01 / Context
Where Core Web Vitals stand in late 2026
Despite a steady stream of blog posts announcing "new 2026 thresholds", the official numbers have not moved. Google's Search Central documentation still asks for a Largest Contentful Paint (LCP) within 2.5 seconds, an Interaction to Next Paint (INP) below 200 milliseconds and a Cumulative Layout Shift (CLS) below 0.1. Each is judged at the 75th percentile of real Chrome users, split by mobile and desktop. The metric set itself has been stable since INP replaced First Input Delay in March 2024.
Google's wording on rankings is also unchanged and worth quoting precisely: good Core Web Vitals "align with what our core ranking systems seek to reward". That is not a promise that a green report lifts you above a more relevant competitor. In practice we treat Core Web Vitals as a tie-breaker in search and as a direct lever on conversion, which for an online store is the bigger prize.
What a store should take from this
- Stop chasing rumoured thresholds; optimise against the documented ones with a safety margin.
- Measure per template (home, category, product, cart, checkout), not per domain average.
- Prioritise mobile: that is where most revenue now happens and where pass rates are lowest.
The rest of this report covers what actually changed in 2025 and 2026, why it matters commercially, and the engineering order of work we would follow on a typical catalogue site with nine days left before Black Friday.
02 / What changed
The real 2025–2026 changes: browsers, soft navigations, INP drift
Three developments matter more than any threshold rumour.
1. LCP and INP now work beyond Chrome
With Safari 26.2 on 12 December 2025, LCP and INP became Baseline Newly available, meaning they are measurable in Chrome, Firefox and Safari. CLS remains Chromium-only. Google's field dataset (CrUX) is still Chrome-only, but your own real-user monitoring can now see iPhone shoppers' loading and responsiveness for the first time. For stores with a heavy iOS audience, that closes a long-standing blind spot.
2. Soft navigations are entering CrUX
On 13 October 2026 Chrome published experimental soft-navigation versions of LCP, INP and CLS in CrUX, built on the Soft Navigations API that shipped in Chrome 151. Single-page and headless storefronts have historically been measured only on the first hard load. Google says it aims to graduate these metrics in May 2027, subject to feedback, and that Search Console will switch to the new data once they graduate. Stable scores are unchanged for now.
3. INP is quietly getting worse
The CrUX release notes show the share of origins with good INP falling from 86.6% in the May 2026 dataset to 85.2% in October 2026, and the August notes describe the continued INP regression as a concern without a definitive cause. Overall, 56.3% of origins passed all three metrics in the October 2026 dataset. Put plainly, almost half the web fails, and responsiveness is the metric moving in the wrong direction.
03 / Revenue
Why speed is a revenue problem, not an SEO checkbox
The best-known evidence is still the Google-commissioned Deloitte study "Milliseconds make millions". Across 37 European and American brand sites and more than 30 million sessions, a 0.1-second improvement in mobile speed metrics was associated with an 8.4% increase in retail conversions and a 9.2% increase in average order value. Two caveats keep us honest: the data dates from late 2019, and the gains required every page in the journey to improve, not just the home page. It is a correlation study, not a guarantee.
The commercial context has only sharpened since. Adobe Analytics reported that US shoppers spent a record $11.8 billion online on Black Friday 2025, up 9.1% year on year, with mobile accounting for 55.2% of those sales. Cyber Monday reached $14.25 billion. When more than half of peak-day revenue arrives on phones, the mobile 75th percentile is not an abstraction; it is a mid-range Android handset on a congested network trying to add a jacket to the basket.
Think in templates and funnel steps
A domain-level "pass" can hide a failing product listing page that carries most of your paid and organic landings. Map each template to its traffic and revenue share, then fix in that order. This is the core of how we approach high-performance e-commerce: performance budgets tied to the pages that actually take money, rather than to a vanity score.
Core Web Vitals for stores: thresholds, culprits, first fixes
The official thresholds, judged at the 75th percentile of real users, with the most common e-commerce cause we see for each and the first fix to try.
| Metric | Good (p75) | Typical store culprit | First fix |
|---|
| LCP | ≤ 2.5 s | Hero or product image discovered late, slow uncached HTML | Preload the LCP image with fetchpriority="high", cache HTML at the edge |
| INP | < 200 ms | Facet filters, variant pickers and tag scripts blocking the main thread | Break up long tasks, defer non-critical third parties |
| CLS | < 0.1 | Promo bars, review widgets and late web fonts pushing content down | Reserve space for injected UI, use size-adjusted fallback fonts |
| TTFB (diagnostic) | ≈ 0.8 s or less | Personalised pages that bypass the CDN on every request | Cache by segment, move personalisation to the client or edge |
LCP, INP and CLS thresholds per Google Search Central; TTFB is a diagnostic metric, not a Core Web Vital.
04 / Loading
Fixing LCP: images, HTML caching and the edge
On most catalogue sites the LCP element is an image: the hero banner on the home page, the first product shot on a product page, sometimes the first tile on a category grid. The fixes are well understood and still routinely skipped.
Images
- Make the LCP image discoverable in the initial HTML, not injected by a carousel script. Add
fetchpriority="high" and never lazy-load it.
- Serve AVIF or WebP with responsive
srcset and sizes, so a 390px-wide phone does not download a 2,400px master.
- Lazy-load everything below the fold, including the rest of the gallery and recommendation carousels.
HTML and the CDN
Image work is wasted if the document itself takes a second and a half to arrive. Category and product HTML should be cacheable at the edge for anonymous users, with short TTLs plus stale-while-revalidate so price and stock updates propagate without cache stampedes. Watch your cache key: tracking parameters, currency cookies and A/B test cookies frequently fragment the cache into near-zero hit rates. Normalise or strip what does not change the response.
For headless builds, check that server-side rendering actually produces the product image and title in the first response, and that hydration does not re-render the LCP element. Our web development team treats this as a release gate rather than a post-launch clean-up.
05 / Responsiveness
Fixing INP and CLS: JavaScript budgets and third-party discipline
INP measures the slowest meaningful interaction a shopper has on a page. On stores, the worst offenders are predictable: applying a filter on a category page, choosing a size or colour, opening the mini-cart, and typing into search. Each tends to trigger a large re-render plus a burst of analytics events on the main thread.
Engineering patterns that work
- Give visual feedback first, then do the heavy work. Update the button state, yield to the browser, and only then recalculate the grid.
- Split long tasks. Use
scheduler.yield() where supported, with a setTimeout fallback, so input is not stuck behind 300 ms of filtering logic.
- Debounce search-as-you-type and move expensive work, such as price formatting across a large grid, off the critical interaction.
Third parties are usually the real budget
Tag managers, consent platforms, chat widgets, review apps, session-replay tools and personalisation scripts often add more main-thread time than the storefront itself. Inventory every script, its owner and its business value. Load non-essential ones after interaction or on idle, and remove anything nobody can justify. Treat this as a governance issue as much as an engineering one; it also reduces the supply-chain risk we flag in a security audit.
For CLS, reserve explicit space for promo bars, cookie banners and review stars, and set width and height on every image. Use font fallbacks with size-adjust so late web fonts do not shove the price and the add-to-basket button down the page.
06 / Peak season
Black Friday readiness: nine days out
Black Friday 2026 falls on 27 November. If you are reading this in the week before, the priority is not a re-platform; it is protecting what works and removing avoidable regressions.
Freeze what can break
Declare a code and tag freeze for the peak window, and include marketing tags in it. The most common peak-week regression we see is not a deploy; it is a new pixel, a countdown timer or a full-width banner added on the morning of the sale. Pre-build campaign banners at fixed dimensions now so they cannot shift layout.
Load, cache, rehearse
- Run a load test against cached and uncached paths, including search, cart and checkout, which rarely benefit from the CDN.
- Warm edge caches for top category and product URLs before traffic arrives, and confirm promotional URL parameters do not bust the cache.
- Check that origin autoscaling and database connection limits match the expected peak, not last year's.
- Rehearse a rollback and a third-party kill switch: if a vendor script slows down, you need to be able to disable it in minutes without a deploy.
Finally, set up a real-user monitoring view that the on-call engineer and the e-commerce manager both watch. CrUX and Search Console will not show you what happened until weeks later.
Pre-peak Core Web Vitals checklist
Ten checks we would complete on any store before a high-traffic campaign.
- [+] Identify the LCP element on home, category and product templates on a mid-range Android phone.
- [+] Ensure the LCP image is in the initial HTML, has fetchpriority="high" and is not lazy-loaded.
- [+] Serve AVIF or WebP with correct srcset and sizes on all product imagery.
- [+] Confirm category and product HTML is edge-cached for anonymous users, with a clean cache key.
- [+] Profile filter, variant and mini-cart interactions for long tasks over 50 ms.
- [+] List every third-party script with an owner; defer or remove those without a clear business case.
- [+] Reserve space for promo bars, cookie banners and review widgets to prevent layout shift.
- [+] Load-test search, cart and checkout, which usually bypass the CDN.
- [+] Prepare a kill switch for each non-essential vendor script.
- [+] Publish a shared RUM dashboard segmented by template and device for the peak window.
07 / Measurement
Measure what Google sees and what your shoppers feel
There are two sources of truth and they answer different questions. CrUX, which powers the Search Console Core Web Vitals report and PageSpeed Insights field data, is a 28-day rolling view of Chrome users. It is what Google's systems see, but it lags, it aggregates, and it cannot tell you which campaign or release caused a regression.
Real-user monitoring with the open-source web-vitals library, or a commercial RUM tool, fills the gap. Since December 2025 it can capture LCP and INP from Safari and Firefox as well. Send the metric together with attribution data (the LCP element, the interaction target, the slowest script) and segment by template, device class and country. That is how you find that one review widget on product pages is responsible for most of your poor INP.
Plan for soft navigations
If your storefront is a single-page app or a headless build with client-side routing, start comparing the experimental soft-navigation metrics in CrUX with your stable numbers now. Google expects most SPAs with many in-app navigations to look better, but heavy work during route changes will surface as poor INP.
Lab tools such as Lighthouse remain useful for debugging, not for reporting. For a full crawl and indexation review alongside performance, see our technical SEO audit checklist for 2026.