600+ location pages. Each one unique. Two human checkpoints.
A financial services company needed unique location pages for every city they serve across Texas. Manual estimate: over a year. Delivered in one month.
600+
Pages live
1,000+
Hours saved
1 month
Delivery time
2 humans
Quality checkpoints
Six hundred cities. One year of writing. Or twelve.
The company needed a landing page for every city they serve in Texas. Over 600 cities. Each page had to be unique, grounded in that city's census data, demographics, and market signals. Google penalises template swaps. Searchers see through them. The pages needed real local data and actual relevance to each community.
Writing 600 unique pages by hand? With a full-time writer, over a year. The agency running the project needed a fundamentally different approach.
An AI research agent. Four steps. Per city.
Four steps run for every city before the page goes live. The research is what makes the page about Corpus Christi different from the page about Plano. Not the template. The data underneath.
Census · Demographics
Census figures, median income, cost of living, population growth, local economic indicators. Verifiable numbers, not filler.
Per-city signals
Local search demand, competitor presence, which offerings match that community's demographics.
Drafted with context
Using all that research as context, the AI writes a landing page. A page about Corpus Christi reads completely differently from a page about Plano.
Review · Auto-publish
Pages flow through Airtable for review, then auto-publish to WordPress via Page Generator Pro. Over 600 pages through a single pipeline.
One month. Two reviewers. Six hundred pages.
- 600+unique, SEO-optimised pages live on the site.
- 1,000+hours of development time saved.
- 1 modelivery. Manual equivalent: over a year.
- 2human checkpoints — content writer plus QA reviewer — on every page before it went live.
- 0template swaps. Every page grounded in real data.
05 / The lesson
Scale creates a quality problem.
The first fifty pages were easy to check. Page 400? Page 600? You need a system for reviewing the system's output, or quality degrades silently.
The human checkpoints weren't optional. They were the reason this worked. A writer caught local nuances the AI missed. A QA reviewer caught formatting issues that compounded across hundreds of pages.
The right answer is keeping humans in the loop at the points where they add the most value. Not at every step. Not at none.
Next step