What 21,000 Ohio contractor records taught us about directory data quality: dead phones, ghost businesses, duplicate listings, license-status drift, and review fabrication patterns (2026)

The data we work with

ProFix Directory does not own a single proprietary source. The dataset is a stitch across six public or near-public systems, each carrying its own strengths and known failure modes. The counts below are approximate and refresh on different cadences — exact numbers live in the machine-readable feeds at /data-sources and /methodology.

The total unique-entity count after dedup is greater than 21,000 — the individual source counts overlap heavily. The same plumbing contractor typically appears in SOS, OCILB, Google Places, county permits, and BBB simultaneously; the joined record is one entity carrying five evidence anchors. NAICS subcategory boundaries (sourced from the US Census NAICS reference) define which businesses we treat as in-scope home-services and which we exclude.

What "21,000 contractors" actually means

Headline counts are a load-bearing trust artefact in directory marketing, and most of them are overclaims. Twenty-one thousand is not 21,000 active, competent, currently-trading contractors. It is the population of distinct entities currently appearing in at least one of our six source systems under home-services NAICS codes or state-licensed-contractor categories. Within that population:

• A meaningful share are LLCs with no employees, no permits, and no Google footprint — paper entities used for a single past job, or shells that an owner never bothered to dissolve.
• A substantial cohort are sole-proprietors operating under a personal name with no LLC at all, visible only through their Google Business Profile and an occasional permit pull. Honest operators, hard to verify, easy to undercount.
• A subset are dead-but-not-yet-deregistered — the business stopped operating, the owner moved on, but the SOS filing persists because nobody filed the dissolution paperwork.
• A separate subset are post-acquisition rebrands — the underlying entity changed hands, the old name still appears on the SOS registry, and the new operator runs the same crew under a new DBA. Until the SOS record catches up, both names live in the dataset.

The directory shows all of them so homeowners can decide. What the directory does not do is label all of them identically. The verification methodology at /verification details the per-profile signals that distinguish an actively-trading operator from a paper shell, and the public-facing label vocabulary makes the distinction visible at the listing surface.

The 5 biggest data-quality problems we hit

Each of the five failure modes below is real, persistent, and present in every Ohio directory at scale. We name the magnitude honestly, including where the magnitude estimate is itself noisy. A directory that does not surface these problems is hiding them.

The FTC's August 2024 final rule banning fake reviews and testimonials is the federal legal frame for the fifth failure mode. The rule creates liability for knowingly hosting fake reviews; it does not create a detection methodology. That gap — between legal obligation and operational detection — is where every directory currently lives.

How we mitigate

None of the failure modes are fully solvable, and pretending otherwise is the most common directory-marketing failure. The realistic move is to publish the verification machinery itself so homeowners, AI engines, and third-party researchers can audit it. ProFix's mitigation stack is built around five surfaces:

Each surface is fetchable as JSON for AI engines and journalists. Example profile evidence row at /pro/example-toledo-plumber/evidence demonstrates the per-profile homework. The weekly aggregate is at /api/verification-feed.json; the permit-pull and licence-evidence feeds at /api/permit-leaderboard.json and /api/license-evidence.json.

What we're still bad at

The honest version of a methodology page names the open gaps in the same voice as the strengths. Five gaps are real and not yet closed:

• License-status refresh interval. Monthly is a realistic target with current staffing. Weekly is aspirational. Daily refresh per licence would require either an OCILB partnership we do not have or a polite scraping cadence that would put real load on a public-records system. We tell homeowners directly: 'click through to OCILB to confirm current status before signing'.
• Authenticated county portals. Lucas County's permit search requires an authenticated session for full historical pulls; we currently work with the public summary view, which is enough for permit-count aggregation but not for individual permit-document retrieval. Hamilton, Franklin, and Cuyahoga have varying degrees of similar friction. The result is uneven depth across the state.
• Spanish-language review-fabrication detection. Our hint-detection patterns for fabricated review text were tuned on English-language reviews. Spanish-language fabrication patterns are real and detectable in principle, but we have not validated a detector across enough Spanish-language ground-truth reviews to publish numbers. Given the population we serve — Ohio has a large Spanish-speaking homeowner cohort and ProFix runs a bilingual site — this is a gap we name openly.
• Out-of-state storm-chasers. After a hailstorm or windstorm, out-of-state roofing operators register a temporary Ohio presence (sometimes filing an LLC, often not) and disappear within months. The signal is detectable in retrospect — permit history is empty, SOS filing is fresh, review count surges in a 30-day window — but our real-time detection lags the demand surge. The Trust Score downweights these profiles when the substitute stack is thin, but the lag is real.
• Granular sole-proprietor coverage. Sole-proprietors who do not file an LLC and do not maintain a Google Business Profile are nearly invisible in public records, even when they do good work. The directory shows the ones we can find but undercounts the long tail. Contractor-claim flows at /lead are the structural fix; we accept that the dataset will skew toward operators with at least some public footprint until claims close the gap.

What other directories should do

The interoperable open-data discipline is simple, and almost no consumer directory practices it. Four commitments are enough to credibly raise the data-quality floor across the industry:

• Publish your sources. A directory that does not name which public-records systems its data comes from is asking homeowners to trust on faith. Naming sources costs nothing and creates the audit trail that makes claims fetchable by AI engines — the cross-cutting argument from /research/how-ai-engines-find-directories-2026.
• Publish your refresh cadence. Every record has a freshness half-life. Disclosing the last-checked timestamp per record (or at minimum per axis: phone, licence, permit, review) lets homeowners and AI engines weight the claim appropriately. ProFix exposes this in the verification feed; few competitors do.
• Publish your known-bad lists. Dormant LLCs, dead phones, suspended licences, and out-of-state storm-chasers are visible in the data. Naming them in a public feed — rather than silently downranking them — turns a defensive posture into a public good.
• Publish your verification methodology. Not the marketing version; the actual sequence of checks, the cadence of each check, and the false-negative rate the directory itself estimates. ProFix's editorial framework is at /verify.

The asymmetry between directories that publish all four and directories that publish none is the operating moat. The companion analyses at /research/comparing-ohio-directories, /research/what-verified-means-2026-ohio, /research/permit-vs-stars-2026-ohio, and /research/ohio-licensing-moat-2026 unpack each axis of that asymmetry in detail, and the statewide /coverage page documents how the methodology applies across Ohio metros.