SEO for Multi‑Location Brands
Templated local pages fail: the data underneath is the same everywhere. We build pages on data a competitor can’t copy.
Why programmatic SEO earned a bad name
The method spent years meaning one template, one paragraph, and a find and replace on the city name. Buyers learned to distrust it, and they were right to.
Programmatic SEO used to just mean Mad Libs. You have a template, you change San Antonio, Texas to Dallas, Texas, and the content underneath is the same. This was slop before slop, and it was just bad content on the web.
Google made the enforcement explicit in March 2024. Its scaled content abuse policy covers many pages generated primarily to manipulate rankings rather than to help people, and it names automation, human effort, and combinations of the two. Bulk content written by hand earns no exemption.
A template can never carry the differentiation on its own. It has to come from the data underneath the page, the design on top of it, and the people running those locations, who know things a competitor can’t look up. That last one is the hard part.
What happens without a data layer
Producing content costs almost nothing now, so volume is cheap, and we have watched the same curve finish the same way enough times to know how it ends.
You get this big push with AI, you climb the mountain, and you can get it out there quickly. It works for 60 days, and then there is the cliff. It is obviously spammy, it is not helpful, and you are just adding noise to the search results. Those sites get nuked. Happens again and again.
What those executions are missing is design and data engineering, and both get badly underestimated in this world. We would go further and say Google is starved for differentiated content right now, though that is our read rather than anything Google has said.
The cut in low-quality, unoriginal content Google projected its March 2024 update would make in search results.
Source: Google, March 2024 updateOf the top 20 results across 500 informational keywords carried AI-generated content by July 2025. That is the flood the March 2024 policy is aimed at.
Source: Originality.ai, ongoing studyConsumers using ChatGPT and other generative AI tools for local recommendations, up in one year.
Source: BrightLocal, 2026 Local Consumer Review SurveyA thousand cities of route pricing, refreshed monthly
One worked example of a build that holds.
We keep route pricing pages running for a national auto transport marketplace across roughly 1,000 U.S. cities, and the price on every page gets rewritten each month through code, off their live operational data. There was never any way somebody was going to manually update the prices of car shipping routes across the U.S. for 1,000 cities every month.
Cities carrying a route pricing page, refreshed monthly through code
Source: A2 engagement records · Read the Sherpa case study: roughly 9x organic traffic →One block for each kind of shipper
The page carries a block for every kind of person who ships a car on that route:
San Antonio → Dallas
Route overview · this month’s pricing
Route pricing, refreshed monthly from operational data
Snowbirds: shipping south for the seasonal window
Students: shipping to school on a term calendar
Reviews from people who shipped this route
Each group gets a heading of its own because each one searches the route differently, and the H2 matches how that group actually searches.
The content under each heading gets assembled out of structured parts, off the brand’s own operational data, instead of a template with the city swapped.
The reviews come from those same people, resolved for that market instead of a national average.
So when a student in Rochester asks an AI assistant how to ship a car to California, the page with a block built for that person is the one that matches.
What sits underneath a page
Six layers sit between an engineered portfolio and a wrapper on a prompt. Anything built to survive a core update needs all six.
Architecture
State pages, city pages, and several service types inside each market, plus landing pages outside the location tree. Running one template for all of it is what earned the method its reputation, so we build many.
Data layer
The brand’s own operational data gets ingested and structured, third-party inputs get sourced by zip, regulatory databases get maintained as the rules change, and reviews get resolved per market.
Human-led design
A model can only remix what already exists, so anything aspirational in the design stays human-led, and somebody here owns that judgment.
Content-block engines
Underneath every page there is a CMS, data components, and engines that build the content blocks, so a page gets assembled out of structured parts instead of a template with the city swapped.
Interactive components with schema
Calculators, toggles, and configurators, each carrying schema on the back end, so a machine can read the values as well as the person clicking through.
Changing every page at once, through code
When the market shifts, the headings, the data components, the pricing, and the calculator toggles all move across the portfolio through code. That is how one national brand moved roughly 300 pages to leasing language in about a week, and how a full redesign later went across every page just as fast.
We do leverage AI, and it is worth being precise about where. All of the design, all of the scaffolding, all of the expertise is built with a lot of human involvement, and then AI fills in the scaffolding.
That scaffolding is also why a model upgrade flows through a whole portfolio at once. If a new model comes out that is two or three times better, we can run 1,000 pages back through it and all of your content gets better. Inside A2 that system is Foundry, our programmatic page platform.
We’ll say the same about the edit loop, because you’d find it out in month two anyway. Sometimes you are going to get frustrated with the designs coming out of Forge, our content system. The system keeps the history of what you liked and what you edited, so every session starts closer.
Hand-built volume against an engineered engine
Building this in house is a real option and plenty of teams should, so here is the comparison we would walk you through, with its assumptions labeled.
One writer gets you roughly 20 pieces a month, or 30 to 40 refreshing work that already exists. So a hundred a month is about five writers, plus editorial, plus SEO, plus the web support to get it live. Speed, quality, and cost: you still pick two.
Estimate: Austin Shrum, A2 AnalyticsThe whole point of the stack is to break that triangle. On one engagement the route pricing stays current across a thousand cities every month, and the refresh is a code change rather than headcount.
Source: A2 auto transport engagement recordsIsn’t this replacing my team?
Our answer is the same every time. AI is not here to replace your team. It is here to create content that was cost prohibitive before, and nobody was going to hand-update 1,000 cities of route pricing.
Why not just feed the pages from an API?
You could. A live pricing API is a real engineering build, and it is also a public endpoint, so a competitor can hit it and read your pricing close to real time. Baking the data into the pages on a monthly refresh publishes the numbers without exposing the feed.
Either path gets judged the same way. Google set out in March 2024 to clear 40 percent of unoriginal content out of its results, and the pages left standing carry something unoriginal content cannot.
Four kinds of data a competitor can’t copy
Take this inventory to your own organization and go looking. Most brands are sitting on far more than they have ever published.
First-party, mastered
Years of operational records, the sales data, the route-completion counts per market, structured into a real data environment. It already exists inside the business and almost never reaches a page, because operational data and content live in different places. A brutal thing for a competitor to go up against.
Third-party, sourced
Environmental and demographic inputs, sourced by zip. On a national gutter protection brand that meant tree and foliage type, rainfall, soil porosity, basement propensity, and home square footage standing in for the size of the job.
Regulatory, living
For a national solar provider we maintain a database of incentives across federal, state, city, and utility levels, tagged to zip with sources cited, and tracked as the legislative schedules move.
Reviews, resolved per market
The system finds the most locally relevant review for each page, so it carries the voice of that market instead of a national average.
Where the proof stops: we went into the solar build hoping for breakaway non-brand success on the generic solar queries and we did not get it. Generic solar is a brutal market. We are still working with them, and we would tell you the same thing on a first call.
A template is copyable in an afternoon, and this layer takes years to accumulate and real engineering to publish. The fifth input is harder than any of them, and it is the owners running the locations. They already get asked to do plenty by the parent company, so we want to be extremely respectful of their time. That is what Alloy is for: a portal, voice to text, and photos tagged back to their profile. Get that part wrong and it reads as one more thing corporate wants.
Sources cited
- Google Search Central: March 2024 core update and new spam policies, including the scaled content abuse policy and the projected 40 percent reduction in low-quality, unoriginal content.
- Originality.ai: Amount of AI content in Google search results, ongoing study, at 19.56 percent of the top 20 results across 500 informational keywords, July 2025.
- BrightLocal: Local Consumer Review Survey 2026, where use of ChatGPT and other generative AI tools for local recommendations rose from 6 percent to 45 percent year over year.
- A2 engagement records: auto transport marketplace route pricing pages across roughly 1,000 U.S. cities, refreshed monthly through code.
- Estimates attributed to Austin Shrum, A2 Analytics: hand-built content production benchmarks.
Frequently asked questions
What is programmatic SEO for multi-location brands?
Generating location, service, and route pages at scale out of structured data, meaning operational, third-party, regulatory, and review data, rather than writing every page by hand.
Won’t templated pages get hit by a core update?
Thin, duplicated templates get hit, and they should. The pages that come through an algorithm change are built on data that answers a question somebody in that market asked.
What data do we need to get started?
More than you think, and most of it is already in the building: location records, service coverage, pricing, reviews, compliance data. A working session maps what you hold and what we would source.
How long until pages are live?
The data model is the slow part, and it usually takes the first few weeks. Generating the pages after that is days.
Start with the data you already have
Send us what you already hold, and we will map it onto a page architecture built to come through the next core update. One session settles what you have and which templates we would build first.