Check this page with an assistantOpens a chat asking it to summarise this article and name the evidence behind each claim.
Claude opens with the prompt on your clipboard: Anthropic does not support prefilled prompts on the web, and we would rather copy it than ship a button that drops it.
What Is Programmatic SEO, in Plain Terms?
Programmatic SEO is one page template plus one structured dataset, multiplied. Instead of writing pages one at a time, you build a single template and feed it rows of data: a page per city, per integration, per product pairing, or per spec. The template handles the structure; the dataset supplies what makes each page worth loading.
The canonical wins are worth naming because they set the standard. Zapier's integration pages, one per app pairing, each answer a real question: does tool A connect to tool B, and what can the connection do. NerdWallet's card comparison pages each carry rates, fees, and a verdict specific to that matchup. Neither company wrote those pages by hand, and neither shipped a page that exists only to exist.
The method itself is neutral. It is a manufacturing process, and like any manufacturing process it amplifies whatever you feed it. Feed it a verified dataset and you mass-produce useful answers. Feed it a thesaurus and a list of city names and you mass-produce spam. Outcomes divide entirely on whether each generated page carries real data a visitor came for, which is the question the rest of this playbook is built around.
Which Template Patterns Hold Up for a Small Site?
| Pattern | Example page | When it works | Minimum real data per page |
|---|---|---|---|
| Location pages | "[Service] in [City]" | You genuinely serve each area | Local proof: service radius, local pricing notes, area-specific details, at least one local reference |
| Integration or compatibility pages | "[Your product] + [Tool]" | Software and trades alike | Real setup steps and honest limitations |
| Comparison pages | "[Option A] vs [Option B]" | You honestly assess both options | A criteria table and a real recommendation |
| Glossary or spec pages | "What is [term]" | Technical niches with real terminology | A genuine definition plus usage context, not a dictionary clone |
| Calculator or price-estimate variants | "[Job type] cost calculator" | Strongest pattern per page | Working inputs and honest ranges |
Pick one pattern, not three, and pick the one backed by the strongest data you already own. A plumber's job history is a location dataset nobody can scrape. A SaaS founder's integration list is verified by definition. A contractor's price book is a calculator waiting for inputs. The pattern you can fill with proprietary rows will beat the pattern that merely looks bigger, because the dataset is the product and the template is just packaging.
How Many Pages Is Too Many? The Thin-Content Line
There is no magic number. Google's March 2024 scaled content abuse policy targets pages produced primarily for rankings rather than readers, at any count: two hundred thin pages violate it and twenty rich ones do not. Twenty rich pages will also outrank the two hundred, and carry no penalty risk while doing it.
The test we use is the survival test. Take any generated page and strip out the shared template text, the paragraphs that appear on every sibling page with only the variable swapped. If what remains would not stand alone as a useful answer, the page should not exist. A location page that survives the strip still has its service radius, its local pricing note, its named reference. A location page that does not survive is a city name wearing a template.
Thin scale carries a second cost that penalties get all the attention for: near-duplicate pages split ranking signals between themselves. When forty pages answer the same query with the same text, the engine has to guess which one deserves to rank, links and relevance spread across all forty, and the page that deserved to win gets a fraction of its earned strength. Cannibalization means thin scale does not just fail to help, it actively hurts your best pages. The same policy line governs auto-published blog posts, which is covered in our guide to automating publishing without wrecking quality.
A 30-Day Build Plan for One Pattern
Thirty days is enough to take one pattern from dataset to indexed batch, provided the sequence is respected: data first, template second, small batch third, and no scaling until Search Console confirms the first batch earned its place. The plan below front-loads the unglamorous work because that is where programmatic projects are actually won.
- Days 1 to 7: assemble and clean the dataset. This is the real work. A spreadsheet with 15 verified columns beats a scraped one with 50, because every unverified cell is a future error published at scale. Pull from your job history, your price book, your support tickets, whatever you can vouch for row by row.
- Days 8 to 14: build and test the template on three hand-checked pages. Each needs a unique title, a unique meta description, and an FAQ specific to that page. If the three test pages read as three different pages rather than one page in three costumes, the template is ready.
- Days 15 to 21: generate the first batch of 20. Submit the sitemap, then review every page by hand once. At batch size 20 a full manual pass costs an afternoon and catches the template bugs that would otherwise ship two hundred times.
- Days 22 to 30: watch Search Console indexing. Fix the patterns the crawler flags, pages excluded as duplicates or crawled and not indexed, and only then scale further.
The one rule that prevents most programmatic disasters: never publish batch two until batch one is indexed and earning impressions. Indexing is the engine's verdict on whether your template earned its place, and scaling before the verdict arrives means multiplying a mistake. If you are unsure which pattern your site can support at all, the Growth Map reads your Search Console data and shows which query clusters already have demand waiting, and if your site is missing foundations, start with the small business SEO checklist before generating anything.
The Data Density Test Before You Ship
Before any generated page goes live, ask three questions. Does it contain at least five facts unique to this page? Would a visitor landing here feel the page was made for their query? Does it link to and from related pages naturally? A page that fails any of the three goes back to the dataset, not to the sitemap.
A useful working threshold for the borderline cases: if less than roughly 40 percent of a page's visible text is unique to that page, merge it with its neighbors instead of shipping it. Two thin city pages become one strong regional page; three overlapping comparisons become one honest roundup. To be clear, neither the 40 percent figure nor the five-fact rule is a measured ranking factor, and nobody outside the search engines knows the real cutoffs. They are planning heuristics that force the right conversation before publishing rather than after deindexing.
Pages that pass this test earn a second dividend: they are exactly the kind of self-contained, fact-dense pages that AI assistants quote. Extractable template pages feed AI answers the same way they feed rankings, a mechanism we cover in how businesses get recommended by ChatGPT, and you can check individual pages with our free tools before committing a whole batch.
Questions People Ask About Programmatic SEO
- Does programmatic SEO still work in 2026?
Yes, when each generated page carries real data a visitor came for. The template sites that died after the March 2024 scaled content abuse policy were producing pages for crawlers, not readers. Pages built from a proprietary dataset, with unique facts and a working answer per URL, still rank, still earn AI citations, and still compound. The method survived; the shortcut version did not.
- How many pages do you need for programmatic SEO?
There is no minimum. The dataset decides the count: if you have 30 genuine city variants or 45 real integrations, that is your ceiling, and quality gates decide which of those pages survive indexing. Twenty pages that each pass a data density check outperform two hundred that share ninety percent of their text. Start with a batch of 20 and let indexing results set the pace.
- Is programmatic SEO against Google's guidelines?
The method is not. Google's scaled content abuse policy, updated in March 2024, targets pages produced primarily to manipulate rankings rather than to help readers, regardless of whether a human or a template produced them. A generated page with unique, useful data per URL is fine at any scale. A thousand near-identical pages wrapped around swapped city names is exactly what the policy describes.
- What is the difference between programmatic SEO and doorway pages?
Doorway pages exist to funnel visitors somewhere else and add nothing themselves; the page is a redirect with extra steps, and Google has penalized the pattern for years. A programmatic page done right is the destination: the visitor gets their price estimate, integration steps, or comparison verdict on that page, without needing to click anywhere else. The test is whether the page answers the query or merely intercepts it.
