Programmatic SEO is having its second hype cycle. The first one gave us the great directory boom. This one, powered by LLMs that can generate a "unique" page for every keyword permutation in minutes, is bigger and more dangerous. I have built programmatic sections that worked beautifully and I have audited domains that programmatic content quietly killed. The difference between those outcomes is not execution polish. It is a decision made before the first page was generated.
The One Question That Decides Everything
Before any programmatic project, I ask: does each generated page contain information that exists nowhere else and that someone specifically wants?
A page per city for a service that genuinely differs by city (pricing, availability, regulations, real local inventory) passes. A page per city that swaps the city name into identical paragraphs fails. A comparison page per product pair backed by real spec data passes. A "best X for Y" page where the LLM hallucinated the list fails.
If the honest answer is "the pages are mostly the same," you are not doing programmatic SEO. You are doing doorway pages with better tooling, and the outcome has been documented for twenty years.
Failure Mode One: The Quality Ratio Collapse
Search engines evaluate domains, not just pages. This is the part programmatic enthusiasts skip. When you publish 8,000 thin pages next to 80 excellent ones, you have not added 8,000 lottery tickets. You have changed what your domain is: it is now 99 percent thin content, and sitewide quality classifiers notice.
I have seen this play out on a domain that added a massive templated section to an otherwise strong site. The templated pages never ranked well, which was expected. What stung was the slow decline of the previously strong pages over the following months. Removing and pruning the section preceded a recovery, though it took the better part of a year. Correlation is not proof, but I have seen the pattern enough times to treat it as real. This is why content pruning is not optional hygiene for programmatic sites. It is the survival mechanism.
Failure Mode Two: Crawl Budget and Index Bloat
Generating pages is free. Getting them crawled, indexed, and kept in the index is not. Large programmatic sections routinely end up with the majority of URLs in "Discovered, currently not indexed" purgatory. Worse, the crawler time spent on your permutation pages is time not spent refreshing the pages that make you money.
My rules here:
- Launch in tranches, not all at once. Publish a few hundred pages, watch indexation rates for weeks, expand only if the index accepts them.
- Internally link only what deserves to exist. If a page is only reachable from an XML sitemap, you have already told Google what you think of it. Real site architecture with hub pages and meaningful internal links is what separates a programmatic library from a URL dump.
- Set removal criteria in advance. Decide before launch: pages with zero impressions after N months get consolidated or removed. Writing this rule after launch never happens, because by then the sunk cost has a vote.
- Ten randomly selected generated pages reviewed by a human against the question "would I be embarrassed if a prospect landed here?"
- A canonical and noindex strategy for near-duplicate permutations.
- Hub pages and internal linking planned as carefully as the templates themselves.
- Indexation, impression, and pruning thresholds written down with dates.
- A named owner for dataset freshness.
Failure Mode Three: LLM Content Without a Data Spine
The 2026-specific trap. Teams point an LLM at a keyword list and generate "unique" prose per page. The prose is unique the way snowflakes are unique: distinct at the character level, identical in information content.
The programmatic projects I have watched succeed all had what I call a data spine: a structured dataset that made each page factually different. Inventory, pricing, specs, locations, availability, real user activity. The LLM's job was presentation, not knowledge. When the LLM is the source of the facts, you also inherit hallucination risk at scale. One fabricated claim per hundred pages across ten thousand pages is a hundred published falsehoods with your brand on them.
There is also a trust dimension I care about more each year: as AI-generated content floods the web, provenance becomes a differentiator. Pages whose facts trace to a real dataset you own are exactly the kind of source machines and humans learn to prefer. I expand on this in first-party truth and data provenance.
When Programmatic SEO Is Genuinely the Right Call
I am not against the approach. I am against the approach without its prerequisites. Programmatic is right when:
1. You own or license a real dataset with meaningful variation across rows.
2. The query space has genuine long-tail demand you can verify, not just keyword-tool mirages. A disciplined keyword research process matters ten times more at programmatic scale because every mistake is multiplied by thousands.
3. Each page can answer its query better than a generalist page could. If one well-made guide would serve the entire query cluster better than your thousand fragments, build the guide.
4. You have the operational capacity to maintain it. Datasets go stale. A programmatic section is a product with an ongoing maintenance bill, not a campaign.
My Pre-Launch Checklist
Before any programmatic build ships under my watch:
If a team cannot get through that list, the project is not ready, and shipping it anyway means borrowing rankings from the rest of the domain at a terrible interest rate.
The Bottom Line
Programmatic SEO is leverage, and leverage amplifies whatever it touches. Applied to a real dataset serving real demand, it builds compounding assets that no manual content operation can match. Applied to keyword permutations and LLM filler, it converts a healthy domain into a liability faster than almost any other mistake in modern SEO. Decide which project you are running before the generator does it for you.