Programmatic SEO Came Back. This Time It Has to Earn It.

Programmatic SEO was a content-farm strategy wearing a database costume. The agent version can be something else, but only if each page is generated from data the site actually owns.

The pattern is old: one template, N rows of structured data, N landing pages. It worked from roughly 2015 to 2021 for comparison pages, integration directories, and location pages. Then it stopped working, because the pages stopped being about anything. Google's spam and helpful-content systems got good at identifying a template with slot-fillers, and whole programmatic footprints got deindexed in single updates.

The tooling never went away. What changed is that agents can now run the whole pipeline, and in our directory the pipeline is well represented: page-play-builder from aaron-he-zhu scores 8.9, with seo-keyword-cluster and seo-content-brief from seranking sitting in the 8.6–9.0 band alongside the rest of that family — seo-plan, seo-content-audit, seo-competitor-gap-analysis.

The step everyone skips

The failure mode of programmatic SEO is not the template. It is publishing 400 pages for what should be 40.

This is where clustering earns its place, and it comes before any page exists. seo-keyword-cluster groups queries by intent overlap, and the output you actually want is not "here are your keywords" but "here is how many distinct pages these keywords require." If 900 long-tail queries collapse into 40 clusters, the correct programmatic build is 40 pages with strong internal linking — not 900 near-duplicates. Most content-farm verdicts were, at root, cluster counts that were ignored.

Once the cluster count is settled, seo-content-brief turns each cluster into a spec: entities to cover, heading structure, the questions the page must answer, internal link targets. This is the contract the generated page has to fulfill. Without a brief, "generate a page for this keyword" degrades to filler with the keyword in the H1.

The build step

page-play-builder takes the spec-plus-data pattern and produces the batch. Our review scored it 8.9, and the score comes with the same caveat the tool itself carries: it is powerful and risky in equal measure. It is the fastest way in the directory to produce hundreds of decent pages, and also the fastest way to produce hundreds of thin ones. The skill does not — cannot — know which one it is doing. That is decided entirely by what you feed it.

The agent version's actual advantage

It is tempting to say the advantage is generation speed. It isn't. Templates with a data source were already fast; a mail-merge from 2018 could produce 500 pages.

The real advantage is that an agent can do per-row work that templates never could:

• Pull the real data for row N. Not "insert city name here," but the actual pricing, the actual API endpoints, the actual compatibility matrix for that row's entity.
• Write internal links at build time. Which of the other 39 pages this page should reference is a per-row decision, and it is what makes the footprint navigable instead of orphaned.
• Flag rows that have no data. If row 217 has nothing real to say, the correct output is "skip row 217," and an agent can be instructed to refuse rather than pad.

That last one is the whole ballgame. The old programmatic SEO died because empty rows got published anyway. An agent that is told "if you cannot name the data source for this row, do not write this page" structurally cannot produce the thing that got sites penalized.

The discipline that keeps it alive

Before you run the batch, answer one question per row: what does this page know that a competitor filling the same template does not?

If the answer is "nothing, but we'll have the page," you are running the 2019 version with 2026 tooling, and the outcome will be the 2019 outcome. If the answer is a data table, a first-party measurement, or an entity relationship you own, the page has a reason to exist and a fighting chance of being the best result for its cluster.

The stack is good enough now. seo-keyword-cluster to get the page count right, seo-content-brief to make each page mean something, page-play-builder to build. What no skill in the directory — or anywhere — can supply is the dataset. If you don't have one, you don't have a programmatic SEO project; you have a content farm with better tooling.