Is programmatic SEO dead?

No, programmatic SEO is not dead. What died is the version that treated a template as the product: thousands of pages where a city name or a product name gets swapped into the same paragraphs and nothing else changes. Google classifies that as scaled content abuse, and the March 2026 core update applied that judgment more consistently than before. Page sets built on a dataset that genuinely changes the answer from row to row still rank, still get cited by AI assistants, and are still one of the cheapest ways to build organic traffic at scale.

The confusion is understandable. If you read the postmortems from sites that lost traffic in 2026, almost all of them were programmatic. But so are Zapier, Wise, TripAdvisor and G2, which did not lose traffic. The method is not the variable. What sits behind the template is.

What Google actually says about scaled content

It is worth reading the policy rather than the commentary. Google's spam policies define scaled content abuse as when "many pages are generated for the primary purpose of manipulating search rankings and not helping users", and give as an example "using generative AI tools or other similar tools to generate many pages without adding value for users".

Two things follow from that wording. First, the method of generation is explicitly not the trigger. Google has been consistent that automation and AI are fine when the output helps someone. Second, the test is about purpose and value, which means it is a judgment about the pages themselves, not a count. There is no page threshold that flips a site from acceptable to spam, which is why the question people usually ask, how many pages is too many, has no answer.

The related policy people conflate with this one is site reputation abuse, sometimes called parasite SEO. That covers third-party content published on a trusted domain to borrow its ranking power, which is a different problem entirely. If you are generating pages on your own site from your own data, that policy does not apply to you.

What changed in 2026

Google ran a spam update on March 24 and 25, 2026, then started a broad core update on March 27 that finished rolling out on April 8. Google described the core update in its usual terms, as a regular update designed to better surface relevant, satisfying content from all types of sites, and did not publish a companion blog post singling out any technique.

That is worth emphasizing because a lot of the coverage framed it as an anti-programmatic update. It was not. The pattern in the aftermath was more mundane: sites whose pages had no per-page substance lost ground, and the loss was more visible in programmatic sets simply because those sets are large. A thin page is a thin page whether you published one of them or forty thousand. Publishing forty thousand just makes the problem impossible to miss.

Does programmatic SEO still work?

Yes, for page sets where each page carries information a searcher cannot get from the template alone. The clearest way to check is to open two pages from your set side by side and read them as a visitor who needed one of them. If the second page would have been useful too, your differentiation is real. If the second page reads as the first one with three nouns replaced, you have a duplication problem that no amount of rewriting the intro will fix.

The programs that keep working share a specific trait: the data came first and the pages came second. Zapier had the integration catalog before it had integration pages. Wise had live exchange rates. TripAdvisor had two decades of reviews. In each case the page exists because there is something to show. Compare that with a page set generated by permuting a keyword list, where the pages exist first and somebody then tries to find something to put on them. That ordering is the tell.

The four tests a page set has to pass

Before you commit engineering time to a template, run the set through these. They are in order of how often they are the thing that kills a project.

Does the axis have demand across its values? This is the failure mode nobody plans for. Teams pick an axis because the template can produce it, not because anyone searches it. Integration pages work because people really do search two product names together. Combining every service with every zip code usually does not, because the long tail of that grid has no searches in it at all. You need evidence of demand spread across the values, not just a big head term at the top.

Do the values share one intent? A single template serves a single job. If two thirds of your values want a comparison and the rest want a definition, one template will produce pages that satisfy neither group. Clustering the modifier set by intent before you build is the cheapest way to catch this, and it usually results in either two templates or a smaller, cleaner set. Our keyword clustering tool exists mostly for this shape of problem.

Is there a dataset behind each row? Prices, availability, specifications, regulations, measurements, first-party benchmarks, real inventory. Something structured, specific and ideally hard to copy. If your dataset lives across several systems, getting it into one place is a prerequisite, and it is often easier to connect the apps, APIs and databases you already run than to maintain another spreadsheet by hand. If no dataset exists and you cannot build one, the honest answer is that this is not a programmatic opportunity, it is a normal content project.

Has a human read a sample? Pull twenty random live pages and read them start to finish. Not the template preview, the live pages, including the ones where a data field was empty and the sentence now reads oddly. Edge cases are where generated sets embarrass themselves, and they are invisible until somebody looks.

How many pages should you publish?

Fewer than the template can produce, at least at first. Publish the 20 to 50 values with the clearest demand, get them indexed, and watch what happens in Search Console over four to eight weeks. You are looking for impressions across the set rather than on one or two pages, and for the pages to get indexed at all, which on a newer domain is not guaranteed. Expand once you have that evidence.

This staged approach also protects you from the worst outcome, which is not a penalty. It is spending three months of engineering time on a template that produces 8,000 pages nobody searches for. That failure is far more common than a manual action and nobody writes case studies about it.

Is programmatic SEO worth it for a small site?

It can be, with two caveats. A new domain will index a large page set slowly, so plan for months rather than weeks and lead with the strongest slice. And crawl budget is real once a set gets large, which means your internal linking has to actively route crawlers along the axis instead of relying on a sitemap alone.

The upside for a small site is competition. The head term in your category is probably owned by someone with a decade of links. The individual modifier queries frequently are not, because the incumbents built their sets years ago and never went back to fill the gaps. Difficulty tends to collapse as you move down an axis, and that gap is where a small site can win. Our long tail keyword tool page covers the underlying arithmetic.

Where AI assistants fit in

One development that cuts in favor of programmatic sets: assistants answer constraint-heavy questions well, and constraint-heavy questions are exactly what a modifier axis is made of. Someone asking an assistant which tool integrates with a specific stack, or what a specific fee is on a specific corridor, is asking a question that a well-built programmatic page answers precisely. Pages with clean tables, specific figures and a visible last updated date get extracted more reliably than prose. If that channel matters to you, our answer engine optimization page covers how to structure for it.

The same discipline applies though. An assistant that reads two of your pages and finds them interchangeable has no reason to cite either one specifically. Differentiation is the requirement in both channels.

The short version

Programmatic SEO is not dead, but the era where a template plus a keyword permutation list was enough is over, and it ended before 2026. What replaced it is unglamorous: do the research first, find an axis with demand spread across its values, make sure one intent runs through it, attach a dataset that changes the answer per page, publish a slice, and expand on evidence. That workflow is what our programmatic SEO keyword research page is built around, and you can run the first step on a seed term in the tool at the top of this article.