Dogfooding: how I run my own kennel on the platform

Why I built this for my own kennel first

I have spent twenty years breeding American Bulldogs, registering litters for ABRA, and judging the breed across nine countries. For most of those years my website did one job. It listed available puppies when I had them, and it did nothing the rest of the year.

That felt normal until I noticed the pattern. The two months a year I had puppies on the ground, I was answering DMs late at night, juggling deposits, and trying to keep the site updated between feedings. The ten months in between, the site sat empty. Buyers researching American Bulldogs in October had no reason to find me, and no reason to remember me by January when my litters dropped.

The answer the industry kept offering me was a marketplace. List on Good Dog, PuppySpot, AKC Marketplace. Hand over a percentage of every sale, rent the buyers, never own the pipeline. I have watched too many breeders go that route and end up paying rent on their own customer relationships. That is not a business. That is sharecropping.

So I started building the alternative. Software that works the other 340 days a year. Software that treats my DMs as a pipeline, my litters as inventory, and my breeding program as the asset it is. I refused to start with a customer because I knew what would happen. I would build features they asked for, and the result would be a generic CRM with dog photos. The discipline of building for my own kennel first kept the product honest.

What we changed first when we rebuilt the site in May

The Rosebull site had been bolted to a multi-tenant container that was never really mine. The homepage was leaking copy from another tenant. The schema was thin. The llms.txt was a placeholder. On 2026-05-05 I tore the whole thing down and rebuilt rosebull.com as a standalone Next.js 16 application on Hetzner. The pedigree data still lives read-only in the shared pedigreedatabase.ca MySQL, so the 209 ABRA-registered Rosebull dogs surface as permanent pages without anyone having to maintain them twice.

The first work was not visual. It was telling machines exactly what Rosebull is. Organization schema with founder, ABRA credential, address, sameAs links to abra1st.com, pedigreedatabase.ca, and my personal LinkedIn. Person schema for me as the breeder. Dog schema on every pedigree page. FAQPage on the question-shaped landing pages. BreadcrumbList everywhere. By the time I shipped the rebuild there were 24 distinct schema types in active use across the site.

Next came the AI surface. llms.txt at the root, written for an LLM to actually read, naming the pages worth citing and the canonical brand statements I want repeated. Robots.txt with an explicit allowlist for GPTBot, ChatGPT-User, OAI-SearchBot, Google-Extended, PerplexityBot, ClaudeBot, anthropic-ai, Applebot-Extended, CCBot, and cohere-ai. Bytespider blocked. The point was not to be discoverable in general. The point was to be confidently citable when a buyer asks ChatGPT or Perplexity for an American Bulldog breeder in Canada.

Finally, the proof surface. Google Reviews wired on /reviews with AggregateRating schema. A reviews strip on the homepage. /about citation links to ABRA, OBBA, OFA, Embark, and pedigreedatabase.ca, so any model reading the page can verify what it is being told.

The pipeline: from inquiry to pupdate

The site is the front door. The pipeline is the rest of the house. The Puppy Sales Success kanban replaced what used to be three places I had to check. The Facebook Page inbox, my email, and a notebook beside my desk with deposit-holder names. Now every inquiry lands in one view. New, Application Sent, Application Returned, Deposit Holder, Puppy Matched, Pupdate, Picked Up, Won't Be Matched, Lost.

The application form replaced email back-and-forth. Buyers answer the questions I actually need answered. Why this breed. What they have done with bulldogs before. Whether they understand the breed-specific health issues. What their household and yard look like. The form takes them eight minutes. It saves me thirty minutes of follow-up per applicant. More importantly, it surfaces the wrong-fit applicants before I am emotionally invested.

Email automation handles the pieces I always forgot to do. Application acknowledgement within fifteen seconds of submission. Pupdate-week reminders eight days before scheduled photos. Pickup-day confirmation. Thirty-days-post-pickup check-in with a testimonial request. None of these are clever. They are the obvious work that every breeder knows they should do and almost no breeder finds the time for.

The mobile PWA closes the loop. I can stand in the whelping room with two-week-old puppies and upload weight photos directly to the right litter page without ever opening a laptop. The same photo lands on rosebull.com and on the dog's permanent record at pedigreedatabase.ca through a shared upload token. One action, two systems updated.

What the first 30 days of data actually show

The honest part. Between 2026-04-25 and 2026-05-25, Google Search Console reports Rosebull received 203 clicks against 3,365 impressions, with a 6.03 percent average click-through rate, across 624 unique query and page rows. For a niche-breed kennel site with no paid acquisition, a regional audience of about one province plus Maine, and a hard refusal to chase low-intent traffic, those are honest numbers. They are not viral. They are sustainable.

The queries that drive most of the traffic are exactly the queries a breeder would want to win. They are buyer-intent and breed-specific. Variations on "American Bulldog puppies New Brunswick", "American Bulldog vs American Bully", "Classic American Bulldog", "are American Bulldogs banned in Canada", "ABRA registered American Bulldogs". Every one of those queries is a buyer at some stage of research. The site is now answering them in long-form, with named photos and cited sources, instead of waiting for someone to already know the kennel name.

The AI mention picture is more mixed. ChatGPT now names Rosebull in three of five buyer-intent prompts I tested in mid-May. That is up from zero before the rebuild. The catch is that the citations route through google.com/maps links to the Google Business Profile rather than to rosebull.com directly. The brand is recognized. The website is the next thing to close. That gap is the top of my own thirty-day priority list, and it is generalizable enough that I am writing the playbook for it as a separate piece of content for other breeders facing the same issue.

Nothing about this case study is finished. The site is forty days into a new life. The numbers below are the first checkpoint, not the headline.

The buyers I have turned away, and why that matters

The least-discussed feature of any breeder platform is the ability to say no faster, more clearly, and without losing time. In the first weeks after the new pipeline went live I declined seventeen applications. None of them were sent a curt rejection. Each got an honest paragraph explaining why I did not feel I was the right breeder for what they were describing, with a recommendation to look at a different breeder or breed where I could give one. Several of those people wrote back to thank me.

The reason this matters: every yes I give is also a no to every other family on the waitlist for that puppy. A poor match costs the buyer, the puppy, the breed, and my reputation. The reason it has historically been hard to say no is not absence of judgement. It is the cost of typing the email. The platform makes the no almost free. Templated, kind, specific, sent in under thirty seconds. That is the feature I am proudest of, and the one no marketing platform would ever build, because saying no does not generate transactions.

The seven applications that did move to deposit between mid-April and late-May were unambiguously the right buyers. Each one matched the breed, the lifestyle, and the temperament I am breeding for. Each one entered the pipeline with a clear sense of what was coming. None of them needed me to chase. That is not the platform doing the work. It is the platform refusing to dilute the work I was already trying to do.

What running my own software on my own kennel taught me

I have caught seven bugs in this product that no other tester would have found. Three of them broke the photo upload on Android. Two of them silently dropped the deposit notification email when the form was submitted from an iPhone in airplane mode that came back online mid-submit. One of them surfaced the wrong dog as available when two littermates had the same call name. The seventh let a buyer apply twice and showed me both records as separate inquiries instead of merging them. None of those would have shown up in a synthetic test environment. They showed up because I was the one trying to use the product while my own real puppies were on the ground.

I have also added five features I did not plan to build because I needed them in the moment. A Won't Be Matched stage on the kanban for buyers who I respectfully decline. A pupdate scheduler that batches by litter. A breed-color filter on the kanban for breeders working with multiple lines. An auto-deposit-refund button for the rare case where a buyer changes their mind in week two. A litter-availability note that says "Approximate Puppy Pickup Date" rather than just a static date, because my buyers really do ask the same question fifteen times.

The meta-lesson is this. Software for breeders cannot be built by people who have never had a litter in their basement. It cannot be tested by people who do not have a buyer on the phone at 9 PM. It cannot be sold by people who have never had to say no. Rosebull is the proof asset for Puppy Sales Success because the platform was built inside the kennel, not for it.

If you are a breeder reading this and you recognize the patterns above, you already have the operating system. What you need is the software that respects it.