Methods and data: SMB Web Practices Study

What the websites of small businesses competing on Google Maps do, which practices go with better results, and how doing them well differs by platform and trade.

Status: final.

Design and sample

The websites of 1,147 owner-operated local service businesses in 9 categories and 18 US metros (54 category-metro pairs) were rendered on 2026-10-24 and compared with each business's best rank in Google's local pack, from a Maps pull on 2026-10-24 at 8 search points per metro, to rank 50. Ranks are grouped into tiers: top 3, ranks 4-10, 11-20, 21-30, and not seen (a business in the area that never reached the top 30; ranks 31-50 count as not seen, as the locked tests define the tiers). The business-name findings use every enriched owner-operated listing (4,194), with or without a website.

Two samples: discovery and confirmation

The associations here were first seen in a discovery sample: round 1's 18 pairs, 279 websites fetched 2026-10-23 against a Maps snapshot of 2026-10-18. Before round 2's data was pulled, six tests were locked in confirmatory_tests_r2.md (SHA-256 recorded in analysis/KIT_RULES.md and committed first). They were then run once on the 36 new pairs, which no association was built from. 2 of the six held; the rest are reported below and their findings were rewritten to say what the new data shows. Round 2 also re-pulled the discovery pairs, with a fresh draw of businesses; every table below uses all round-2 sites unless a finding says it uses the 36 new pairs. Organic results and backlinks were bought for the discovery sample only, so the off-site findings describe it.

The GBP local-pack study (releases/v1.x) uses the same kind of businesses and focuses on the profile. This study's findings are about the website; where one extends a GBP finding, its caveat or methods line says so.

The six locked tests

Test	What it tests	Pass rule	Estimate (95% CI)	n	Result
RT-01	City relevance (0-5) goes with top-10 rank	OR per point > 1, CI excludes 1	1.09 (0.97 to 1.22)	746	FAIL
RT-02	Service word in the business name goes with top-10 rank	OR > 1, CI excludes 1	1.85 (1.40 to 2.44)	2589	PASS
RT-03	Service words on the homepage (0-4) do not go with rank	OR per point CI inside [0.67, 1.5]	1.33 (1.16 to 1.51)	746	FAIL
RT-04	Fewer than a third of tracked sites count conversions	Wilson upper 95% bound < 1/3	0.22 (0.19 to 0.27)	421	PASS
RT-05	Sites with no measurement rank lower	OR < 1, CI excludes 1	1.21 (0.80 to 1.83)	746	FAIL
RT-06	A canonical tag goes with top-10 rank	OR > 1, CI excludes 1	0.90 (0.62 to 1.30)	746	FAIL

How each site was read

• Render. Headless Chromium through datanephew's /try renderer: the homepage loads, the page is scrolled, then a known cookie banner's accept button is clicked. Each analytics or ad tag records the phase it first loaded and first sent data in. No hit is forwarded to the owner's analytics: each is answered locally. The page the Google listing links to is also rendered when it is not the homepage.
• Measurement. Every tag /try recognises (Google, ad pixels, session replay, call tracking, CRM, consent tools), the site builder's or host's own visitor statistics (Wix, Squarespace, GoDaddy, Duda and other Snowplow builders, Cloudflare), Google ids written in the page, and what the published Google tag container holds and listens for.
• On-page SEO. Title, meta description and keywords, canonical, H1, Open Graph, verification tags, image alt text and schema types, from the rendered homepage.
• Local relevance. Whether the metro's city name appears in the title, H1, meta description, schema address and visible text (a 0-5 score), and the category's service words in the same places (0-4).
• Off-site (discovery sample). Organic results for " " searched from the metro (SerpAPI, top 30), and the DataForSEO backlink summary of each own domain.
• Speed. One mobile Lighthouse run per homepage through the PageSpeed Insights API: the performance score and lab metrics, the accessibility, best-practices and SEO category scores, network weight, and Chrome UX Report field data where Google has enough real-visitor traffic to publish it.
• Domain, hosting and email. The domain's registration date (RDAP), the response headers of the homepage (server, HSTS, HTTP/3; a rate-limited or failed response counts as missing, not as "no"), and the email provider and records in DNS.

Strategies (Part 3): how the patterns were found

Part 3 comes from exploring how the website variables relate to each other, not only to rank: a Spearman correlation matrix over 68 variables (population-wide and within each market), clustering of the variables into themes, k-means clustering of the sites on 57 site-practice variables (no rank or review variables), with the number of groups chosen by bootstrap stability, and the correlation of each variable pair within each platform, trade, review tercile and level, about 9,000 tests with Benjamini-Hochberg q. A pattern is kept only if it holds, with the same sign, in both samples. The levels are points along a continuum (silhouette under 0.1), not distinct types. These are patterns with limited evidence: each is to be tested on new markets in the next study.

Analysis

Each signal is compared between businesses ranked in the top 10 and the rest with a logistic regression adjusted for review count (percentile within the pair), claimed status and distance to the nearest search point, with standard errors clustered by pair. Headline associations are re-estimated under pair fixed effects, without the distance control, on top 3 vs not seen only, on own domains only, without plumbers, and leaving out one pair at a time (table W1_sensitivity_matrix); a "flip" there lowers the grade, as in the GBP study. Exploratory directional tests carry a Benjamini-Hochberg q within their family. "No association" findings need their interval inside an equivalence band of 0.67-1.5 to grade High.

Grading

The GBP study's rules, unchanged (High / Medium / Low; consumer gating by grade). A finding a locked test confirmed as declared is graded on that test and is not capped. Every other association was found while exploring, so it carries a "post hoc" caveat that caps it at Medium, including findings rewritten after a locked test failed. This release: 45 Medium, 4 Low, 1 High.

Limitations

• One cookieless visit per site from a datacenter; tags gated on other consent tools or on visitor behaviour can be missed.
• Homepage and listing landing page only; city matching uses the metro's city name, so a suburb's own town does not count.
• Sites and ranks were read on the same day (2026-10-24); the off-site data is from the discovery sample (2026-10-23).
• The study set draws equally from every rank tier, so tier shares describe the sample, not how common a signal is.
• Associations only. Nothing here shows that changing a website moves a ranking.
• Part 3 (strategies) is limited evidence: patterns found while exploring this data (a correlation matrix, clustering of the sites, and ~9,000 within-segment correlations), kept only where they held in both samples. Each is to be tested on new markets in the next study.

Findings

About the study

The sample

• WEB-C1-01 (Medium, descriptive): The study rendered 1147 websites of owner-operated local service businesses in 9 categories and 18 US metros, each matched to its map-pack rank. Homepage plus the listing's landing page; one visit per site.

Part 1. What goes with better results

Small businesses get better or worse map-pack results with a series of things; some things owners worry about make no clear difference.

What goes with a higher map-pack rank

• WEB-S1-01 (Medium, post hoc): Homepages with a meta description (75% of sites) go with higher map-pack ranks (odds ratio 1.34), but the interval barely clears no difference. Barely clears no difference; exploratory.
• WEB-S1-03 (Medium, post hoc): Homepage titles that name the service (71% of sites) go with higher map-pack ranks (odds ratio 1.83), after reviews, claimed status and distance. Exploratory; four in five of these titles repeat the business name.
• WEB-L1-01 (Medium, post hoc): Naming the city in more places on the homepage shows at most a small association with map-pack rank (odds ratio 1.09 per place). Pre-declared test RT-01 expected a clear association; not met.
• WEB-L1-02 (Medium, post hoc): Homepages naming the service in more places (title, H1, description, text) go with higher map-pack ranks (odds ratio 1.33 per place), also after reviews and distance. Not the pre-declared direction: RT-03 expected none.
• WEB-N1-01 (High, confirmed): Businesses whose name contains their service reach the map-pack top 10 more often (odds ratio 1.85), also after reviews, claimed status and distance. Report only: keyword-stuffed names break Google's guidelines.
• WEB-L3-01 (Medium, post hoc): Homepages naming the service in 3+ places sit higher on the rank ladder: odds 1.93 per rung across six bands, from 77% at ranks 1-3 to 55% not seen. Exploratory; the study set draws equally from each band.
• WEB-L3-02 (Medium, post hoc): Homepages naming the city in 3+ places sit higher on the rank ladder: odds 1.44 per rung across six bands, from 52% at ranks 1-3 to 26% not seen. Exploratory; the per-point city score (RT-01) did not confirm.
• WEB-M5-01 (Medium, post hoc): Sites with owner-installed tags sit higher on the rank ladder: odds 1.33 per rung across six bands, from 69% at ranks 1-3 to 47% not seen. Exploratory; a marker of a maintained site, not a lever.
• WEB-T4-01 (Medium, descriptive): Sites in the top fifth of their market for reviews reach the map-pack top 10 66% of the time; the bottom fifth, 16%. Sample drawn equally from rank tiers; reviews lead the GBP study too.

What makes no clear difference

• WEB-M4-01 (Low, post hoc): Sites with no measurement show no clear association with map-pack rank (odds ratio 1.21); the interval is wide. Pre-declared test RT-05 expected lower ranks; not met.
• WEB-S1-02 (Low, post hoc): 13% of sites still carry a meta keywords tag, which Google ignores; it shows no clear association with map-pack rank. Google stopped using meta keywords in 2009.
• WEB-S1-04 (Medium, post hoc): A canonical tag (80% of sites) shows no clear association with map-pack rank (odds ratio 0.90) in the confirmation sample. Pre-declared test RT-06 expected higher ranks; not met.
• WEB-N1-02 (Medium, post hoc): A city in the business name shows no clear association with top-10 rank (odds ratio 1.10); only 211 businesses have one. Few such names.
• WEB-O2-02 (Medium, post hoc): A site's spam-adjusted backlink strength (domain rank) shows no association with its map-pack rank in the first sample. Own domains only; first sample; links measured after the snapshot.
• WEB-L3-03 (Medium, post hoc): Naming the service in 3+ places does not separate sites at ranks 21-50 from those never seen (odds ratio 0.95); the gradient is in the top of the ladder. Exploratory; no website signal separated these two groups.
• WEB-T1-01 (Medium, post hoc): A faster homepage shows no association with a top-10 map-pack rank once reviews are counted (odds ratio 1.00 per 10 Lighthouse points). One mobile lab run per homepage; agrees with the GBP study's CT-11.
• WEB-T1-02 (Medium, post hoc): Older domains reach the map-pack top 10 no more often than new ones (odds ratio 0.95 per 10 years of domain age). Domain registration date, not the business's founding date.

Ratings

• WEB-T2-01 (Medium, post hoc): Older domains carry lower Google ratings (-0.15 stars per 10 years, after reviews): 66% of young domains are rated 5.0, 30% of 20-year-old ones. Young listings with few reviews are often rated 5.0.

Part 2. The best practices

What a sound website setup is, how many local businesses have it, and what doing it well involves.

Measure your visitors

• WEB-M1-01 (Medium, descriptive): 76% of local business websites have some visitor measurement: tags the owner installed or the site builder's own statistics. Builder statistics count; we cannot see whether anyone reads them.
• WEB-M1-02 (Medium, descriptive): 24% of local business websites have no visitor measurement at all: no analytics tag, pixel or builder statistics. Server-side logs and hosting dashboards are invisible from outside.
• WEB-M1-03 (Medium, descriptive): A check for five common tags finds none on 48% of sites, but only 24% have no measurement; builder statistics cover 20%. Five tags: GA4, Tag Manager, Google Ads, Meta pixel, Microsoft UET.
• WEB-T4-02 (Medium, descriptive): Wix, Squarespace and Duda sites always carry at least the builder's own statistics; 26% of WordPress and 42% of GoDaddy sites measure nothing. Builder statistics are the platform's, not the owner's analytics.

Make sure the tracking actually works

• WEB-M2-01 (Medium, descriptive): On 86% of sites with a Google tag, a GA4 visit was recorded when visited like a person; 10% recorded only after a scroll or cookie accept. One cookieless visit from a datacenter.
• WEB-M2-02 (Medium, descriptive): Speed plugins and cookie banners delayed Google Analytics on 10% of tagged sites: their tags waited for a scroll or an accept. Visitors who leave without scrolling go uncounted there.
• WEB-M3-02 (Medium, descriptive): 10% of the Tag Manager containers we could read hold no tags at all: installed, but running nothing. The Google tag may still run separately.
• WEB-M3-03 (Medium, descriptive): 3% of sites with a Google tag still send data only to Universal Analytics, which Google stopped processing in July 2023. Small count.

Count the calls and forms, not just visits

• WEB-M3-01 (Medium, confirmed): Only 22% of tracked sites in the confirmation sample count calls, forms or link clicks (interval 19% to 27%); 16% listen for phone taps. Read from the published Google tag; conversions set elsewhere are missed.
• WEB-W1-01 (Medium, descriptive): Only 6% of businesses tag their Google listing's website link, so most cannot tell visits from their profile apart from other traffic. GA4 can still attribute some Google traffic without tags.
• WEB-T4-03 (Medium, descriptive): Among tracked sites, counting calls and forms ranges by trade from 14% of handymen to 40% of plumbers. Each trade holds 51-93 tracked sites.

Say what you do and where, where Google reads it

• WEB-S2-01 (Medium, descriptive): Where a homepage title names the service, 80% of the time it is because the business name already contains it. Service words per category from a fixed list.
• WEB-L2-01 (Medium, descriptive): Listings that link to a location page instead of the homepage land on pages naming their city in about 0.9 more places. Small group of listings.

Keep the site light

• WEB-T3-01 (Medium, descriptive): What a homepage loads goes with how slow it is: the number of third-party services correlates 0.62 with the time the page is blocked. One mobile Lighthouse lab run per homepage.
• WEB-T3-02 (Medium, descriptive): Homepages with a tag manager score a median 50 on Lighthouse's mobile test, against 64 for those without one. Tags cost speed; speed shows no association with rank.
• WEB-T4-04 (Medium, descriptive): Median Lighthouse mobile scores range by platform from 70 (Next.js) to 25 (Weebly/Square Online); Squarespace sites score a median 52. One lab run per homepage; platforms with 20+ sites.

Know who you are up against online

• WEB-O1-01 (Medium, descriptive): Only 13 of 279 local business sites in the first sample appear in Google's organic top 30 for their own service-and-city search. First sample only; searched once per pair, 2026-10-23.
• WEB-O1-02 (Medium, descriptive): Directories and platforms hold at least 29% of the organic top 10 for local service searches; the businesses studied hold 14%. First sample; directory list is hand-made, true share is higher.
• WEB-O2-01 (Medium, descriptive): 74% of local business domains in the first sample carry the same spam network's footprint, about 2,450 referring domains each, inflating raw backlink counts. First sample; footprint from one vendor's index, 2026-10-23.

Part 3. Strategies: not one-size-fits-all

Sites sit at three levels of investment along a continuum, and what doing it well looks like depends on the platform and the trade. Limited evidence, to be tested in the next study.

Patterns found while exploring this data. Each is to be tested on new markets before it is treated as a finding.

The practices come as a package

• WEB-T4-05 (Medium, descriptive): Review count goes with 28 of 32 website practices measured: larger businesses tend to do more of everything at once. Correlations, q < 0.05 over the whole matrix.

Three levels of investment

• WEB-A1-01 (Medium, pattern): 23% of sites are instrumented: tag manager, conversion tracking and ad tags; 53% reach the top 10, with a median 48 reviews. Limited evidence: levels found while exploring; to be tested next.
• WEB-A1-02 (Medium, pattern): 49% of sites are tidy but untracked: schema, meta descriptions and canonical tags, rarely a tag manager; 42% reach the top 10. Limited evidence: levels found while exploring; to be tested next.
• WEB-A1-03 (Medium, pattern): 28% of sites are bare: little on-page SEO and usually no measurement; 31% reach the top 10, with a median 8 reviews. Limited evidence: levels found while exploring; to be tested next.
• WEB-A1-04 (Medium, pattern): Once reviews are counted, instrumented sites reach the top 10 little more often than bare ones (odds ratio 1.25): the levels largely mark business size. Limited evidence; the interval is wide.
• WEB-A1-05 (Medium, pattern): Instrumented sites are the slowest (median Lighthouse 47, against 68 for bare sites): tracking costs speed, and speed shows no association with rank. Limited evidence; one mobile lab run per homepage.

Same strategy, different platform

• WEB-A2-01 (Medium, pattern): On Squarespace, heavier homepages go with a top-10 rank (correlation 0.42, against 0.09 across all sites), likely because they carry more content. Limited evidence: one platform, found by scanning; to be tested next.
• WEB-A2-02 (Low, pattern): On GoDaddy's builder, a review widget comes with a missing H1 (correlation 0.70): features arrive bundled by the template, not chosen one by one. Limited evidence: one platform, found by scanning; to be tested next.
• WEB-T4-06 (Medium, descriptive): Builder templates set the accessibility floor: the median Lighthouse accessibility score is 95 on Squarespace and 94 on Wix, against 89 on WordPress. Automated Lighthouse checks only; one run per homepage.

Same strategy, different trade

• WEB-A3-01 (Medium, pattern): Plumbers who run a Google Ads tag also tend to run a booking or quote tool (correlation 0.45, against 0.10 across all trades). Limited evidence: one trade, found by scanning; to be tested next.
• WEB-A3-02 (Low, pattern): Accountants' sites with a stale copyright year usually measure nothing either (correlation 0.60, against 0.10 across all trades): a neglected-site pattern. Limited evidence: one trade, found by scanning; to be tested next.
• WEB-A3-03 (Medium, pattern): In home health, older domains belong to businesses with more reviews (correlation 0.34); across all trades the two are unrelated (0.02). Limited evidence: one trade, found by scanning; to be tested next.

How to cite

datanephew (2026). SMB Web Practices Study, data release v1.0 (Maps snapshot 2026-10-24). https://datanephew.com/research/local-website-study/

License and data-source terms

• License: the data files and charts are released under Creative Commons Attribution 4.0 International (CC BY 4.0). Reuse them freely, including commercially, with credit to datanephew and a link to the study.
• Data sources: Maps results and backlink data were collected through DataForSEO, which confirmed that publishing these aggregates is permitted. Organic search results come from SerpAPI. Page-speed scores come from the Google PageSpeed Insights API, with real-visitor field data from the Chrome UX Report (CrUX, CC BY 4.0) where Google publishes it. Domain registration dates come from public RDAP records. Every file holds aggregate statistics only: no business names, identifiers, addresses, phone numbers or website addresses, and any group of fewer than 10 businesses is suppressed.
• Google, Google Maps and Google Business Profile are trademarks of Google LLC. This study is not affiliated with or endorsed by Google.