The Smallest Possible Proof

The claim

Northwest Lagotto is a Lagotto Romagnolo breeding program on ten acres in Lynden, Washington, run by one person, with no marketing budget and no industry glamour. As of today, it is the answer a live web search gives, first and with the most detail, to the question a Pacific Northwest family actually asks: who is a good Lagotto Romagnolo breeder near me. I know this because I ran the search, dated it, and I am reporting exactly what it returned — including the parts that do not flatter the argument.

I also built the site that made this true, which is the whole reason this case study exists. Clarion Studio's practice is built on a claim: that AI search does not check the size of a business before it checks the structure of the evidence behind it. Clarion's own 130-firm research audit tested that claim against the wealth-management industry, at scale, with a stratified sample and a documented methodology. This piece tests the same claim at the opposite extreme — one entity, zero budget, a niche narrow enough that most people have never heard the breed's name — because if the mechanism is real, it should not care about the difference.

The confound, disclosed

Here is the complication, stated before anything else, because it is the complication a careful reader will spot anyway. I am the breeder. I am also the builder. Northwest Lagotto is a business I own and operate; Clarion Studio is the digital-infrastructure practice I founded, and this case study is, among other things, an advertisement for what Clarion sells. Every claim below should be read with that interest in mind.

In most case studies, that arrangement would be disqualifying. A vendor demonstrating the value of its own work on its own asset is not independent evidence. The reason I am publishing this one anyway is that the confound runs the other direction than it usually does.

A typical client engagement has to manufacture, artificially, something Northwest Lagotto already had for free: the compression between what an expert actually knows and what gets written down. When Clarion works with an independent registered investment adviser, the process starts with extraction — interviews, document review, a slow translation of years of tacit judgment into language a schema can hold — and information gets lost at every step, because the person who knows the thing and the person building the site are not the same person.

Here, they were. Everything I know about Lagotto Romagnolo health testing, breeding-line provenance, and what actually distinguishes a responsible program from a marketing one went directly into the entity graph, because I wrote both. There was no handoff, no interview transcript to lose nuance in, no builder guessing at what the expert meant. The same person who knew the answer built the page — which is either the weakest part of this case study or the fastest possible proof, and I think it's the second, for one reason: the site is checkable independent of who built it. You do not have to trust my account of what NWL knows about Lagottos. You can read the site, check the schema, and run the searches yourself.

I am disclosing this in the second section rather than a footnote for the same reason NWL's own site discloses when a health test is still pending rather than waiting to be asked: an omission volunteered late reads like a confession, and one volunteered first reads like a policy. This piece follows that policy throughout: the entity-graph drift Clarion's own audit caught and fixed in NWL's schema, the eye certifications still pending, the litter-status language search engines are still serving stale, all of it is in here, not because it helps the sales case, but because leaving it out would make the rest of the piece unverifiable by the standard it's arguing for. This is not a testimonial. It is a specification, applied once — and specifications are meant to be checked, not believed.

Why a kennel of one is a legitimate stress test

Clarion's own research report audited the wealth-management industry specifically because independent financial advisers sit at an extreme: high-consideration purchases, credentialed principals, a regulatory record that exists whether or not the firm bothers to link it. Northwest Lagotto sits at a different extreme, on nearly every dimension that report used to describe difficulty.

There is no marketing budget in the ordinary sense. What NWL has instead is my own time, spent building the site by hand instead of billing it to a client. There is no brand recognition outside the breed itself — Lagotto Romagnolo is a rare Italian truffle-hunting breed most Americans have never heard named, let alone seen. There is no glamour: nobody writes trend pieces about dog breeding the way they write them about wealth management or private equity. And the geography is narrow by design: a regional buyer pool in the Pacific Northwest, not a national or global market.

What NWL does share with the audited RIAs is the actual mechanism the report tested: a named specialization (Lagotto Romagnolo, specifically, not "dogs"), content depth around that specialization (a journal running dozens of essays on health testing, temperament, and breed history), an external footprint, and a verifiable identity behind the claims. If the four-signal framework from Clarion's research really describes how AI search evaluates any entity, and isn't a wealth-management-specific artifact, it should predict the same outcome here — at zero budget and zero glamour — that it predicted at scale across 130 audited firms. That's the actual test this piece runs. Not whether a good breeder can build a nice website. Whether the mechanism generalizes.

What was true before

The current Northwest Lagotto site went live in April 2026. Before that, NWL ran on a Wix-built site, the templated, drag-and-drop kind that gets a small business online quickly and gives it almost nothing else. I know the date because I built the replacement and moved the domain myself; independent corroboration isn't hard to find either. A site:nwlagotto.com search run July 11, 2026, still returns URL patterns that don't match the current site's architecture at all — /available-puppies, /meet-the-breeder, /post/searching-for-lagotto-romagnolo-puppies..., /breeding — the slug conventions of a page builder, not the hand-built HTML the site runs on today. Those pages are gone. Google's index hasn't fully caught up to that yet, which is itself a small, live demonstration of the lag this piece returns to later.

What the old site didn't have is the actual subject of this case study. No entity graph. No Organization or Person schema binding claims to a checkable identity. No structured link to AKC Marketplace, to the Lagotto Romagnolo Club of America, to an OFA health record — nothing an AI system could verify, only prose it could read and set aside, the same condition Clarion's research found in ninety-five percent of audited advisory firms, the share scoring the lowest possible E-E-A-T signal strength. The dogs were the same dogs. The health testing was the same testing. The expertise didn't change in April 2026. What changed was whether any of it was written in a form a machine could check, rather than merely a form a human could believe.

That's the actual before-and-after this piece can honestly claim, and it's a narrower claim than "traffic went up" or "we get more inquiries now," deliberately narrower. I'm not publishing inquiry counts or revenue figures here, and I'd be skeptical of a case study that did, because neither is independently verifiable by a reader and both would be too easy for a founder to shade upward. What's verifiable is the record itself: what a Wix site with no structured data looked like to a machine, and what a hand-built entity graph looks like to the same machine, three months later, on the same underlying business. The gap between those two states isn't a rebrand. Templates and hand-built sites can look similar to a human visitor — same photos, similar copy, a comparable sense of polish. To a machine reading for verifiable structure, they aren't close. One publishes claims. The other publishes claims and the means to check them.

The method, applied at n=1

Clarion's research report specifies seven layers a complete entity graph needs: Organization, regulatory verification, Person, credential, content, discovery file, and propagation. Zero of the 130 audited RIA firms have all seven. Here is what those same seven layers look like, built once, by hand, for a business two orders of magnitude smaller than the smallest firm in that sample. The seven-layer architecture itself, taught from first principles with worked examples across two other professions, is walked in detail in How AI Verifies You're Real, including a live tool that scores any site against the same rubric.

Two of those seven rows deserve more than a checkmark. The regulatory-verification layer is the one Clarion's research calls the most striking finding in the study, precisely because it's the cheapest layer to build and the one almost nobody builds: zero of 130 audited RIA firms link a single regulatory body from their own site. Northwest Lagotto links three — AKC Marketplace, LRCA membership, and per-dog OFA records — not because dog breeding carries more regulatory infrastructure than registered investment advice (it doesn't; there's no SEC-equivalent for Lagotto breeders), but because the equivalent authorities exist and somebody bothered to link them.

The propagation layer earns a note, because it's the row where the audit process itself is visible. Clarion's July 2026 audit of NWL's schema — the same audit standard applied to client sites, run on the studio's own reference asset — found the canonical Organization, Person, and Website entities defined inline, redundantly, on multiple pages instead of declared once and referenced by @id everywhere else. The kind of drift that accumulates quietly on any site that grows. It was remediated across the full site the same month, before this piece published, and the check that caught it now runs as a standing part of the audit protocol, so the same drift doesn't come back unnoticed. That's the pattern worth taking from the row: entity graphs aren't built once — they're maintained, and the maintenance is checkable too.

This isn't a Clarion-only claim, and it's worth being precise about how much weight the wider category evidence can bear. Industry analyses — not peer-reviewed, and held to a lower confidence than Clarion's own audit — suggest that roughly sixty-one percent of ChatGPT-cited pages carry rich schema markup, against roughly twenty-five percent of standard search-result pages, and separately suggest a thirty-six percent lift in AI-summary appearance tied to schema deployed sitewide rather than on a single page. Read those two numbers as directional, sourced to technical-SEO agency content rather than independent research, not as findings Clarion verified itself. The 130-firm audit is the number I'd actually stand behind; the industry figures just point in the same direction.

What's findable today

Everything above is what NWL built. This section is what a search actually returns, checked and dated, including the results that complicate the story.

Ten live web searches ran July 11, 2026 — standard web search, not an AI-engine query. I'm holding that distinction through this section rather than letting the two blur together; the AI-engine transcripts follow separately, below.

For "Lagotto Romagnolo breeder Pacific Northwest," NWL appears twice on the first page — homepage and About page — alongside five genuine regional competitors, and the synthesized answer names NWL first and with the most specific detail: Puppy Culture, CHIC testing, the ten acres, the years with the breed. For "best Lagotto Romagnolo breeder Washington state," the same pattern holds: NWL first, ahead of four named alternatives. Both are real fields, not empty ones. The honest claim is that NWL is the most consistently surfaced answer to a real regional query with real competition, not the only result in an empty field.

For "Northwest Lagotto reviews," the picture is less flattering, and I'm reporting it that way on purpose. The synthesized answer mostly pulls testimonial language from NWL's own site, plus a Yelp listing and an unaccredited BBB profile. Most of what's currently findable about NWL's reputation traces back to NWL's own domain, not to an independent third-party review platform. That's a real gap, not a rounding error, and it isn't one a schema fix solves by itself. A related, smaller finding: the same search surfaces lagottonw.com, a different, unrelated breeder in Vancouver, Washington, confirmed as a distinct business, not a redirect or an NWL property. It's a name-confusion risk for a buyer typing carelessly, not a technical problem, but it's the kind of gap an entity graph doesn't close on its own.

For a generic breed-fact query — "the only breed developed for truffle hunting" — NWL isn't cited at all. The synthesized answer draws on Wikipedia, the LRCA, and general breed-authority sources. This is the negative finding, and it's the one that actually sharpens the claim rather than undermining it: NWL wins geography-plus-breeder queries, the ones an actual buyer asks. It doesn't win generic breed-education queries, the ones with no purchase intent behind them. A case study that claimed the second kind of win would be overclaiming. This one doesn't.

Those ten searches used standard web search as a proxy for framework fit. The AI-engine queries are a separate exercise, and they were run separately: eleven captures across ChatGPT, Perplexity, and Google's AI Overview on July 11, 2026, each logged with its engine, account condition, and timestamp, with full transcripts and screenshots archived. Two conditions are disclosed because they matter. The first ChatGPT runs came from a logged-in account, and the model itself volunteered that it knew the asker was associated with Northwest Lagotto — so the money queries were re-run in a memory-off temporary session, and both conditions are reported. And map-card results may lean on geolocation, so the text of each answer, not the map, is the datum. Every query below can be re-run by any reader, on any date, against what's reported here.

One last finding cuts against Clarion, not NWL, and it belongs here because the discipline in this piece only means something if it runs in both directions. NWL's robots.txt carries an AI-crawler allow-list naming GPTBot, ClaudeBot, PerplexityBot, Google-Extended, and more, explicitly welcomed rather than blocked by default. Clarion's own site had the opposite problem until this same week, when a default-on Cloudflare setting quietly blocking those same crawlers was found and fixed during the audit that produced this case study. The subject of this piece had the plumbing right before the studio auditing it did.

What the audit found, and what happened next

A case study that only shows what works isn't evidence — it's marketing. So here is what the audit process found when it was pointed at the studio's own reference asset, and what happened next.

It found drift. Canonical identity entities had come to be defined inline on multiple pages instead of declared once and referenced everywhere — the kind of entropy any growing site accumulates. It was found by the same audit standard Clarion runs on client sites, fixed sitewide before this piece published, and the check that caught it is now a standing part of the protocol. Finding your own drift before a client or a competitor does is what the discipline is for.

The more interesting finding is one no site fixes by deploying anything: the external record trailing the entity by weeks. NWL's own internal audit corrected two claims that were live and wrong — a stale summer-litter announcement contradicting an already-closed litter elsewhere on the site, and language describing CHIC certification as complete when the annual eye exams a current CHIC number requires were still being scheduled. Both corrections are deployed to production; I checked by fetching the live site directly on July 11, 2026, not by trusting my own memory of having fixed them. The current homepage now reads "Autumn 2026 Litter, Inquiries Open," and the health-testing language now says exactly what's true: DNA-clear or DNA-carrier status confirmed for the two breed-specific conditions, OFA hips completed, eye exams pending, certification expected this summer. Search engines, that same morning, were still serving the pre-correction claims — the old litter status, the "CHIC-certified" language the corrected page no longer uses. Two of the AI-engine captures in the evidence above still carry that stale phrasing. But by the end of the same day's capture session, Google's organic snippet for NWL already read from the corrected page: "health tested, publicly verifiable at ofa.org." The entity changed, the external record trailed it by weeks, and then it caught up — morning and evening of the same day, both states dated and screenshotted. That is what the lag looks like when you can actually see it move.

The owner of an entity graph doesn't control that lag. What the owner controls is which version of the record the recrawl finds.

Borrowed visibility erodes on a lag measured in weeks to months, not years, and the firm that owns its own record closes that lag faster than one that depends entirely on someone else re-crawling it.

That gap has a name and a shelf life, and it's the same shelf life Clarion's research argues AI visibility itself has. NWL isn't exempt from the lag. It is, right now, a dated, checkable example of it.

What this replicates

None of the seven layers above are Lagotto-specific. The Organization entity, the regulatory-verification links, the Person entity, the credential declaration, the content-to-author binding, the discovery files, the propagation discipline — every one is a generic mechanism instantiated here with dog-breeding vocabulary. Swap in a CFP designation for an AKC credential, an IAPD record for an OFA record, a client-facing essay on Social Security timing for a journal essay on health testing, and the structure is identical.

That's the actual service Clarion sells, described plainly rather than through a sales page: extracting the tacit expertise a principal already has and building the entity graph that makes it checkable by a machine — the same work this case study performed on my own business, minus the confound, because for a client the expert and the builder are two different people, and the extraction has to happen on purpose rather than by accident of who's typing.

The honest version of this section isn't "hire Clarion and get NWL's results," because results here are explicitly the wrong thing to promise. I've said above that I'm not publishing inquiry numbers, and a different business in a different niche with different competitors would see a different outcome from identical infrastructure. What replicates is the method, not the number. A firm with a genuinely narrow specialization, real content depth behind it, and a principal willing to have actual credentials checked rather than merely stated has the same raw material NWL had before April 2026. What NWL did next is what an engagement does for someone else's expertise instead of mine.

The argument, restated

The argument this piece has been running is narrower than it might look from a distance: not that AI search will make any business a market leader, and not that a website is destiny. Only that the mechanism Clarion's 130-firm research documented — AI search rewarding verifiable structure over prose, at whatever scale it finds it — held up when tested against the smallest, least glamorous entity I could put it in front of, and it held up with real gaps still showing, not despite them.

AI search does not check the size of the business before it checks the structure of the evidence. Zero of 130 audited RIA firms link a single regulatory body from their own site. A ten-acre dog-breeding program in Lynden, Washington links three. Neither of those facts is about dog breeding, and neither is about wealth management. Both are about whether the evidence was ever built to be checked.

I own Northwest Lagotto. I founded Clarion Studio. I have a direct financial interest in this case study being persuasive, and I've tried to write it in a way that would still be true if it weren't. Read the schema. Run the searches. Check the dates. That's the whole offer.

FAQ