The archive that stopped receiving questions

The number folkfox counted itself#

A fox does not mourn a vanished hedgerow. It reads the wind, finds the next one, and moves before the frost. Marketers manage that move far less often, because the hedgerow they hunt in belongs to somebody else and it thins slowly enough to pass for weather. The plainest proof on the open web is not a publisher at all. It is a question archive.

Stack Overflow received 1,476 new questions in July 2026. In March 2014, its peak month, it received 207,199. That is a fall of 99.29%, and folkfox counted it rather than quoting it, using the public Stack Exchange API on 9 August 2026, counting questions created inside each calendar month.

One property of that count matters more than it looks. The endpoint returns questions not deleted as at the moment of the query, so any historical month drifts downwards as moderation catches up. Deletions subtract. They never add. Hold on to that, because it does a lot of quiet work in the next section.

Read the shape rather than the shock. The archive did not lose its readers first, it lost its writers. Contributors stopped arriving because the answer they wanted arrived somewhere else, and once the supply side stops showing up, no amount of clever distribution can refill the well. That is a content distribution strategy problem wearing a developer-tooling costume.

Three numbers, one month, and why you cite the query#

The figure circulating this week is 1,442 for July, published by PPC Land on 8 August 2026. The folkfox query returns 1,476. Both numbers are honestly arrived at, and the gap between them is the genuinely interesting part.

Here is the awkward arithmetic. Deletions can only push a monthly count down over time, never up, so a later query returning a higher July figure than an earlier published one cannot be drift. It points to a different query definition: a timezone boundary, a different treatment of deleted questions, or a pull taken part-way through the month. Different definition, different number, same direction of travel.

One claim does not survive contact, though. A line repeated widely this week says July was a 34% rise on June, and the first significant month-on-month increase since September 2025. On the folkfox count, June was 2,017 and July was 1,476, a fall of 27%. The 99% headline holds firmly. The monthly direction does not.

December 2025 shows the same instability from the other side. The devclass report on 5 January 2026 put the month at 3,862. PPC Land put it at 3,314. The folkfox query today returns 4,217. Three counts, one calendar month, and nobody being careless.

The shape survives all three. Writing on 15 May 2025, The Pragmatic Engineer reached the same conclusion through a different route, a Stack Exchange Data Explorer query shared by Marc Gravell, with monthly units and roughly 200,000 questions at the 2014 peak. Two methods, one trend, which is what corroboration is supposed to look like.

So the durable lesson is dull and it is worth more than the drama: cite the query and the date, never the number alone. Any content distribution strategy that quotes a competitor's stat without a method is one screenshot away from a correction, and corrections cost more scent than they save time. That habit is the cheapest credibility a content marketing agency can buy.

The publisher rebuilding revenue away from sessions#

An archive of unanswered questions is a vivid anecdote. Audited accounts are better evidence, and one set landed this month with the numbers a content distribution strategy actually needs.

Reporting People Inc.'s second-quarter results on 4 August 2026, Digiday put Google search at 21% of the publisher's traffic, down from 25% the previous quarter, with search traffic down 40% year on year and core sessions down 22%. Those are not panel estimates. They sit inside earnings.

Look at what the company did about it rather than what it said. Sessions shrank and non-session revenue grew, so the revenue line stopped depending on the traffic line. That is a content distribution strategy decision taken at board level, and it is the generalisable move in this whole story.

That is the trap in one sentence. The crawler that feeds the answer engine is the crawler that feeds the index, so blocking one blocks both, and the publisher keeps feeding a machine that repays it in a fraction of the clicks. Anyone weighing that call should read the crawler names carefully first, which is the argument folkfox made in blocking AI crawlers and the Googlebot risk.

The behaviour behind the referral traffic decline#

The best measurement of that behaviour is not a survey. On 22 July 2025 the Pew Research Center published browsing-tracker data from 900 US adults on the Ipsos KnowledgePanel Digital who consented to share their March 2025 web activity, covering 68,879 unique Google searches, of which 12,593 produced an AI summary. Observed clicks, not remembered ones.

The findings are blunt. A traditional result was clicked on 8% of visits with an AI summary present against 15% without, links inside the summary were clicked on 1% of visits, 18% of all searches produced a summary, and users ended their browsing session on 26% of pages carrying one against 16% of pages without, all per Pew. Zero click search stopped being a theory the moment somebody watched the tab close.

The wider picture points the same way. The Reuters Institute's Digital News Report 2026, published 16 June 2026 across 48 markets with roughly 2,000 respondents each, found weekly use of AI chatbots for news rising from 7% to 10% year on year, reaching 16% of under-35s, with the UK lowest at 4%. In the same report the Institute reports a third-party analytics dataset, not its own survey, showing Google organic traffic to over 2,500 sites down 33% globally and 38% in the United States between November 2024 and November 2025.

Worth saying plainly: none of this August movement belongs to a confirmed Google ranking update. Google has confirmed no ranking update since the June 2026 spam update finished on 26 June, so a content distribution strategy built to chase the next algorithm announcement is chasing a scent that is not there.

What the crawl costs when nobody comes back#

Referral traffic falling is only half the ledger. The other half is what the machines take on the way past, and one operator has counted it in unusual detail.

On 8 August 2026, PPC Land reported figures from Nick Gray of PatronView, drawn from a year of server logs and firewall data across a 1.5 million page database: 35,000 crawls by Claude-SearchBot for every human visitor referred, against 46 to 1 for Googlebot and 406 to 1 for Bingbot. Amazon's Amzn-SearchBot reached roughly 117,000 requests a day before it was blocked, serving 4.63 GB a week to the crawler against 175 KB to humans, with a CAPTCHA solve rate of 0.24%, or 252 of 106,437 challenges.

Attribution matters here and folkfox will not fudge it. The underlying PatronView post returned HTTP 403 when folkfox tried to read it, so those figures are cited as PPC Land reporting PatronView's numbers, from one operator's own logs, not as an independent audit. Take the ratios as a signal of scale rather than an industry benchmark.

The macro picture rhymes. Writing on 4 August 2026, Digiday set out Cloudflare data showing more than half of all web traffic is now bot-based, AI scraper traffic up 597% between January and December 2025, AI-driven traffic up 187% across 2025, and scraping attacks touching nearly 20% of the median organisation's site traffic. Half the hedgerow is machines.

The bill that is not law yet#

Legislators noticed. New York's Stealth Crawler Prohibition Act carries penalties of up to $15,000 per day per violation, as the same Digiday explainer notes. It is a bill, not a statute. The New York State Senate record for S9934A shows it passing the Senate on 2 June 2026 by 60 votes to 1 and the Assembly on 5 June 2026, then being delivered to the Governor with no signature, no veto and no chapter number recorded. It takes effect 90 days after becoming law, if it becomes law.

For a marketing team the practical read is simple. The cost side of publishing is rising while the referral side falls, so any content distribution strategy that measures success in sessions alone is measuring the half of the ledger that is shrinking fastest. The publisher economics underneath that squeeze ran through folkfox's piece on a $54.8m write-down in health media.

Newsweek notes that over 60 percent of Google searches now end without a click, the ground truth under the publisher revenue story.

A content distribution strategy that owns more than it rents#

Enough diagnosis. The useful question is which surfaces a brand actually holds, and which it merely borrows from a landlord who can change the locks between quarters.

The centre is unglamorous and it is yours: an email list, a well-kept archive, a product surface, a community you host. The edges are wonderful while they last. A content distribution strategy that treats the edges as infrastructure will keep rebuilding for whoever owns them, and Stack Overflow is what the far end of that road looks like when the supply side quietly stops arriving.

Measure the answer layer, then measure it again#

Something to measure it with shipped on 6 August 2026. Cloudflare added an AEO Visibility Dashboard reporting Citation Rate, Mention Rate, Prominence, Share of Voice and Industry Fit. Its engineering write-up by Jack Galilee explains the method: repeated realistic category prompts, scored by exact text analysis rather than a model marking its own homework, against a pre-computed category baseline rather than live queries on every scan.

Read the limits before the marketing. It covers Anthropic's Claude and OpenAI's GPT only, with no Gemini, no Perplexity and no Copilot, and it is early access by request. Pricing is not stated anywhere, so treat it as undisclosed rather than free. As Cloudflare's Chief Strategy Officer Stephanie Cohen put it, being discoverable used to mean ranking on a page, and that is not enough any more.

The measurement gap is real, and so is the temptation to fill it with whatever dashboard arrives first. Better to pick one measure of the answer layer, hold it for two quarters, and let a folkfox SEO and GEO practice argue about the method rather than the vendor. A brand that only exists inside somebody else's summary has a visibility problem no dashboard can fix on its own.

Five honest moves for the next two quarters#

First, own an audience you can reach without permission: a list, an app, a community. Second, publish where the platform itself is the index, because platform properties now carry weight a website cannot, an argument folkfox set out in where your best content actually lives. Third, write the pages a machine can quote cleanly, which is the point of the English page the answer layer actually retrieves.

Fourth, move a slice of revenue off sessions entirely, exactly as People Inc. has. Fifth, restate the promise so it survives being summarised by a stranger, which is brand strategy work rather than a channel tweak. None of that is a rebrand, and none of it needs a new platform. It needs a content distribution strategy with a centre.

The fox in the sieve is the right picture for this year. Traffic pours through, the sieve keeps the few solid things worth keeping, and a smart content marketing strategy is mostly a set of decisions about what the sieve is made of. If you want that built rather than described, start a conversation with folkfox and bring your last four quarters of numbers with you.

Frequently asked questions#