10 November 2026 · 8 min read

Back/forward cache: the page views GA4 never counted

On 9 November 2026, Google Tag Manager changed one line of behaviour that reaches a large share of GA4 properties on content sites. Web measurement tags that subscribe to page views on history change now record a page view when a page is restored from the browser back/forward cache. Google's wording is that the tags now mirror what they already did on back and forward navigation that reloads the page.

No new event, no new parameter. The change closes a case where the view was never sent at all, which is why it arrives as a rise in one number rather than a new row in a report.

What the release actually covers

The release applies to tags that subscribe to a page view on history change. That subscription exists in two places, and most properties have at least one of them switched on.

If both are off for your property, nothing changed and the rest of this article is background. If either is on, the Views number in your reports is now measuring something slightly different from what it measured last week.

Why a cached page produced no page view

Back/forward cache, usually shortened to bfcache, is a browser optimisation. When a visitor navigates away, the browser does not destroy the page. It pauses JavaScript and keeps a snapshot of the entire page in memory, including the JavaScript heap. Go back, and the page reappears in milliseconds with no network request at all.

That is the reason the view went missing. GA4's history listener waits for one of the three history API calls, and a restore from cache fires none of them. Nothing reloads, so the config command never runs again. Nothing calls pushState, so the listener never wakes up. The page is on screen, the session continues, and no page view is recorded for the navigation the visitor just performed.

The event that does fire is pageshow, with its persisted property set to true. That is how you tell a restore from an ordinary load, and it is the hook Google's tags now use.

Google's developer documentation has described this gap for years. The bfcache guide on web.dev states that many popular analytics libraries do not measure restores as new pageviews, and it walks through a manual fix built on the pageshow event. The November release moves that fix into the tags themselves.

How much the number moves

Chrome usage data quoted by web.dev puts back and forward navigation at about one in ten navigations on desktop and one in five on mobile. That is the size of the pool, not the size of your increase.

Two filters sit between the two. Some back and forward navigation reloads the page instead of restoring it, because the page was never eligible for the cache or was evicted from it. Those visits were always counted. And a restored page only counts from 9 November onward.

So the rise you see is the share of back and forward navigation that came from memory. Sites that keep pages cache-friendly, meaning no unload listeners, no Cache-Control: no-store on the document and no long-lived connections left open, see the largest effect. Sites with third-party scripts that register unload handlers see less, because those pages reload instead of restoring.

The trap: double counting on sites that fixed this by hand

For the past few years the documented workaround was a short listener on your own pages. Register pageshow, check persisted, send a page view yourself. Plenty of sites run that code, either in the site's own bundle or through a Custom HTML tag in the container.

If yours does, the listener did not go anywhere. It still fires on every restore, and now the Google tag sends its own view at the same moment. Two page views, same URL, same session, one navigation.

Checking takes about two minutes. Open Preview in GTM, load a page, click through to a second page, then press back. Count the page_view events that arrive on the return. One means you are fine. Two means the site owns the problem described in this section.

There are two sensible fixes, and the choice depends on why the snippet was added.

What you should not do is switch off the history page view setting to stop the second event. That setting is also what gives a single-page application one view per screen change. Turning it off trades a small double count for a much larger hole.

What the extra view carries

A restored page is the same document that was frozen, so the view it sends describes the page as it was at the moment of the freeze. The URL, the title and the referrer come from that frozen state, not from a fresh request.

That has a practical consequence worth knowing before you read the new numbers. A visitor who returns to a page from cache does not look like a new entry to the site, because they are not one. The view carries the address they originally landed on, including campaign parameters in it, and the same referrer the first load recorded.

The second consequence is about the data layer. Variables that read a value at the moment of the push read the page state that was frozen. If the visitor switched currency, accepted a consent banner or picked a variant before navigating away, that state is still there when the view fires. Usually that is what you want. Occasionally it explains a mismatch against a server log.

Neither behaviour is a defect. They are the shape of the gap that the release closed: the page was never really reloaded.

Pages that never reach the cache

Expectations need this list, because a cache hit rate is never 100 percent and a page that fails one condition simply reloads.

The practical reading is that the new page views are a floor, not a total. If your site blocks caching in one of these ways, the count still understates back and forward navigation, and the fix is on your side rather than Google's.

Sizing the change on your own property

Do not guess from someone else's percentage. Measure the step on your own traffic.

Compare Views by device for the seven days before 10 November against the seven days after. Mobile should move more than desktop, because mobile navigation leans on back and forward. If neither moves, check that the setting in question is actually enabled for the stream, and check that the pages are eligible.

Then look at Views per session. This is the ratio dashboards usually hold steady, and it is the first thing a definition change disturbs. A jump of a few points across the whole site, concentrated on content pages and blog posts, is what a correct implementation looks like.

One more comparison is worth making, because it separates the two possible causes. If a landing page that takes search traffic rises while a page reached only from internal links does not, the difference is the cache, not a new campaign. If a page that used to be your best performer jumps by a third, check for the manual listener first.

Mark the date in your annotations so a reader six months from now does not read the step as a campaign result. If you already keep annotations for releases and site changes, this one belongs next to them.

The one-line explanation for stakeholders

Traffic did not change. The count of pages seen did. A visitor who pressed back onto a page the browser had kept in memory was always visiting that page again, and until this month the tag did not record it.

That sentence matters more than it looks. It stops someone comparing November against last November and concluding that a redesign worked, or that a channel suddenly improved. The generator of the number changed, and the honest way to present that is a re-baseline rather than a victory.

A short checklist for this week

  1. Confirm whether page views on history change are enabled, in GA4 enhanced measurement and in the Google tag settings.
  2. Search the site's JavaScript and the container's Custom HTML tags for a pageshow listener that sends page_view. Decide which one owns the event.
  3. Run the Preview test: load, click, back. One view is correct, two needs a fix.
  4. Re-baseline Views per session for any dashboard that reports engagement.
  5. Add an annotation for 9 November.
  6. If the increase is smaller than expected, audit eligibility: unload listeners, no-store on documents that do not need it, and connections left open.

None of this requires a reimplementation. It requires knowing which of the two page view paths your site uses, and cleaning up the older workaround if it is still running. The measurement gap that took a custom script to close a year ago is now closed by the tag. Leaving both in place is the only way to get this wrong.

Want to know whether your page views doubled up in November? Book a free tracking audit and we will check the history settings, the custom listeners and the numbers your dashboards are built on.