |
|---|---|---|---|---|---|
| stackoverflow.com | 8.3M | 7.5M | 7.1M | 11.1M | 8.6M |
| w3schools.com | 16.1M | 20.5M | 15.7M | 15.1M | 13.4M |
| developer.mozilla.org | 7.3M* | 7.4M* | 7.3M* | 6.3M | 6.1M |
| geeksforgeeks.org | 19.2M* | 20.4M* | 19.4M* | 25.7M | 21.4M |
| chatgpt.com | n/a | n/a | 14K | 117M | 765M |
| perplexity.ai | n/a | n/a | 3.0M | 7.0M | 8.7M |
Global organic traffic estimates. *MDN and GeeksForGeeks data prior to Feb 2024 is from global estimates. ChatGPT.com domain launched mid-2024.
Related Articles
Tech Media Lost 58% of Google Traffic: Original Research
Ten major tech publications lost more than half their organic search traffic. We analyzed the data.
ResearchReddit Backlinks Research: How Reddit Ranks Without Links
Reddit pages rank in the top 10 with zero backlinks. We analyzed the data behind this anomaly.
ResearchThe Publisher Affiliate Collapse
Google's policy changes decimated publisher affiliate sites. Here's the data on who survived.

Yuval Halevi
Helping SaaS companies and developer tools get cited in AI answers since before it was called "GEO." 10+ years in B2B SEO, 50+ cybersecurity and SaaS tools clients.