Nobody Reads the Documentation Anymore. Here Is What That Cost Three Companies.

Stack Overflow, Chegg, and Tailwind Labs all report fewer visits to the pages meant to answer a question, even as the products behind those pages kept growing. Stack Overflow's own data puts new questions at 1,125 in August 2026, the lowest full month in the site's history outside its 2008 private beta.¹ Chegg's revenue fell 51 percent year over year last quarter, to $51.8 million.² Tailwind Labs' founder says the company laid off three-quarters of its engineering team in early January 2026, and that revenue was down close to 80 percent.³

These are not the same kind of evidence. Chegg's come from audited SEC filings, the strongest proof here. Stack Overflow's come from a live public database anyone can query.¹ Tailwind Labs' come from one founder's account of his own company, with no audit behind it. None deserve the same weight.

The short version

• Chegg's revenue fell 39 percent in FY2025 and another 51 percent in Q2 2026, in filings naming AI Overviews a stated business risk.⁴⁵
• Stack Overflow's live data shows new questions down from a March 2014 peak of 207,172 to 1,125 in August 2026, over 99 percent, though the decline started years before ChatGPT.¹
• Tailwind Labs' founder says docs traffic is down about 40 percent since early 2023 and revenue close to 80 percent, in an unaudited GitHub comment.³
• The circulated "lowest month" figure for Stack Overflow is wrong. It is 1,125 in August 2026, not 1,076 for June.¹
• Ahrefs' 58 percent click-through figure is a correlation against a modeled forecast, not a measured before-and-after.⁶
• Google says organic clicks have stayed relatively stable and traffic is shifting between sites, not vanishing.⁷
• Two studies from the same firm, Seer Interactive, disagree on whether click-through on non-AI-Overview searches is falling or rising.⁸⁹
• Alphabet's Search revenue rose about 17 percent year over year the same quarter Chegg's fell by half.¹⁰

Three companies, three signals of documentation traffic decline

Chegg sits at the top. Its numbers come from a Form 10-K, a Form 10-Q, and a Form 8-K, filed with the SEC and checked by a public company's own auditors.⁵²¹¹ The dollar figures are not in dispute; what Chegg attributes the fall to is its own read on causation.

Stack Overflow sits second. Its own public API returns the exact monthly count of new questions, back to 2008.¹ Not audited, but live and reproducible; anyone can query it.

Tailwind Labs sits third, well below the other two. Everything reported about its numbers comes from one comment founder Adam Wathan posted to a GitHub pull request on January 7, 2026.³ No second source confirms his percentages. His numbers carry the weight of one person's word about his own company. Trusting a 10-Q over a GitHub comment is reasoning correctly.

Stack Overflow, measurable from its own data

The Stack Exchange API returns the exact number of new questions asked on Stack Overflow, month by month, queried directly for this post on September 21, 2026.¹

The peak was March 2014, at 207,172 new questions. Annual volume held in the millions for years: 2,183,175 in 2016, still 1,335,310 in 2022. ChatGPT launched publicly on November 30, 2022,¹² and that month Stack Overflow logged 109,076 questions, its last normal month. 2023 fell to 787,778 and 2025 to 109,763, roughly a twentieth of the 2016 total.¹

Recent months keep falling: March through August 2026 read 3,177, 2,545, 2,076, 1,730, 1,185, 1,125, the lowest full month outside the July 2008 private beta and 99.46 percent below the 2014 peak.¹

One correction worth stating outright: several outlets cited June 2026's 1,076 as the low point, but the API shows June at 1,730, above August's 1,125, a straight decline rather than a dip and recovery. And 2016's volume was already below the run rate the 2014 peak implied, years before ChatGPT existed.

Chegg, in its own SEC filings

Chegg is publicly traded, so a misstated filing is a securities problem.

Total net revenue fell from $617.6 million in FY2024 to $376.9 million in FY2025, a 39 percent decline, with a net loss of $103.4 million.⁴ Q2 2026 revenue was $51.8 million, down 51 percent from $105.1 million a year earlier, though the net loss narrowed to $3.0 million and Chegg's newer Skilling division grew 2 percent to $17.5 million.²

Chegg's 10-Q attributes the drop to lower subscription and advertising revenue, "primarily related to reduced traffic which led to fewer subscribers."² That is Chegg's own attribution, from a company with an antitrust suit against Google and so a financial interest in the framing. Its 10-K names a cause directly: "Google's roll out of Artificial Intelligence Overview (AIO) has created and is expected to continue to create headwinds for our industry and our business, most notably reductions in traffic to our website and customers subscribing to our services."⁵ Its investor release is softer, naming no AI Overviews, saying only "changes in search interfaces continue to impact our traffic."⁴

Chegg cut about 388 employees, 45 percent of the workforce, on October 21, 2025; CEO Nathan Schultz stepped down, and Dan Rosensweig resumed as President and CEO.¹¹ It filed an antitrust complaint against Google on February 24, 2025, over the AI Overview expansion.² The case is pending; no docket, judge, or ruling is public record.

Tailwind Labs, self-reported

Tailwind Labs is private and publishes no financial statements. Everything known about its recent numbers comes from four comments Adam Wathan posted to one GitHub pull request in January 2026.

The day before the layoffs became public, Wathan closed a pull request proposing an llms.txt endpoint for the Tailwind CSS docs. His reasoning, posted January 6, is the clearest statement of the mechanism in this evidence base: "making it easier for LLMs to read our docs just means less traffic to our docs which means less people learning about our paid products and the business being even less sustainable."³

The next day he posted the numbers that circulated widely: docs traffic "down about 40% from early 2023 despite Tailwind being more popular than ever," and revenue "down close to 80%."³ The traffic baseline, early 2023, is roughly three years before the comment, and it is his own unaudited estimate; no base figure or timeframe was given for the revenue claim. In the same comment: "75% of the people on our engineering team lost their jobs here yesterday because of the brutal impact AI has had on our business."³ The layoffs happened the first week of January; his comments do not resolve whether it was the 5th or the 6th. It would be easy to read Wathan as opposed to LLM-friendly documentation; he was not, writing the same day that he wanted to offer LLM-optimized docs once he could do so safely, but could not prioritize it yet.³

A 40 percent drop at the top of a documentation funnel does not, on ordinary funnel math, produce an 80 percent drop in revenue. Wathan never claims it does, and nothing in the record explains the gap. In September 2026, Tailwind Labs folded its commercial business into Shopify, closing new sign-ups but keeping the framework free; that deal is a separate story.

Where the traffic actually went

The clearest answer comes from a Pew Research Center browsing-panel study: 900 US adults with a tracker installed, watching 68,879 actual Google searches in March 2025, of which 12,593 produced an AI summary.¹³ Users clicked a traditional result about 8 percent of the time when a summary appeared, versus about 15 percent when it did not. The clearest number in this research base: only about 1 percent of visits to a page with an AI summary ended with a click on a source the summary itself cited. Being cited and being visited are not the same event.

Server logs tell a similar story. Mintlify, which hosts documentation for many software companies, reported AI agents at 66 percent of measured traffic in July 2026, 213 million agent requests against 105 million human page loads.¹⁴ By August, agent requests reached 257 million against 131 million human page loads.¹⁵ Page loads rose even as agent share climbed.

Cloudflare's crawl-to-referral ratios show the same shift. Anthropic's ratio was measured at 70,900 to 1 for one week in June 2025.¹⁶ By June 2026, 52 percent of all crawler requests were for AI training, up from 22 percent in spring 2025.¹⁷ Treat any single figure as a snapshot, not a constant.

Which of these numbers you should not trust

Ahrefs found an AI Overview's presence "correlates with a 58% lower average clickthrough rate," from 300,000 keywords and Search Console data.⁶ That gets repeated as "AI Overviews reduce clicks by 58 percent," which is not what Ahrefs measured: the gap compares December 2025's actual position-one click-through rate against a modeled forecast without AI Overviews, a correlation against a counterfactual, not a before-and-after.

Seer Interactive has published two versions of the same study that disagree with each other. The first, 3,119 terms across 42 organizations, found click-through on AI-Overview queries fell 61 percent while a control group with no AI Overview fell 41 percent.⁸ A later update, on 53 brands and 5.47 million queries, found the opposite for that same control group, a 36 percent rise.⁹ Two good-faith studies from the same firm cannot both be right about which way the baseline moves.

What Google says about it

Google disputes the dramatic-decline framing directly. Liz Reid, Google's VP and Head of Search, wrote that "total organic click volume from Google Search to websites has been relatively stable year-over-year," with click quality up. Her clearest line: "While overall traffic to sites is relatively stable, the web is vast, and user trends are shifting traffic to different sites, resulting in decreased traffic to some sites and increased traffic to others."⁷ The post has no numbers of its own.

Some independent research backs a version of that. Graphite, using Similarweb data validated against Search Console across eight sites, found Google organic traffic down only about 2.5 percent year over year across more than 40,000 large US sites, with the top 10 and smallest sites gaining while sites ranked roughly 100th to 10,000th took the steepest losses.¹⁸ Author Ethan Smith disclosed a conflict of interest: "it is in the financial interest of the author to argue that SEO is a large channel and is not decreasing."¹⁸

A separate study found the opposite, with the same uncertainty about cause. Reuters Institute, using Chartbeat data across 2,576 sites, found Google organic referrals to news sites down 33 percent globally and 38 percent in the US, but stated plainly: "it is not clear how much of this is down to AI overviews."¹⁹

The broadest counterweight is Alphabet's own earnings: Q2 2026 Search and other revenue was $63.3 billion, up about 17 percent year over year.¹⁰ Even Wikimedia, reporting an 8 percent pageview decline, credited "generative AI and social media" together, not AI alone.²⁰

What to measure instead of sessions

Total sessions is the wrong number to watch alone. Query your own numbers directly instead of trusting a circulated figure, the way this post did with the Stack Exchange API, which caught the wrong "lowest month" claim. Validate any third-party estimate against your own analytics the same way; Graphite's 2.5 percent figure is trustworthy because it was checked against real Search Console data, with the correlation published alongside it.

Split agent traffic from human traffic; Mintlify's data shows agent share and human volume can move independently in the same month (both rose from July to August 2026), which a single traffic total cannot show you. And track citation rate against click rate, not as one metric: Pew's 1 percent figure is the clearest evidence that being mentioned by an AI answer and being visited by a human are different outcomes. Our free AI visibility checker gives a starting read on where your pages get cited.

Frequently asked questions

Why is documentation traffic declining?

AI answer engines increasingly answer directly rather than send the reader to the page that explains it. Pew Research found only about 1 percent of AI-summary visits end with a click on its cited source, and more requests hitting documentation servers now come from agents, not people.

Is Stack Overflow really dying?

New questions hit 1,125 in August 2026, the lowest full month outside the 2008 private beta, against a March 2014 peak of 207,172. The decline predates ChatGPT by years; AI accelerated it rather than starting it.

Did AI Overviews cause Chegg's revenue decline?

That is Chegg's own claim, made in SEC filings and an antitrust suit against Google, not an independently verified fact. Its 10-K names AI Overviews a business risk; its investor release is softer and does not name them at all.

How much did Tailwind Labs lose to AI?

Founder Adam Wathan wrote in a January 2026 GitHub comment that docs traffic was down about 40 percent since early 2023 and revenue close to 80 percent, with 75 percent of engineering laid off. None of it is audited; the company is privately held.

What should I measure instead of total sessions?

Split human page loads from AI agent requests in your server logs, track citation rate against click rate, and check any third-party traffic estimate against your own Search Console data.

Figures checked on 22 September 2026.