
How Web Analytics Has Changed
On this page
- 2017 to 2020: Safari shortened cookie life and changed returning users
- 2021: Apple changed ad identifiers and email opens
- 2023 to 2024: Universal Analytics ended and GA4 redefined bounce rate
- 2023 to 2025: Chrome reversed its third-party cookie plan
- 2024: Consent requirements changed audience coverage and conversion labels
- 2026: AI referrals gained a channel and Ads collection control changed
- Current rules: log changes, reset baselines and reconcile counts
Analytics & Data
On 5 May 2026, bot mitigation vendor Imperva reported that bots accounted for over 53% of all web traffic. In Iron Goo's own server logs, 31.5% of 10,044 requests over about 41 hours carried a self-identifying automated user agent. How web analytics has changed is mostly a story of definitions and visibility changing under familiar chart labels. Browsers cap cookies, Apple hides some signals, Google has redefined sessions and bounce rate, and consent affects what gets reported. A five-year chart can therefore contain several different metrics with one name.
The Iron Goo count comes from nginx logs for irongoo.com, from 00:00 UTC on 28 October to about 17:00 UTC on 29 October 2026, excluding requests from our own server. A user-agent string match identified 3,161 automated requests, including 844, or 8.4%, bearing a named AI or large-platform crawler user agent. These are requests, not visits, and the automated count is a lower bound because bots posing as browsers are not counted.
Most of those automated requests likely never run the JavaScript that sends data to Google Analytics 4, or GA4. Server logs and GA4 therefore count different populations. GA4 also excludes known bots using Google research and the International Spiders and Bots List; you cannot disable that exclusion or inspect how much traffic it removed.
2017 to 2020: Safari shortened cookie life and changed returning users
On 5 June 2017, WebKit introduced Intelligent Tracking Prevention, or ITP. For domains classified as cross-site trackers, cookies could be used in a third-party context for 24 hours after user interaction, then were partitioned; without interaction for 30 days, they were purged. The practical effect was to limit the persistence of cross-site recognition used for advertising measurement.
On 21 February 2019, ITP 2.1 capped persistent cookies created through document.cookie at seven days. Google Analytics sets its analytics cookie through JavaScript, so a Safari visitor who returns after more than seven days without an intervening visit can be counted as a new user. That is an inference from the cookie rule: new-user counts can rise and the returning-user share can fall even when the same people come back.
On 24 March 2020, WebKit announced that cookies for cross-site resources were blocked by default across Safari. That further restricted cross-site identification and the attribution that depended on it. A change in a browser rule can therefore move an acquisition report without a corresponding change in demand.
2021: Apple changed ad identifiers and email opens
From 26 April 2021, apps on iOS 14.5 had to request permission through App Tracking Transparency to track users or access their device advertising identifier. Apple's developer notice states: "Unless you receive permission from the user to enable tracking, the device's advertising identifier value will be all zeros and you may not track them."
The missing identifier reduces what an ad platform can observe across apps. Our reading is that platform-reported conversions from iOS app users consequently became partial and could include modeled results, so a reported drop does not by itself establish that orders fell. On Meta's 2 February 2022 earnings call, CFO David Wehner estimated the 2022 revenue impact of Apple's iOS changes at "on the order of $10 billion," as reported by Fortune. That was Meta's revenue estimate, not a measured number of lost conversions.
Apple announced Mail Privacy Protection on 7 June 2021. Apple says it downloads remote content privately in the background; our reading is that this can register an open even when nobody reads the message. An open-rate increase after adoption can reflect that preload; report clicks and replies alongside opens, since the protection does not extend to links.
2023 to 2024: Universal Analytics ended and GA4 redefined bounce rate
Standard Universal Analytics, or UA, properties stopped processing data on 1 July 2023. Analytics 360 properties with a current 360 order could process data longer, but access to the UA interface and API ended for everyone from the week of 1 July 2024. A chart that joins UA and GA4 data crosses a change in measurement as well as a change of product name.
Bounce rate is an especially easy place to make a false comparison. UA counted a session with one page view as a bounce. GA4 counts a session as engaged if it lasts longer than 10 seconds, has a key event, or has at least two page or screen views; its bounce rate is the percentage of sessions that were not engaged.
A bounce is a session with one page view. A session restarts at midnight or when new campaign parameters are encountered.
A bounce is a session that was not engaged. Sessions do not restart at midnight or when new campaign parameters are encountered.
Say a person views one page for 11 seconds and leaves. The hypothetical visit is a UA bounce because it has one page view, while it is a GA4 engaged session because it lasts longer than 10 seconds. The same behavior can lower the reported bounce rate after migration without any improvement to the page.
Session totals can differ for a separate reason: one visit spanning midnight could be split into two UA sessions and remain one GA4 session. Google's comparison of the two products says user and session discrepancies of up to 20%, and pageview discrepancies of up to 10%, are expected. Treat those as context for an investigation, not a guarantee that every difference inside those limits is harmless.
2023 to 2025: Chrome reversed its third-party cookie plan
On 14 December 2023, Google announced a Tracking Protection test for 1% of Chrome users starting 4 January 2024. At the time, it planned to phase out third-party cookies for everyone in the second half of 2024. That test could reduce cross-site measurement for its participants, but the announced path did not become the final policy.
On 22 July 2024, Google changed direction: "Instead of deprecating third-party cookies, we would introduce a new experience in Chrome that lets people make an informed choice". On 22 April 2025, Google said it "will not be rolling out a new standalone prompt for third-party cookies" and would maintain its existing choice approach.
As of 29 October 2026, third-party cookies still work in Chrome by default, with a user choice in settings. Chrome Incognito blocks them by default, and Safari has blocked cross-site cookies by default since March 2020. Check the browser and context before attributing a tracking change to cookie removal.
On 17 October 2025, Google announced the retirement of most Privacy Sandbox technologies, including Attribution Reporting API, Topics, Protected Audience and IP Protection. CHIPS and FedCM continue, and Private State Tokens are maintained. The distinction matters when planning future attribution: a proposal from the original cookie-removal plan is not automatically a dependable future measurement source.
2024: Consent requirements changed audience coverage and conversion labels
Google's March 2024 EEA consent requirements affected which users could enter advertising audiences. From early March, unless action was taken, GA4 audiences used in linked advertising products included only users outside the EEA. Personalized advertising and remarketing required affirmative consent signals, and Customer Match lists could degrade from March without consent.
Consent mode v2 is the industry name for the version with ad_user_data and ad_personalization signals. It is a configuration for communicating consent choices, not a single law or a single Google deadline. The details of consent settings and their legal context belong with your privacy, consent and first-party data decisions.
Without behavioral modeling, events from users who decline analytics storage are not reported. Google's behavioral modeling requirements include at least 1,000 events per day with analytics_storage denied for at least 7 days and at least 1,000 daily users with analytics_storage granted for at least 7 of the previous 28 days. A smaller company may not meet both thresholds, so it should not assume a model fills every gap.
If a consent change reduces reported sessions while recorded leads remain steady, the reported lead conversion rate can rise because its denominator shrank. That is a measurement change, not proof that the site became more persuasive. The same caution applies when bot filtering changes a denominator; reading data without fooling yourself examines the failure of single numbers in more detail.
On 21 March 2024, Google announced that conversions in existing Analytics reports and explorations would become key events. The rename changed the Analytics label, not historic counts; marking an event as a key event does not change historic data. Google now uses conversion for ad measurement and bidding, so record which definition a dashboard uses before comparing its columns.
2026: AI referrals gained a channel and Ads collection control changed
On 13 May 2026, GA4 added the AI Assistant channel to its default channel group. Google's release notes describe traffic from chatbots such as ChatGPT, Gemini and Claude, while its channel definition includes arrivals from sources such as Deepseek, Copilot and Grok. Traffic from AI Overviews and AI Mode remains in Organic Search.
Whether older data is reclassified is unknown. Our inference is that Referral can fall when traffic that would previously have appeared there is classified as AI Assistant, even if the underlying visits have not fallen. Check the two channels together around the change before concluding that referring sources stopped sending visitors.
On 15 June 2026, Google Analytics moved control of Google Ads cookie and identifier collection from the combination of Google Signals and Consent Mode to Consent Mode within Google Ads. Google did not state an effect size. If an Ads-related count or coverage ratio shifts around that date, check the settings before assigning the movement to campaign performance.
Bot estimates also continued to rise in vendor reporting. Imperva's 2025 report said automated traffic reached 51% of web traffic; its 2026 announcement reported over 53%. Neither figure is a substitute for measuring your own requests, and neither tells you how many sessions appeared in your GA4 property.
- 5 June 2017Safari introduces ITP
Cross-site tracker cookies receive limits that can shorten recognition used in advertising attribution.
- 21 February 2019Safari caps JavaScript-set cookies
A seven-day cap can increase reported new users when returning visitors lose their analytics identifier.
- 24 March 2020Safari blocks cross-site cookies
Default blocking further restricts cross-site recognition used by attribution reports.
- 26 April 2021Apple requires tracking permission
Unavailable advertising identifiers reduce observed cross-app signals used in ad conversion reports.
- 7 June 2021Apple announces Mail Privacy Protection
Background loading can inflate recorded email opens without a person reading the message.
- 1 July 2023Standard UA processing stops
The move to GA4 puts different session and bounce definitions under familiar report labels.
- March 2024Google changes EEA audience requirements
Missing consent signals can reduce advertising audience coverage and reported analytics activity.
- 21 March 2024Analytics conversions become key events
The report label changes; the rename itself does not change historic counts.
- 22 July 2024Chrome changes its cookie plan
Google reverses its planned third-party cookie deprecation, changing assumptions about future attribution.
- 13 May 2026GA4 adds the AI Assistant channel
Some traffic may appear under a new channel, potentially reducing Referral through classification alone.
- 15 June 2026Consent Mode controls Ads collection
Google changes the control for Ads cookies and identifiers, with no effect size stated.
Current rules: log changes, reset baselines and reconcile counts
A small team does not need a new dashboard for every vendor announcement. It needs a record of what changed, a visible marker where people make decisions, and a recurring comparison with a count that exists outside analytics. Assign one person to maintain the record even if several people make changes.

The chart keeps one name while its definition changes at the markers. Read each segment against its own baseline.
- →1. Keep a measurement change log outside analytics
Use columns for date in UTC, system, what changed, metrics affected, expected direction, source or link, annotation added, and new baseline start. Iron Goo's website foundation build instruments analytics on the site, where tags and consent defaults need to be set up. The check is simple: someone who did not make the change can identify the affected metric and its source from the log.
- →2. Annotate the charts people read
Add the date and a plain description of every relevant break to the charts used in reports and decisions. Mark the key-event rename as a label change with no expected count change; mark a channel definition change where acquisition trends are reviewed. Check that the marker appears on the chart a decision-maker actually opens, and keep the external log as the record.
- →3. Start a new baseline after a definition change
Do not compare across the break as though the definition stayed constant. Use at least four full weeks on the new definition to cover weekday cycles, and use year-over-year comparisons only when both periods share that definition. The check is that the baseline start date and definition are recorded; use the new baseline when resetting KPI targets.
- →4. Compare analytics with a source of record
Each month, divide the analytics count by the corresponding count from the payments system or CRM. Use paid orders for purchase events and the CRM as the count of record for leads and customers, with the definitions and date window aligned. The check is a recorded monthly ratio; data hygiene for small teams explains how to keep the CRM usable as a count of record.
- →5. Re-audit on a schedule
Review the coverage ratio monthly. Each quarter, check that tags fire, consent defaults are correct, the key events list and channel groups match the reporting plan, and the CRM sync works. Repeat the relevant checks immediately after a vendor release note changes a definition; the check is a dated audit record with any differences entered in the change log.
The following example rows show how to record two real GA4 release dates. The annotation and baseline fields are illustrative: enter the actual status and start date for your own reporting.
| Date UTC | System | What changed | Metrics affected | Expected direction | Source/link | Annotation added | New baseline from |
|---|---|---|---|---|---|---|---|
| 2024-03-21 | GA4 | Conversions renamed key events | Report labels | None on counts | Google Analytics announcement | N | Not required for rename alone |
| 2026-05-13 | GA4 | AI Assistant channel added | Referral, AI Assistant | Referral may fall; AI Assistant may rise | GA4 release notes | N | Date chosen after review |
For a hypothetical monthly check, say the payments system records 120 paid orders and GA4 records 96 purchase key events for the same period and definition. The coverage ratio is 80%. It is a diagnostic measure, not an instruction to multiply GA4 revenue by a correction factor.
coverage_ratio = 96 GA4 purchase key events / 120 paid orders = 80%

Compare the analytics count with the payments count of record. Watch the ratio between them for a change around a dated release or configuration update.
A step change in coverage near a dated release points to a possible measurement break; investigate the tag, consent settings and event definition. A business change may move both counts together, leaving the ratio more stable, though neither pattern proves a cause on its own. Start the log with the last known tracking change and add its date to the chart your team uses most.
Put your next steps in order
Build on what you have just read with a free ebook on semantic SEO and AI visibility. Get worked examples, decision rules, checklists and a do-this-week action in every chapter, plus a six-week plan for your site.

