The state of Australian small business websites 2026
We ran 12,621 Australian small business websites through Google Lighthouse (mobile), the same test behind Google's own PageSpeed Insights, between 25 May and 5 September 2026. This is what we found.
Only 11.7% of sites pass the 2.5-second LCP bar under Lighthouse's lab test, against 59.9% among the subset with real Chrome user data (n=2,887, about 23% of the sample). Lighthouse's lab throttling is deliberately harsher than most real connections, and CrUX only reports on sites with enough traffic to measure — so the field number is both more forgiving and biased toward the more established sites in the sample. Neither number is the whole picture on its own.
Most sites land in the middle, not the bottom
The largest single band is 60–70 (2,635 sites), and the distribution skews toward "needs improvement" rather than outright failure — but a full 27.3% still score under 50, Lighthouse's own poor threshold.
Squarespace and Shopify score worst, Wix scores best
Among platforms with at least 100 sites in the sample, WordPress is the largest single group by far (5,554 sites) and sits below the overall mean. "Custom" means no CMS fingerprint was detected — it is a default label, not proof of a hand-built site — and "Unknown" means detection failed outright; both are reported separately and excluded from platform comparisons in our own sales conversations. Framer (n=15) was too thin to chart at this threshold.
663 distinct industry labels appear in the sample. The ranking below is limited to industries with at least 80 sites, so single unusual results don't skew a small group.
Best 10 by mean score
Worst 10 by mean score
Reports by industry
The six largest industries with at least 100 sites each get their own page, with the same numbers set against all sites and against their sector group.
Retail and real estate trail every other sector
Sites collapse into 12 keyword-matched sector groups. Retail/e-commerce and real estate score lowest and have the highest poor rates; trades and professional services score highest. "Other" is industries that didn't match a keyword rule, not a catch-all judgement.
Tasmania scores best of the states we can report
State is derived from the postcode or state text in each business's listed address; 9,572 sites had no derivable state and are excluded here. Only states with at least 100 matched sites are shown — WA, SA and NT were all too thin to publish individually.
State × industry
A state × industry cell is only published when it holds at least 30 sites. Industry here is the business's own listed category, not the sector groups above. None of the 1,044 combinations across 3,301 state-matched sites gets there — the largest is Plumber in VIC, at n=26 — so no state × industry figure is reported.
| State | Industry | n | Median score | % poor (<50) | Median LCP |
|---|---|---|---|---|---|
| No state × industry cell reaches n≥30. | |||||
What this means for a business owner
A Lighthouse score in the 50s or 60s does not mean a site is broken. It means the site takes noticeably longer than it should to become usable on an average mobile connection, and that shows up as visitors leaving before the page finishes loading rather than as an obvious fault anyone would notice by eye.
The gap between lab and real-user results in this data is a reminder that a site can look fine to the person who owns it — opened on a fast office connection, on a laptop, on a site they already trust — while performing very differently for a stranger on a phone deciding whether to stay. Median lab load is 6.9 seconds; Google's own threshold for a good result is 2.5.
None of this is platform-proof or industry-proof. Every platform and every industry group in this sample has sites that pass comfortably and sites that fail badly — the averages describe a pattern, not a guarantee. The only way to know where a specific site stands is to test it. Run your own site through the same test.
Domains are Australian small and medium businesses gathered by Pryce Digital's own lead research (Google Places and Australian Business Register sourced), each run once through Google Lighthouse via the PageSpeed Insights API. This is not a random sample of all Australian websites — it is skewed toward the small local service businesses we prospect (trades, health, hospitality, professional services and similar), and skewed away from large enterprise or purely e-commerce-at-scale sites.
All runs use mobile strategy: mobile emulation with Lighthouse's default network and CPU throttling. Each domain has a single Lighthouse run (the latest successful run when a domain was checked more than once). Individual Lighthouse scores are known to vary run to run, typically by five to ten performance points, from lab network and CPU variance; aggregate statistics across thousands of domains are stable, but no single domain's exact score should be read as precise.
Platform is detected heuristically from HTML, header and asset fingerprints, with a confidence score. We only treat a match at 0.7 confidence or higher as a confirmed CMS. "Custom" (0.4 confidence) means no fingerprint matched — it is not proof of a hand-built site — and "Unknown" (0.0) means detection failed outright.
Page weight statistics (median 5.0 MB) come from a Lighthouse diagnostic that only fires when a page's payload is large enough to warrant a warning, covering just 63 of the 12,621 domains. That is a biased sample of the heaviest pages, not a census of page weight across the study. The LCP-over-4-seconds and LCP-over-10-seconds figures, by contrast, are calculated across the full sample.
Lighthouse's lab Interaction to Next Paint metric was null for every run in this dataset and could not be reported; the field INP figures above come from CrUX only. State is derived from a postcode found in each business's address text, using official Australia Post ranges, falling back to a state-abbreviation word match when no postcode is present. Only states with at least 100 matched domains are published individually.
A field-data flag is set to true for 100% of domains, but the actual Chrome UX Report metrics behind it are only populated for 22.9% (2,887 domains) — the flag defaults true rather than confirming a real match, so only the presence of an actual field metric value should be read as "we have real-user data."
This study is released under CC BY 4.0. Use the numbers and charts freely in your own reporting, with attribution to Pryce Digital, 2026.
Press enquiries: studio@prycedigital.com