# 200 insurance agency websites: 50-state audit methodology

Collected September 23, 2026. Unit of observation: one directory-linked page on each of 200 distinct website hosts. Allocation: four websites per state across all 50 US states. DC and territories are outside this design.

## Selection

The sampling frame is the public Travelers independent-agent directory, https://agent.travelers.com/, using its published sitemap agency-detail URLs. A listing must supply an agency name, an office address in the assigned state and an external website. State refers to that listed office, not necessarily corporate headquarters. Directory affiliation establishes eligibility; corporate ownership and licensing were not independently audited.

Within each state, candidates are ordered by SHA-256 of `webdimonia-us-50-states-2026-v1|` followed by the complete directory URL. States are processed in alphabetical postal-code order. Hostnames are lowercase with the `www.` prefix removed. Previously selected requested or final hosts are excluded. Different hosts can belong to the same corporate group.

Select the first four eligible, readable agency destinations per state. Replace unavailable responses, non-agency destinations, unresolved identities and duplicates with the next candidate from that state. Exclusions include bank and general professional-services homepages and agency membership networks, even when the directory identifies an insurance subsidiary. Selection does not depend on contact or technical feature counts.

The selection log records 431 candidate attempts, including candidates considered during duplicate resolution. Rows beyond a state's final quota are labeled accordingly. Seven destinations were excluded during identity review; the reasons are recorded in the log. Collection failures do not establish that a website is unavailable to ordinary visitors.

## Collection

Inspect one directory-linked page per website, preserving its path and recording its final address. Prefer HTTPS for directory links supplied as HTTP and strip query parameters. Check public robots directives; skip when robots cannot be checked successfully, except a 404 or 410. Requests use bounded timeouts, a 3 MB body limit and at most four state workers. No credentials, access-control bypass, calls, emails or form submissions are used.

The public dataset records the directory listing, listed agency name, city, state, requested URL, final URL, collection time and SHA-256 hash of the received body. Agency names retain directory labels; an agency's current public brand can differ. Saved source bodies remain private.

## Measures

Every flag is true or false for the 200 included pages. All charts use 200 as the denominator.

| Flag | Detection rule |
| --- | --- |
| `https` | Final response URL uses HTTPS |
| `title_present` | Nonempty HTML title |
| `description_present` | Description metadata with nonempty content |
| `viewport_present` | Viewport metadata exists |
| `canonical_present` | Canonical link element exists |
| `jsonld_present` | At least one application/ld+json script exists |
| `jsonld_parseable` | At least one such block parses as JSON |
| `tel_link` | An anchor has a tel: destination |
| `mailto_link` | An anchor has a mailto: destination |
| `form_element` | An HTML form element exists |
| `quote_control_text` | A link or button contains the standalone word quote |

These are source detections. JavaScript-rendered content, other pages and embedded systems may not appear. A form can be for search or newsletter signup. Parsing JSON does not establish valid business schema. Presence does not establish visibility, usability, correct settings or successful contact. Speed, Core Web Vitals, mobile usability, lead volume and conversion rates were not measured.

## Validation

All 200 requested hosts and all 200 final hosts are unique. Directory address states match assigned states. Each of the 50 states contributes exactly four included pages. Agency names, page titles and redirected destinations were reviewed, with closer source inspection of ambiguous identities. Every saved response hash was checked, and all eleven flags were independently re-extracted from all 200 saved responses before publication.

## Interpretation

This is an equal-state sample of accessible websites linked from one carrier's directory. It is not population-weighted or a random sample of all US agencies. It includes local agencies and larger brokerages. Four observations per state support balanced geographic coverage, not state rankings or state-level prevalence estimates.

The findings help identify checks for an agency's own website. They do not rank agencies, establish customer service quality or measure commercial outcomes.