The pitch is seductive. Take a dataset, build a template, multiply: "best [service] in [city]" for every city in India, or "[tool A] vs [tool B]" for every pair of tools, and wake up to fifty thousand pages. Some version of this works for a handful of well-known companies, which is why it keeps getting proposed. Most attempts I have seen end with a large number of thin pages, a crawl budget spent on nothing, and in some cases a sitewide quality problem that drags down pages that were doing fine.
The template does not create value. The data does.
A programmatic page is only as good as the thing it is built from. If the dataset has something a searcher genuinely wants, for example a verified list of clinics in each locality with hours and services, actual prices for a route or a product, a real comparison drawn from structured attributes, then the template is just the delivery mechanism for information that has value. If the dataset is a list of city names and a paragraph with the city swapped in, the template is a mail merge, and search engines have been good at recognising mail merges for a long time.
So the first question I ask about any programmatic idea is not "how many pages can we make?" It is "what does each page know that the searcher cannot get from the page ranking today?" If the answer is nothing, there is no project.
Programmatic SEO scales a good answer. It cannot manufacture one.
Where it goes wrong
The failures are consistent enough to list. There is no real query behind most of the pages, so they are generated for combinations nobody searches. The pages differ by a few tokens, so they compete with each other and with the hub. Internal linking is an afterthought, so there are thousands of pages reachable only through the sitemap. Nobody set a rule for which combinations deserve a page, so the long tail is full of empty or near-empty results. And no one planned for maintenance, so the data goes stale and the pages become wrong rather than just thin.
The last one is underrated. A page that confidently lists prices from two years ago is worse than no page, for the searcher and for the brand.
Doing it properly is slower than it looks
When it works, programmatic SEO is a product, not a content tactic. It needs a data source with real differentiation, kept fresh. It needs demand validation: evidence that the query pattern exists and that each page's variant has intent behind it. It needs thresholds for when a page should exist, be indexed, be canonicalised to a parent, or not be generated at all. It needs a template with genuinely unique, structured content per page, not padding. It needs internal linking built into the architecture, with hubs, related pages and a logical path from the home page. And it needs crawl and index monitoring, because the first signal that something is wrong is Google quietly declining to index most of what you made.
Done this way, the compounding is real. Every improvement to the template improves every page. Every addition to the dataset creates new pages that inherit all the work already done. That is what makes it worth doing, and it is also why it is not a shortcut: the leverage comes from the quality of the system, and the system takes serious work to build.
If you are weighing a programmatic build, start with the question of what each page knows, and be willing to hear the answer. The way I take an idea from that question through validation, templating and launch is described in my programmatic SEO framework.