Findable was a pSEO product for Webflow and Shopify: turn structured data into pages, keep those pages synced, and stop teams publishing thin keyword shells. The hard parts were the parts most landing pages hide: state, platform limits, source data quality, editorial refusal, and AI workflows aging while we were building them.
These are the lessons that still feel useful now that the product sits in the archive.
Build for a moving AI frontier, not a fixed workflow
When Findable started, AI product work still meant hand-building the primitives: prompt chains, review queues, retries, fallback rules, and the points where a human had to keep control. We were designing workflows while the model layer underneath them kept changing shape.
That frontier moved from replacing one manual step with one model call to running agent workflows across planning, implementation, and verification. The same shift changed our own software development: better coding models, stronger APIs, and autonomous agents made yesterday's product assumptions feel old quickly.
The lesson is not to chase every model release. It is to design for churn: isolate AI steps, keep data contracts explicit, and assume today's clever orchestration will become tomorrow's default API.
Fully dynamic sync is a product, not a script
The first publish can look like a spreadsheet, a template, and a button. Every later change is distributed systems work wearing CMS clothing.
Dynamic sync means handling changed rows, deleted records, partial failures, field renames, CMS limits, rate limits, retries, duplicate prevention, and platforms that do not behave the same way twice. Webflow and Shopify make publishing easier; they do not remove the product work around state, review, and recovery.
If you build this from scratch, treat sync as reconciliation. Users need to know what changed, what failed, what was skipped, and how to replay a run without creating duplicate or stale pages.
Webflow and Shopify break in different places
Webflow breaks at the design and CMS boundary. Teams get visual control, then run into collection structure, CMS limits, publish flows, and the reality that editors still need to maintain the system after launch.
Shopify breaks at the commerce and search-intent boundary. Products, variants, collections, locations, and use cases sit close to revenue, which is useful. They also make it easy to create pages that exist only because a keyword exists.
The platform adapter should encode what each platform makes easy to abuse. The real question is not "can we publish this page?" It is "does this repeatable intent deserve a page, and does the source data make it more useful than a generic answer?"
Proprietary data is the durable advantage
AI made writing abundant; it made evidence scarce. That makes proprietary data more important, not less.
The pages worth indexing come from material competitors cannot easily copy: real products, locations, service areas, customer questions, job examples, reviews, pricing signals, fulfilment details, stock, use cases, and internal expertise. The model can shape the page, but the advantage comes from the source material.
Data advantage is not just possession. It is structure, freshness, permission, and enough context for an editor or model to turn facts into a page that could not have been written by anyone else.
The slop is real, and restraint is the filter
AI slop is not just ugly writing. It is content with no cost of refusal: pages published because generation was cheap, not because searchers needed them.
Programmatic SEO needs the opposite habit. Some keywords deserve a page. Some belong inside a guide. Some should be ignored. The difference comes from search demand, internal links, proof, and the editorial nerve to leave easy traffic on the table.
If you are applying AI search optimisation to a local trade, construction, or home-service website, BackPocket's AI Website Readiness Checker is a useful place to start. It checks whether answer engines can read the site, understand the services and service areas, and find enough proof to recommend the business.
Findable now sits in the Workhorse project archive. The lesson that survived the product is simple to say and hard to operate: build around real data, keep sync boring, and publish fewer pages with more reason to exist.