Methodology
Every figure in this report can be traced to a step described here. Data collected 29 July 2026. Corrections and data requests: hello@webnua.com.
What was studied
2,775 accountancy practices across Ireland. This is not a claim to have surveyed every accountancy practice in the Republic — no such complete public list exists. It is the deduplicated set of practices discoverable in Google's business-listing data at the collection date, which is the population that matters for a study about online visibility, but which necessarily excludes any practice with no listing at all.
Sampling frame, and its known bias
Firms were sourced from a licensed mirror of Google business-listing data, queried as 108 overlapping 30km-radius circles on a grid covering the Republic of Ireland, across ten accountancy-related Google business categories. Northern Ireland was excluded by postcode and country code. Records were deduplicated on Google place ID: 6,829 raw records resolved to 2,775 practices after exclusions.
Independent cross-check. A separately built list of 2,454 Irish accountancy firms, compiled earlier by a different method, gives 32.5% with no website. This study's independent census gives 32.9%. The agreement between two independently constructed datasets is reported as a robustness check, not as proof of completeness.
County assignment
Google populates a region field for only 49% of listings, with a strong rural bias — country addresses carry “Co. Tipperary” while Dublin city addresses routinely omit the county. Using that field alone undercounts Dublin roughly fivefold. Counties were therefore resolved in three passes, most authoritative first.
| Pass | Method | Resolved |
|---|---|---|
| 1 | Explicit region field | 1,365 (49%) |
| 2 | Town-to-county lookup from a hand-checked 2,454-firm list, plus county names parsed from the address | 1,325 (48%) |
| 3 | k-nearest-neighbour on latitude/longitude against pass 1–2 labels | 85 (3%) |
Validation. 300 firms with authoritative region labels were held out and re-predicted by the geometric pass alone. Accuracy: 98.0%. Since that pass was needed for only 3% of records, county-level error from this step is under 0.1% of the sample.
Website crawl
Every firm listing a website received exactly two HTTP requests: its homepage and its robots.txt, from an identifying user agent. Only publicly served HTML and response headers were examined. No authenticated areas were accessed, no forms were submitted, no vulnerability probing of any kind was performed, and no access control was circumvented.
Key definitions: does not load means no response after two attempts or a status of 400 or above. No structured data means no JSON-LD block in the homepage HTML. Stale copyright means the latest year beside a © symbol is 2024 or earlier, computed only across the 697 sites that display one. Accessibility checks are homepage-level proxies for European Accessibility Act readiness and are explicitly not a WCAG conformance audit.
<label for="id"> — and overstated the failure rate at 86.9%. All reachable sites were re-crawled resolving proper label association. The corrected figure is 52.9% of sites with a form having at least one control with no accessible name. Because a placeholder is not a valid accessible name but is a materially different failure from no label at all, a conservative floor treating placeholders as labels is also published: 27.9%. Both are reported; neither is used alone.Search visibility
For each of the 122 towns containing three or more practices, the query accountant [town] was run against Google using that town's own centroid coordinates, derived from the mean position of the firms in it — not the operator's location. Depersonalised, no logged-in session, desktop, depth 20. Firms with no website are counted in the denominator and recorded as not ranking, since that is the substantive reality.
What is deliberately excluded
The AI-recommendation layer — what ChatGPT, Gemini, Claude and Perplexity name when asked to recommend an Irish accountant — was designed but not run for this edition. Single-shot LLM queries are not reproducible; doing it properly requires repeated sampling and variance reporting, so it is deferred rather than published thin. Mystery-shopper enquiry response rates were also not run. Core Web Vitals field data is not reported, because the field dataset has insufficient traffic for the large majority of sites in this sample.
Publication rules
Aggregate only. No individual firm is named, ranked, scored or identified in any public output, in either direction. The published dataset carries no firm name, domain, address or Google place ID — each row has a surrogate identifier and county-level geography, with review counts reduced to bands. Findings are reported at county level only where a county contains at least 20 firms. No dataset about named individual accountants was constructed at any point.
Limitations
- The frame cannot see practices with no Google listing (§02).
- Homepage-only crawling understates firms whose schema, pricing or accessibility work lives on deeper pages. All such metrics are therefore upper bounds on absence.
- Consent and accessibility checks read static HTML and cannot observe JavaScript-injected banners, server-side gating or runtime ARIA. This makes the cookie finding conservative: a site whose tracker and consent banner are both injected by JavaScript registers as neither, so it is not counted as a violation.
- Security-header checks measure presence, not strength or correctness of policy.
- Search results are a single point-in-time observation and will drift.
- Category assignment is Google's, not a professional body's; some bookkeeping and tax-agent businesses are included that a chartered body would classify differently.
Reproducibility
The anonymised dataset is published free under CC BY 4.0 with no registration. Journalists, academics and analysts may request the full methodology walkthrough, including the collection scripts, at hello@webnua.com. We will publish corrections to this page if any error is identified, and re-run the study in July 2027.