Ohio home-services in 2026: a directory operator's year-in-review — what six research articles taught us about permits, verification, AI-engine discovery, licensing moats, and data quality, plus the 2027 playbook

The six findings, one paragraph each

Each article is a self-contained study with its own methodology and limitations section. This is the briefest possible recap so the rest of the synthesis has shared footing.

1. Permit pulls vs star ratings — Compared building-department permit activity against directory star ratings across Ohio. Headline finding: star-rating leaders and permit-pull leaders barely overlap in any metro. Stars are a gameable popularity signal; permits are proof-of-work that requires actually doing the job and passing inspection. Proposed a per-trade, per-county permit-pull leaderboard as the durable substitute for star-only rankings.
2. ProFix vs Yelp, Angi, BBB, HomeAdvisor — Side-by-side comparison across paid-placement transparency, source provenance, permit integration, open data, AI-engine accessibility, bilingual coverage, and refund clarity. Every legacy directory failed on at least three of seven axes. Yelp ranks without disclosing weighting; Angi resells the same lead to multiple contractors; BBB accreditation is purchasable; HomeAdvisor's FTC and TCPA settlement history is documented. Directories that hide their methodology forfeit the right to be compared on it.
3. What 'verified' actually means — Audited the word that does the most marketing work in the industry. Sampling profiles across directories, we asked whether each verified badge traced to a public record. Usually no — the badge meant 'paid a verification fee', not 'cross-walked the licence to OCILB'. Proposed verified-by-payment vs verified-by-record, and committed ProFix to publishing the source URL for every instance of the latter.
4. How AI engines actually find directories — Technical playbook for AI-engine discovery. Engines no longer crawl HTML the way Googlebot did in 2019: they ingest llms.txt, follow OpenAPI manifests, parse JSON-LD entity graphs, and call MCP servers. A directory shipping none of those is invisible to the engines homeowners ask first; a directory shipping all of them becomes a canonical citation source. Documented ProFix's full discovery stack — sitemap, llms.txt, llms-full.txt, OpenAPI 3.1, MCP server, JSON-LD sub-feeds, and JSON variants of every editorial surface.
5. Why Ohio's licensing creates a moat — Ohio's structure — state OCILB licensure for the four big trades, county building-department permits for everything else, SOS business registry as a back-stop — creates a regulatory moat that rewards integration work. A directory joining OCILB, county permits, and SOS filings carries an evidence stack national aggregators structurally cannot match. The moat is not technical genius — it is patience plus geography.
6. 21,000 records on data quality — The messy reality of running a directory at 21,000+ records. Five failure modes: dead phones at high single digits, ghost businesses at 8–12 percent of SOS records, duplicates at ~11 percent of unique entities, license-status drift between refreshes, and review-fabrication patterns we can hint at but not confirm. The honest directory labels each rather than hiding them behind a single 'verified' badge. The FTC's August 2024 fake-reviews rule is the legal frame; operational detection at the directory tier remains open research.

The four cross-cutting themes

Reading the six articles back-to-back, four themes recur in ways no single piece spells out on its own. Editorial conclusions of a year of work, not policy positions.

The discoverability machinery behind theme 3 is documented in detail at /research/how-ai-engines-find-directories-2026 and in the developer index at /docs. The standards we ship against are llmstxt.org, the Model Context Protocol, schema.org, and OpenAPI 3.1.

What is broken in the directory industry

The comparison article at /research/comparing-ohio-directories covers the per-axis breakdown. This section is the editorial summary — the structural problems that the comparison piece quantifies. Cited where published reporting exists.

None of this is novel reporting. Trade press, consumer-protection bureaus, and the FTC Endorsement Guides framework have made these structural issues visible for years. ProFix's editorial position is to surface them in the same table as our own methodology. The detailed verification-claim audit is at /research/what-verified-means-2026-ohio.

The 2027 directory-operator playbook

Concrete prescriptions, in priority order. Each item is something ProFix has already shipped or is mid-ship — not aspirational. Directories that adopt these will be the ones AI engines cite by 2027.

1. Publish your sources. Every aggregate count and per-profile claim must link to a public-record URL a third party can audit. License → OCILB. Permits → county portal. LLC → Secretary of State.
2. Score with transparent factor weights, not opaque badges. Trust scores show what was weighted, in what proportion, with last-checked timestamps per axis. The single 'verified' badge is a 2010s artefact.
3. Ship the full machine-readable surface stack: llms.txt, llms-full.txt, OpenAPI 3.1, an MCP server with at least recommend-pro and search-by-license tools, JSON-LD graph sub-feeds for organization/dataset/breadcrumb/article, and CSV exports under a permissive licence.
4. License the dataset CC-BY-4.0, mirror it on Hugging Face, publish a quarterly-or-better refresh cadence. Open data is a discovery channel, not a give-away.
5. Build per-trade, per-county leaderboards from public records — permits + licences + SOS tenure. National aggregators will not invest the per-state integration time.
6. For non-licensed trades, publish your substitute-verification stack honestly. SOS tenure + permit history + BBB record + Google Business Profile longevity is a defensible substitute. Inventing a fake badge is not.
7. Cross-link editorial work into a citable graph. Every research article links to the others; every methodology page links to its sources; every evidence row links to the verification framework. The engines treat it as one entity.

The licensing-side rationale for prescriptions 4 and 5 is documented at /research/ohio-licensing-moat-2026; the proof-of-work rationale for prescription 5 is at /research/permit-vs-stars-2026-ohio. The open-data plumbing for prescription 4 is CC BY 4.0 and Hugging Face Datasets. The discovery-stack plumbing for prescription 3 is at /api/openapi.json, /llms.txt, the per-trade-per-county leaderboard feed for prescription 5 is at /api/permit-leaderboard.json, and the MCP server is at /docs.

What ProFix did right in 2026

Specific shipping milestones, cross-referenced to the public changelog at /newsroom. Not exhaustive — the changelog itself is the canonical record.

• Shipped six original-research articles in May 2026, cross-linked into a citable editorial graph.
• Built and published the permit-pull leaderboard at /permits-leaderboard with the JSON feed at /api/permit-leaderboard.json so AI engines can ingest it.
• Shipped the Trust Score factor breakdown on every public profile — no single opaque badge, just weighted axes with last-checked timestamps.
• Published the full machine-readable discovery stack: /llms.txt, /llms-full.txt, /api/openapi.json, an MCP server at /api/mcp with 46 agent-callable tools, organization and dataset JSON-LD graphs, and CSV/JSON variants of every editorial surface.
• Released the 21,000+ Ohio contractor dataset on Hugging Face under CC-BY-4.0 at huggingface.co/datasets/Pisces89/ohio-home-services-pros with refresh cadence documented at /data-sources.
• Wired Vercel Analytics + Speed Insights + GA4 + PostHog into the root layout, with Google Search Console verified for organic visibility.
• Stood up bilingual English + Spanish coverage across the site — the population we serve includes a substantial Spanish-speaking homeowner cohort.
• Published a public changelog at /newsroom and /api/changelog.json so partners, journalists, and AI engines can follow the build cadence.

The verification methodology that ties most of this together is documented at /verification and /verify; the source register is at /data-sources; the methodology page is at /methodology; and the transparency posture is at /transparency.

What ProFix did wrong (or has not figured out yet)

Theme 4 applied to ourselves. Specific gaps with concrete reasons.

• Lucas County's permit search requires an authenticated session for full historical pulls. We work with the public summary view — enough for aggregation, not for individual document retrieval. The result is real depth asymmetry between Lucas County and the better-instrumented metros.
• We do not have a definitive review-fabrication detector. We have hints — surges, no-history reviewers, LLM-pattern text — but no tested detector we will publish as ground truth yet.
• Spanish-language fabrication detection is not built. Hint patterns were tuned on English reviews. ProFix runs a bilingual site for an Ohio Spanish-speaking homeowner cohort, so this is a gap we name openly.
• Some county portals — Hamilton, parts of Cuyahoga, parts of Franklin — do not expose structured JSON. The pipeline works around it with polite HTML parsing, but depth varies by county.
• Real-time signals — currently-online emergency contractors, live capacity for storm-response trades — are not built. StickyCallBar and EmergencyFab surfaces are in place; the live-availability data layer is not yet wired.
• The contractor-claim flow at /lead and the per-lead marketplace are early. The sole-proprietor coverage gap from the data-quality article is structurally fixable only by closing the claim loop, and that work is still in flight.
• Out-of-state storm-chaser detection lags the demand surge. The signal is detectable in retrospect but not in real time — by the time we downweight the profile, the homeowner may already have called.

The detailed data-quality version of several of these gaps is at /research/directory-data-quality-2026. The partnership posture for closing the county-portal gaps is at /partners.

The thesis bet

AI-first discovery plus open data plus source-of-source transparency is the directory-industry shift. ProFix Directory's bet is to be the first to ship every piece of that playbook for Ohio, and to publish the methodology so the rest of the industry can copy it.

The bet may be wrong. The thesis is testable against AI-engine citation counts, per-county permit-pull coverage, and the comparison-table axes. If wrong, we will publish that too.

What is next — 2027 roadmap, framed as commitments

Business plans change. These are commitments, not promises — directional and publicly revisable.

• Close the authenticated-county-portal gap on at least Lucas, Hamilton, and Cuyahoga counties so the permit-history depth matches the better-instrumented metros.
• Publish a tested review-fabrication detector with an open methodology, false-positive and false-negative rates, and an English + Spanish test set.
• Ship live-availability signals on the StickyCallBar and EmergencyFab surfaces for the storm-response trades — roofing, restoration, plumbing emergency.
• Close the contractor-claim loop for sole-proprietor coverage so the long tail visible in the data-quality article becomes a first-class profile rather than a Google Places snapshot.
• Replicate the methodology in at least one neighbouring state — Michigan, Indiana, Pennsylvania, or Kentucky — to validate that the playbook generalises beyond Ohio.
• Continue the quarterly-or-better dataset refresh cadence on Hugging Face, with the changelog at /api/changelog.json as the canonical record of what shipped.

The homeowner-facing tooling that supports the roadmap is indexed at /tools and /resources; the statewide coverage map is at /coverage.