Stack Overflow recorded 1,442 new questions in July 2026, according to query results from the platform's own Stack Exchange Data Explorer. That is 0.7% of the 207,204 questions posted in March 2014, the highest month in the site's history. Developer Daniel Lockyer published the comparison on X on August 7, 2026, describing it as the end of an era.

The post carried a single line of commentary and a chart. According to Lockyer, "Stack Overflow has gone from a peak of 207k questions in march 2014, down to 1.4k in july 2026". The image reproduced a graph generated by the Stack Exchange Data Explorer, the public query interface that exposes the network's post database to anyone willing to write SQL against it. Asked in a reply where the data came from, Lockyer pointed to data.stackexchange.com. By the following day the post had recorded 35.7 thousand views and 463 likes.

The underlying query is not complicated. It selects the first day of the month from each post's creation date, counts rows where PostTypeId equals 1, groups by month and orders ascending. PostTypeId 1 identifies a question rather than an answer. The result set runs to 218 rows, beginning in July 2008 and ending with a partial August 2026. The query page shows it was created on January 3, and the interface reports the result returning in under one millisecond.

What 218 rows show

The series divides cleanly into three phases.

Growth ran from the September 2008 public launch through to the middle of the following decade. July 2008, the private beta month, produced four questions. By January 2010 the site was taking 44,923 questions a month. Annual volume passed one million in 2011 and two million in 2013.

A long plateau followed. The single highest month remains March 2014 at 207,204 questions. Annual totals stayed above two million from 2013 through 2017, peaking at 2,183,926 in 2016. Volume softened after that without collapsing: 1,876,061 questions in 2018, 1,755,877 in 2019, then a pandemic-year uptick to 1,855,259 in 2020, with April 2020 alone contributing 182,854.

The third phase is the one that produces the shape in Lockyer's chart. From 1,336,266 questions in 2022, the annual figure fell to 788,512 in 2023, 398,648 in 2024 and 108,981 in 2025. The whole of 2025 therefore generated fewer questions than an average two-month stretch of 2016. Through the first seven months of 2026 the site logged 12,322 questions in total.

Month by month, the recent numbers compress further. December 2024 recorded 17,938 questions. December 2025 recorded 3,314. January 2026 delivered 2,910, falling to 1,229 by May and 1,076 in June. July 2026 returned 1,442, which is 78.2% below July 2025 and 99.3% below the March 2014 peak.

One detail that complicates the trend line

July 2026 was higher than June 2026 by 366 questions, an increase of 34%. That is the first sizeable month-on-month rise in the series since September 2025, when volume ticked up from 5,814 to 6,040 before resuming its decline. A single month does not establish a floor, and the partial August 2026 figure in the dataset stands at 40 questions at the point the query was run. Whether the July number reflects seasonality, a data artefact or genuine stabilisation cannot be determined from the series alone.

What the dataset does not measure

The Data Explorer queries a live copy of the Posts table. Deleted questions are not present. Because moderation and self-deletion remove posts over time, and because older content has had more years of exposure to that process, historical months are understated relative to recent months by an unknown margin. The direction of that bias matters: it means the true peak-to-trough decline is likely steeper than 99.3%, not shallower.

The count also measures question submissions only. It says nothing about read traffic, which is the larger part of how the site has historically been used, nor about answers, edits, or the reference value of the roughly 24 million questions and 36 million answers already accumulated. A site can lose almost all new contributions while retaining substantial readership on its archive.

A separate discrepancy sits in the public record. The Wikipedia entry on Stack Overflow, last edited on 21 July 2026, states that questions submitted by users fell 78% between December 2024 and December 2025. The figures in the Data Explorer dataset put that decline at 81.5%. The gap is consistent with post deletion between the two measurement points, but neither source documents its snapshot date precisely enough to resolve it.

Infrastructure follows the traffic

Two posts from the same author, seven months apart, bracket the operational consequence. On January 5, 2026, Lockyer wrote that "StackOverflow has so little traffic now that they're going to be hosting on a $5 VPS". The remark was attached to a Stack Overflow blog post dated December 24, 2025, titled "The Great Unracking: Saying goodbye to the servers at our physical datacenter", carrying the subheading "So long and thanks for all the bits!". The hosting claim is hyperbole; the datacenter exit is not.

Ownership context sits in the Wikipedia record. Stack Overflow was created by Jeff Atwood and Joel Spolsky, launched publicly on 15 September 2008, and raised $6 million from investors led by Union Square Ventures on 3 May 2010. Prosus, the Netherlands-based subsidiary of Naspers, acquired the company on 2 June 2021 for $1.8 billion. That transaction closed roughly eighteen months before the steepest section of the decline began.

The site is written in C# on ASP.NET MVC with Microsoft SQL Server and the Dapper object-relational mapper. None of that has changed. What changed is the volume of writes.

Why a programming forum belongs in a marketing news cycle

Stack Overflow is not an advertising business, and its question count has no direct bearing on media budgets. The reason it registers is that it functions as an unusually clean instrument for a change affecting every ad-supported site on the open web.

Most publisher traffic data is contaminated. Algorithm updates, paywall changes, social referral shifts and editorial decisions all move the numbers at once. Stack Overflow's question count strips most of that away. It measures a single behaviour: a working programmer deciding whether to type a question into a public web form or into a chat interface. When PPC Land documented the collapse in January 2026, monthly volume had fallen under 50,000. Seven months later the figure has fallen by a further factor of thirty.

The Wikipedia entry attributes the decline to two causes: the site's perceived hostile moderation culture, and competition from generative artificial intelligence tools. The first predates the second by years. Research cited in that entry found that among a sample of 968 posts from new users, 49% encountered obstacles including closure, no response, or unexplained negative scores. A 2019 paper examined hostility and gender on the platform. The moderation problem was documented long before ChatGPT existed, yet annual volume held above 1.3 million questions through 2022.

The referral pattern is not confined to forums

The same directional movement appears across categories that depend on informational search. A randomized field experiment covering 1,065 desktop Chrome users found that AI Overviews cut outbound organic clicks by 39.8% and raised zero-click searches by 34.5% when the feature appeared, with no measurable improvement in how users rated their experience. Ahrefs research published in February 2026 correlated AI Overviews with a 58% reduction in click-through rates for top-ranking pages, close to double the 34.5% the same firm measured in April 2025.

Market-level figures follow the same line. SISTRIX analysis put German losses at 265 million organic clicks per month, with click-through rate at the first organic position dropping from 27% to 11%. Index Exchange recorded year-over-year declines in ad opportunities across 69% of publishers on its platform through 2025, averaging 14%. Teads told investors on November 6, 2025 that publishers across its network saw pageview declines of 10% to 15% in the third quarter, and that higher revenue per thousand impressions did not compensate for the volume lost.

Consequences have arrived at the level of headcount and closure. Bauer Media Group announced in April 2026 that it would shut its German digital subsidiary on September 30, 2026, eliminating 160 jobs, citing reduced clickthroughs from AI Overviews among three converging pressures. Raptive launched a product tier in July 2026 after publishers in its network lost as much as 60% of search traffic. Overfishing.org, a non-commercial reference site, shut down after 21 years. A Berlin relocation guide reported losing 75% of its traffic.

The supply question nobody has answered

Stack Overflow's archive was training material. Its content sits under Creative Commons Attribution-ShareAlike licences, versions 2.5, 3.0 or 4.0 depending on when a given post was contributed. That permissive footing made it one of the most widely ingested technical corpora available, and the models that ingested it are now the reason new contributions have stopped arriving.

This is the part with commercial weight for platform owners. Reddit has spent two years converting the same asset into contracts and litigation, suing Anthropic in June 2025 over alleged unauthorised training use, and more recently asserting that its content accounted for roughly a third of one major model's training set while its own data licensing revenue reached 36 million dollars in a quarter against 549 million dollars in advertising. Reddit executives have also been weighing whether continuing to supply Google is worth the declining referral return. Wikipedia recorded an 8% decline in human traffic over a year while total visits rose, because crawler activity more than replaced the people.

Stack Overflow's licensing terms give it far less bargaining power of that kind, and its question flow has now fallen to a level where the flow itself has limited residual value as a data product. What remains is an archive whose most recent entries increasingly describe an older generation of tools.

The developer behaviour driving this is documented in the platform's own survey work. According to figures from the Stack Overflow Developer Survey summarised in PPC Land's coverage of an OECD working paper, the 2025 edition drew more than 49,000 responses across 177 countries, with roughly half of professional developers using or planning to use AI agents. Earlier coverage recorded 84% of respondents using AI tools in their development process, up from 76% the previous year, and 51% of professional developers using them daily. Google responded to the same shift by opening a documentation API covering roughly 40 million documents for machine consumption, on a 24-hour re-indexing cadence.

What marketers can read into 218 rows

For media buyers and publisher-side operators, the value of this series is diagnostic rather than tactical. It shows what the end state of an informational-query business looks like when the answer layer moves upstream, without the confounding variables that make publisher traffic data arguable. Google has disputed the interpretation of traffic-decline research repeatedly, arguing that clicks arriving through AI features carry higher intent. That argument concerns the quality of remaining clicks. It does not address a case where the behaviour producing the click has been substituted entirely.

The distinction between substitution and dilution is the one that matters for planning. A 39.8% click reduction is a monetisation problem. A 99.3% reduction in the input that made a site worth visiting is a category problem, and the categories most exposed are the ones where a machine-generated answer is functionally equivalent to the original: reference material, technical documentation, definitional content, how-to material. Health and education verticals have already registered the sharpest declines in the ad-opportunity data.

Stack Overflow's site design changed in February 2026, with a beta at beta.stackoverflow.com opening on February 24 and a new logo appearing the same month. The company retains Prashanth Chandrasekar as chief executive, a position he has held since 2019, and remains a Prosus asset. Neither Stack Overflow nor Prosus has published a public statement addressing the July 2026 volume figure.

Timeline

Summary

Who: Daniel Lockyer, a developer posting on X, published the comparison. The subject is Stack Overflow, the programming question-and-answer site created by Jeff Atwood and Joel Spolsky and owned since June 2021 by Prosus, with Prashanth Chandrasekar as chief executive. The data comes from the Stack Exchange Data Explorer, the platform's own public query interface.

What: Monthly new-question volume on Stack Overflow reached 1,442 in July 2026, against a March 2014 peak of 207,204, a decline of 99.3%. Annual volume fell from 1,336,266 questions in 2022 to 108,981 in 2025, with 12,322 recorded across the first seven months of 2026. The dataset counts undeleted questions only, so the historical peak is understated and the true decline is likely steeper.

When: The post appeared on August 7, 2026. The dataset spans July 2008 to a partial August 2026 across 218 monthly rows. The query behind it was created on January 3, 2026.

Where: The figures come from data.stackexchange.com and were circulated on X. Stack Overflow operates globally and exited its physical datacenter in December 2025.

Why: The series matters beyond software because it isolates a behaviour change that publisher traffic data usually obscures. Informational queries that once produced a page view now terminate inside an answer interface. Independent research places the click reduction from AI Overviews between 34.5% and 58% depending on method and period, and consequences have reached closures and job losses at Bauer Xcel Media, Overfishing.org and elsewhere. Stack Overflow shows the same mechanism at the point where substitution is close to complete, which makes it a reference case for any category built on reference content.