On November 10, SparkToro published new research with Similarweb data: visits to the top million websites have fallen 12.2% from their peak in early 2022, while the number of people online grew about 24%. Visits per internet user are down 29.2% (SparkToro). Rand Fishkin's answer to why: "We can't say for certain."
We think much of the answer sits outside search. Put SparkToro's traffic data next to layoff counts, fuel and grocery prices, retail results and app usage, and most of the mystery goes away. What's left is a measurement problem, and that one can be engineered.
Prefer the numbers at a glance? View the infographic, or explore the interactive report and click through every chart.
Why Web Traffic Isn't a Zero-Click Mystery
More people online, fewer visits to the open web
SparkToro's own data cuts against the idea that big tech suddenly swallowed the open web. Traffic share across the top million sites barely moved over five years. The top 100 sites went from about 33% of visits to about 36%, which SparkToro itself describes as stagnation, not consolidation.
Two structural shifts explain more of the per-user drop. First, the denominator. DataReportal's count of internet users rose from 4.95 billion at the end of 2021 to 6.15 billion in 2026. Part of that jump came from DataReportal revising its estimates for India and China in 2025, not only from new people coming online. And much of the real growth came from mobile-first markets, where people spend their time inside WhatsApp, TikTok and YouTube. Similarweb's panel counts browser traffic only, so those users add to the user count without adding open-web visits. Divide a post-lockdown traffic peak by a much larger, app-first user base, and visits per user fall, with or without AI search.
Second, geography. SparkToro calls it "plausible" that traffic to U.S., U.K. and EU sites fell while other regions grew, but its published data has no regional split. That's an open question worth answering before anyone blames AI for all of it.
Layoffs and prices hit both engines of web browsing
Open-web traffic runs on two engines: knowledge workers browsing at their desks during the workday, and households with spare income comparing and shopping at night. From 2023 on, both took a hit at the same time.
U.S. employers announced about 3.3 million job cuts from January 2023 through October 2026, according to Challenger, Gray & Christmas. Technology, the sector that spends its whole day in a browser, cut 134,000 jobs in 2024, 154,000 in 2025 and 172,000 through October 2026, 28% of all cuts announced this year. Every laid-off knowledge worker takes daily research, software comparison, news reading and lunchtime shopping sessions out of the web with them.
At home, prices squeezed the second engine. Regular gas peaked above $5 a gallon in June 2022 and averaged $3.95 for that year, against $3.01 in 2021, according to the EIA. Compounding BLS December-to-December changes, food-at-home prices rose about 26% and shelter about 28% from December 2020 to December 2025. When fuel, rent and groceries eat the paycheck, people stop casual multi-tab browsing and consolidate where they buy.
The retail split: shoppers consolidated where they spend
You don't have to guess how shoppers reacted. Company filings and foot-traffic data show it.
Costco kept growing. Its U.S. comparable sales rose 7.3% in fiscal 2025, excluding gas and currency. Instead of browsing a dozen retail sites, budget-conscious families route groceries, gas and household spending into one trusted membership and its app.
Target went the other way. Its comparable sales fell 3.7% in fiscal 2023, rose 0.1% in 2024 and fell 2.6% in 2025, when transactions dropped 2.2%. Its visits also fell after it announced on January 24, 2025 that it was ending some DEI initiatives, which drew a boycott. Placer.ai measured Target's visits down 9% year over year in February 2025 and 6.5% in March, with seven straight months of declines through August, while Costco's visits rose in each of those months. Analysts disagree on how much of Target's drop was the boycott and how much was broader economic uncertainty. One said the boycotts "might play a minor role."
The bottom line: most people aren't making shopping decisions around corporate politics. They're worried about their jobs, the price of gas and groceries, and getting real value for their money. That pulls traffic out of broad open-web browsing and into a few high-intent destinations.
Discovery moved into apps and social feeds
Similarweb's panel can't see what happens inside apps, and that's where a growing share of discovery now happens.
Meta said in July 2026 that Threads passed 500 million monthly active users. During the 2025 Target boycott, Threads communities like Black Threads shared lists of Black-owned and independent businesses to buy from directly. That spending went from the app straight to checkout on small sites far below the top-million cutoff. Trusted family and lifestyle creators like Ms Rachel can sell out a small children's brand or toy maker through links in the app, with no Google search involved. And attention that once flowed through news portals now lives in Substack newsletters, which reported 5 million paid subscriptions in March 2025, on Bluesky, in Discord servers and in in-app browsers.
The result is fewer exploratory pageviews per person, and more buying concentrated on destinations people already trust.
"We can't say for certain" is a measurement problem
SparkToro tells marketers they don't have the tools to know whether a website change caused an AI visibility lift. That's true if you measure AI search with ad-hoc prompts or open-ended agent sessions, where the same input can return a different answer from run to run. In our own testing, a batched graph step flipped 13.6% of its verdicts between identical runs, and at 300 steps, prompt-only setups passed four runs in a row 0% of the time.
Engineered, isolated steps fix that. Our rebuilt judging step gave the same answer twice on 99.1% of 1,000 keywords. When measurement noise drops to 1% or 2%, a lift in AI citations after you change your site is far more likely to be your work, not the model's randomness. And with real search demand, competitor classification and repeatable AI visibility tracking in one pipeline, you can see whether a traffic dip reflects layoffs and inflation, or a real shift in competitive share.
What This Means for Your Business
The macro picture explains why the web's traffic fell. For your business, the question is narrower and more useful than "where did the web's traffic go?" It's this: when your customers search, do you show up, and do you show up ahead of your competitors?
What the Pressfit Playbook Does
The Pressfit Playbook is the roadmap for AI visibility. It connects three things most teams look at separately:
- What your customers are likely to search for. We start from real search demand and the questions buyers ask, not from a guess about what sounds good.
- Your content against your competitors' content. We compare what you've published with what they've published, page by page and topic by topic, to find where they're winning and why.
- Where each of you shows up. We check how you and your competitors appear for those searches across the major engines: Google, Gemini, ChatGPT, Claude and Perplexity.
Put together, that shows how your competitors show up in front of your customers, and where you don't.
Keywords and content are still the key. AI search didn't replace them. It raised the bar on how clearly your content answers what buyers ask, and how well your site explains who you are. In March 2026, Ahrefs found that 31% of the pages Google AI Overviews cite don't rank in the top 100 traditional results at all (Ahrefs). Ranking alone doesn't get you cited. The right content, structured the right way, does.
That's why the Playbook doesn't end with a report card. It gives you prioritized steps to get there: what to fix first, what to write, what to restructure, and how to close the gap with each competitor. It's the roadmap to showing up, and to beating the competition when you do.
Measuring What Works Is Different Now
Here is where most teams get stuck. They do the work, open Google Analytics, and see nothing.
Two things changed.
The analytics tag doesn't see everyone. The Google Analytics tag runs in the visitor's browser, and a growing share of browsers stop it. Brave blocks it by default, ad blockers block it, and consent banners keep it from firing until someone clicks accept. The detailed data many marketers were used to isn't there anymore. Moving measurement server-side helps, because your server and your CDN see every request, whether or not a script was allowed to run.
AI assistants often don't send a referrer. When someone clicks a link inside the ChatGPT or Claude app, copies a link out of an answer, or opens it in an in-app browser, the visit usually arrives with no referrer and lands in "Direct." And much of what AI engines do never sends a visit at all: they fetch your page, read it and answer the question on their side.
So you have to measure at a different layer. Over 30 days, Google Analytics showed us about 6,500 human visits to Pressfit.ai with almost no AI attribution. Our firewall logged 7,321 requests from AI and answer-engine crawlers in the same window, including 127 live ChatGPT-User fetches and 57 live Claude-User fetches, each one triggered by a person asking an AI a question that needed our page to answer.
Our companion post walks through exactly how to measure it, layer by layer: Measuring AIO and GEO: Why Prompting Agencies Fail and How Deterministic Engineering Closes the Analytics Blind Spot.
The How-To, Then the Proof
The Playbook gives you the how-to: the prioritized steps to show up where your customers search and to beat your competitors there. When you're done, we measure it, with server-side and edge data, not a cookie report that can't see the traffic you earned.
Stop guessing why traffic moved. Get the roadmap, do the work, and measure what changed.
Need help putting it to work? Contact our team at redevteam@pressfit.ai.