Low bfcache restore rate: find what blocks the back button

The back button often reloads the page instead of showing it from memory. Field data shows why: the page kept itself out of the bfcache, or the browser dropped it.

Arjen KarelArjen KarelUpdated November 22, 2026

Why is my bfcache restore rate so low?

In short

  • Your restore rate is the share of back and forward visits that come back from memory instead of loading again. Few do: across 92 sites on CoreDash (25 October to 21 November 2026) the middle site restored 28.8%, and even a clean site stays well below 100%.
  • When Chromium gives a reason, nine times in ten it is masked (90.9% across 97 sites, 14 to 21 November 2026): a cross-origin iframe such as an ad or chat widget, or the browser dropping the page. A no-store header on public HTML and open requests follow far behind.
  • Find the blocking frame with Chrome DevTools, take no-store off public pages but keep it on private ones, move cleanup code to pagehide, and check your share again a week later.

The back/forward cache (bfcache) keeps a complete copy of the page in memory when someone clicks away, so Back can show it at once. Most back button visits load the page again instead: an embedded frame, a response header or an open connection kept the page out, or the browser dropped its copy. Across 137 sites on CoreDash only about 1 in 5 came back from the bfcache.

Across 151 sites on CoreDash, 9.0% of all page views started with the back or forward button. The back button is my favourite fix on a site, because every visitor uses it.

What is a normal restore rate?

Across 92 sites on CoreDash with at least 100 back and forward navigations (25 October to 21 November 2026), the middle site restored 28.8% from the back/forward cache, the bottom quarter 5.1% or less and the top quarter 54.6% or more. Pooled across all 137 sites, 19.7% of all 1,224,141 back and forward navigations came back from memory.

Even a page with nothing blocking it stays below 100%. Browsers drop a cached page after a timeout (10 minutes in Chrome, 3 minutes for a no-store page) or when a phone runs low on memory, and start fresh after a restart or in a duplicated tab (Chrome docs, web.dev). On Adevinta's classifieds marketplaces, phone back and forward visits restored from the bfcache went from 16% to 68% in CrUX on one marketplace once its blockers were gone.

Restore rates also differ per browser. Safari restored 45.3% of back and forward navigations (117 sites), Firefox 25.7% (105 sites), Edge 25.1% (105 sites) and Chrome 10.3% (129 sites); desktop restored 24.0% against 18.5% on mobile. Each browser has its own rules. Safari, which in CoreDash includes Chrome on an iPhone (it runs on WebKit), keeps pages with unload listeners or an open WebSocket. Firefox skips pages with an open WebSocket, and on desktop pages with an unload listener. Chrome ignores unload listeners since Chrome 155 and keeps pages with open WebSockets since Chrome 149. Comparing browsers puts different visitors on different sites side by side, and that is a correlation: the groups also differ in device memory, habits and iPhone traffic.

Restore rate per browser across 137 sites on CoreDash, 25 October to 21 November 2026. The groups differ in more than the browser.

CrUX reports the share of each navigation type per origin over 28 days, in Chrome; RUM splits LCP, FCP and TTFB by navigation type in every browser (how CrUX and RUM differ).

How do I measure my restore rate with real visitors?

Your restore rate is "Back/forward from bfcache" divided by the sum of "Back/forward" and "Back/forward from bfcache". Every browser fires pageshow with event.persisted set to true when a page comes back from memory, so RUM counts restores in Safari, Firefox and Chrome alike.

In CoreDash, open the FCP or TTFB page and look at the "Arrival" card. It shows "Back/forward" (the page loaded again) next to "Back/forward from bfcache" (the page came back from memory), and "Navigation type per page" opens the "Navigation type" tab in the Pages table. The "Restored" row is something else: tabs Chromium loaded again after discarding them or after a restart. You can also pick "Navigation type" under More on the LCP page, or split and filter by "Page label", "Folder", "Device" or "Browser" in Explore.

Navigation type in the Arrival card on the FCP page, with both back/forward rows.

CoreDash also checks each page group (the page label when set, otherwise the URL folder) and opens a "Low bfcache use" issue titled "<share> of back/forward navigations use the bfcache" when fewer than 25% of at least 100 back and forward navigations over the last 7 days came from the bfcache (Poor below 5%). The issue card names the most frequent reason on that page group and links to "Open filtered view" in Explore.

The Low bfcache use issue for one page group, with the reason the browser gives most often.

How much faster is a back button visit from the bfcache?

A restore from the bfcache makes no network request, so the page is back on the next frame. The web-vitals library reports a TTFB of 0 for it, and FCP and LCP both equal the time from pageshow to the next paint (web.dev, web-vitals). A back button visit that loads again is a full page load with a warm HTTP cache: across CoreDash sites (25 October to 21 November 2026), "Back/forward" had an LCP p75 of 1.42 s (135 sites) against 1.71 s for "Page load" (130 sites).

Which reason keeps my pages out of the bfcache?

Across 97 sites on CoreDash from 14 to 21 November 2026, 136,425 Chromium back and forward loads carried a reason, and 90.9% of those 136,425 loads reported masked ("Hidden by the browser, often a cross-origin iframe").

ReasonShare of back/forward loads with a reason
Hidden by the browser, often a cross-origin iframe (masked)90.9%
Cache-Control: no-store4.3%
An open network request1.7%
A browser feature1.2%
Unload event handler0.2%
Everything else1.7%

masked means the browser hides the specific cause (Chrome, MDN). It covers two cases: a cross-origin <iframe> such as an ad slot or a chat widget blocked caching, or Chrome dropped the page itself (after its 10-minute timeout, when its cache slots filled up, under low memory, or because an extension injected scripts). That is why all 97 sites report it: everyday evictions land in the same bucket as blocking frames.

To tell a blocking frame from an ordinary eviction, run the back/forward cache test in Chrome DevTools on one URL of that template: it names the frame, and Lighthouse's bfcache notice runs the same check.

Only Chromium browsers send a "Not restored reason (bfcache)"; Safari and Firefox show up in the restore counts. Of the 210,157 Chromium "Back/forward" loads that week, 35.1% sent none ("Not set"), as in duplicated tabs and tabs restored after a browser restart (MDN).

To read the reasons for your own Chromium visitors in CoreDash, open Explore, pick FCP, filter "Navigation type" to "Back/forward" and set "Split by" to "Not restored reason (bfcache)". Five reasons, an open network request and an HTTP error page among them, have no label in CoreDash and show under the browser's own name.

Explore split by Not restored reason (bfcache), on back button visits that loaded the page again.

What do I change to raise my restore rate?

Each cause below starts with what you can change without touching code, then the one line for your developer. Start with the reason behind most "Back/forward" loads on the templates where people press Back all day, such as search results.

Cross-origin iframes: ads, chat widgets and embeds

Once DevTools names the blocking frame, check that vendor's settings in Google Tag Manager, the ad setup or the chat plugin. Update to the vendor's current script, or load chat and video embeds only after a click. When you can't change the frame, take the widget off the templates where people press Back most, keep it on the pages where they use it, and send the vendor the reason DevTools shows.

Google Publisher Tag refreshes the ad slots in view after a restore by default. If your setup needs its own refresh, call it on pageshow when persisted is true:

window.addEventListener('pageshow', (e) => { if (e.persisted) googletag.pubads().refresh(); });

See iframes and ads with bfcache on web.dev and the Google Publisher Tag refresh reference.

Cache-Control: no-store on public HTML

Check whether a plugin (such as a WordPress plugin calling nocache_headers() sitewide) or a CDN rule sends Cache-Control: no-store or sets changing session cookies on public pages. Safari and Firefox skip every no-store page. Chrome keeps one for up to three minutes when no cookie changes and no no-store request or WebSocket is open. Turn the sitewide cache blocking or the anonymous session cookie off for logged-out pages, or ask your host or developer to limit no-store to logged-in and checkout URLs.

Keep no-store on logged-in, account and checkout pages that show personal data, and check the session again on pageshow when event.persisted is true, so signing out never leaves a private page in memory. The workaround people try first is stripping no-store from every template, account and checkout included. On a shared device, Back after signing out then shows the previous person's page from memory.

On public pages that still need revalidation on every fresh load, send no-cache instead:

Cache-Control: no-cache

See Chrome's Cache-Control: no-store bfcache rules and minimising no-store on web.dev.

Unload event handlers

Chrome ignores unload listeners by default since Chrome 155 and Safari ignores them when it caches a page, but desktop Firefox still skips a page that has one. Turn off Rocket Loader in your Cloudflare dashboard, which adds an unload listener, and pause or update old analytics and A/B testing tags in Google Tag Manager that listen to unload. A tag that still needs it and has no update goes on the one conversion template that requires it, instead of every page.

Move cleanup code and exit beacons from unload to pagehide:

window.addEventListener('pagehide', () => navigator.sendBeacon('/log', data));

See configuring Cloudflare for Core Web Vitals and Chrome's unload deprecation guide.

Open network requests and connections

Live price tickers, notification pollers and exit-tracking tags can leave a request, a BroadcastChannel, a Web Lock or an IndexedDB transaction open when someone clicks a link. Turn background polling off on listing and article pages, or keep live widgets on the one template that uses them.

Close channels and database handles on pagehide, reopen them on pageshow when event.persisted is true, and abort requests in flight with an AbortController:

window.addEventListener('pagehide', () => controller.abort());

See closing open connections before navigation on web.dev.

How do I check that my fix worked?

A week after the change, open the same "Arrival" card or Explore view and check that "Back/forward from bfcache" has a larger share of back and forward visits and the blocking reason has dropped. In CoreDash, mark the "Low bfcache use" issue "Fix applied": the run seven days later reads only page views from after the fix and closes it as "Fixed" or opens it again.

The values of both filters are in the CoreDash reference under Navigation type and Not restored reason (bfcache). Related: CrUX vs RUM and prerendering with Speculation Rules on corewebvitals.io.

Track your first site free.No card needed. Setup takes about 5 minutes.

Start free trialBook a demo