We Re-Checked 4,527 Business Websites 3 Months Later. Here Is What Died and What Changed.

The setup

Our shared observation log records what our enrichment pipeline saw on each business website and when it saw it. For this study we selected 4,527 domains whose most recent observation was between 60 and 180 days old (median gap: 106 days) and re-fetched every one of them on the same day. Each domain belonged to a business found through Google Maps sourced prospecting, and, importantly, every site in the cohort loaded successfully at baseline. Whatever we found now is what a quarter of real time does to working business websites.

One methodology choice matters up front: our re-check used a deliberately conservative plain-HTTP fetcher with no headless browser. Sites that answered with bot walls or render entirely in JavaScript (1,739 of them, 38.4 percent) were excluded from comparison rather than measured badly. The numbers below are computed only on outcomes the method can actually prove.

Finding one: about 1 in 47 working sites stopped loading

Of 4,527 sites that all worked a quarter ago, 96 (2.1 percent) now fail with a fatal error. Within that group, 39 sites (0.9 percent of the cohort) returned a definitive 404 or 410: the page is gone, not hiding. The rest failed with server errors, refused connections, and assorted terminal responses.

Scale that against any list you bought or scraped in spring. A 5,000-row list from April carries roughly a hundred rows today that point at nothing at all, before you evaluate a single email address or data field. Dead sites are the loudest form of data decay because they take every enriched field down with them.

Finding two: 1.9 percent of readable sites switched a core platform

Among the 2,674 sites we could fully re-read and compare, 50 (1.9 percent) had switched at least one core detected platform within the window: a CMS, booking tool, CRM, live chat widget, or email marketing platform. We counted only clean A-to-B switches, where both the old and the new tool were positively identified.

The breakdown of the 51 switch events: content management systems 21, booking tools 15, CRM platforms 9, live chat 5, email marketing 1. The single most visible migration pattern in the sample: booking flows moving from Calendly to GoHighLevel, which accounted for 9 of the 15 booking switches. If you sell against or alongside either product, that is a trend worth knowing, and it was invisible until two snapshots were put side by side.

Why this matters for outreach: platform detections are pitch fuel. "I noticed you run X" is only a good opener while they still run X. At nearly 2 percent core-platform churn per quarter, technographic data from last year is fiction at the edges.

The numbers we refused to publish

Our diff engine also produced thousands of "added" and "removed" tool detections across the same window. We are not publishing any of them as churn, and the reason is the entire point of this blog: the baseline observations were made by an older generation of our detector pipeline, which could not yet see some of the things the current one sees. A tool "appearing" between an old detector and a new detector is not a business changing. It is a ruler changing length.

Our own diff engine flags every one of these cross-version comparisons and holds removal events as unconfirmed until a same-version pass verifies them, so none of that noise reached this analysis, or our users' change feeds. We mention this because it is exactly the kind of confound that quietly inflates statistics in this industry. If a vendor ever shows you a churn or accuracy number, ask whether their measuring instrument changed between the two measurements.

Add the layers together

This study measures the website layer. In our companion study, we measured the email layer: 14.2 percent of decisively checkable scraped emails were dead, rising from about 10 percent when fresh to 13 to 17 percent at three to four months of age. Stack the layers and the compounding is brutal for any static list: within one quarter, roughly 2 percent of your targets' sites vanish, another 2 percent of the readable ones swap a core tool, and one in seven checkable emails on the older rows bounces.

That is the argument, backed by measurement, for treating lead data as something you observe continuously rather than something you buy once. It is why our platform re-observes businesses through live search and enrichment, records changes as labeled signals, and verifies emails at send time instead of shipping a static export with an adjective attached.

Methodology notes for anyone citing this: cohort of 4,527 domains with successful baseline observations 60 to 180 days old (median 106 days), re-fetched July 30, 2026 with a plain-HTTP fetcher; bot-walled and JavaScript-only sites (38.4 percent) excluded from comparison; platform switches counted only where both old and new values were positively detected; cross-detector-generation additions and removals excluded by design. Cite freely with a link.