Programmatic SEO: When It Works, When It Fails

Somebody has shown you a case study where a company launched 40,000 pages from a spreadsheet and tripled their traffic. Now you want to know whether you can do the same thing.

Maybe. I have built these systems and I have also been hired to clean up after them, which is the less glamorous half of the work. The pattern is consistent enough that I can usually tell within one conversation which way a programmatic project is going to go.

Here is the honest version.

What Programmatic SEO Actually Is

Programmatic SEO is generating a large set of pages from a single template plus a structured data source. One template, one database, thousands of URLs.

You already meet it constantly. Every “flights from X to Y” page, every “best restaurants in [city]” listing, every “[software] vs [competitor]” comparison on a SaaS site. Nobody hand-wrote those. A template did, filled from a table.

The technique itself is neutral. It is not a loophole and it is not spam. Google has no rule against templated pages, and some of the most useful pages on the internet are generated exactly this way.

What decides the outcome is not the template. It is what goes into it.

The One Condition That Decides Everything

Every programmatic project I have seen succeed had genuinely different data on every page. Every one that failed had the same page with the city name swapped.

That is the whole test, and it is worth being brutal with yourself about it. Open two of your generated pages side by side. Cover the title. If a reader could not tell which page they are on, Google cannot either, and you have just built several thousand near-duplicates pointing at each other.

Ask it this way instead: what does someone learn on page 400 that they could not learn on page 399? If the answer is “the name of a different city”, stop building and go find better data.

Where the Keyword Pattern Comes From

Programmatic SEO only works on queries that arrive in a repeating shape. Three shapes carry most of it:

• Modifier plus entity. “cheap [thing] in [place]”, “[service] for [industry]”
• Entity versus entity. “[tool] vs [tool]”, “[material] vs [material]”
• Entity plus attribute. “[car] fuel economy”, “[city] cost of living”

The test for a real pattern is whether the tail actually exists. Take twenty combinations at random from the middle and the bottom of your list, not the top. If most of them show no meaningful search demand, you do not have a pattern with 5,000 pages in it. You have a pattern with 200 pages in it and 4,800 pages of liability.

Cutting the list is the single most valuable thing you can do at this stage, and almost nobody wants to do it, because the big number was the point.

Building It Without Creating a Liability

The mechanics are not complicated. The discipline is.

1. Get the data first, template second. If you design the template before you know what data you have, you will design fields you cannot fill, and you will fill them with padding. Padding is what gets the whole directory devalued.
2. Write three pages entirely by hand. Pick your best case, your worst case, and something average. If the worst case reads badly by hand, it will read worse at scale.
3. Set a data threshold, and enforce it. Decide the minimum a page needs to justify existing. If a row does not clear it, that page does not get built. Not built thin, not noindexed later. Not built.
4. Vary structure, not just values. Let sections appear and disappear based on what the data supports. A page with four data points should not have the same six headings as a page with forty.
5. Launch in waves. A few hundred pages, then wait. Watch what gets crawled, what gets indexed, and what gets ignored. Google will tell you what it thinks of the template before you have spent your entire crawl budget finding out.

That last point is where most of the damage gets avoided. Publishing 20,000 pages on a Tuesday is not a growth strategy, it is an irreversible bet placed before you had any feedback.

The Internal Linking Problem Nobody Plans For

A generated page with no internal links pointing at it is a page Google may never see, and will rarely value if it does. Sitemaps get URLs discovered. Links get them ranked.

So the link graph has to be generated too. Related items by shared attribute, parent hub pages per category, breadcrumbs that actually resolve. Not a random “you may also like” block, which distributes authority arbitrarily and teaches Google nothing about how your set is organised.

If you are working out how these pages should sit inside a wider structure, my write-up on topical authority covers the hub and cluster side of it in more depth.

What Changed Now That AI Reads These Pages

AI systems summarising your content have made the differentiation problem sharper, not softer.

A page that exists to catch a keyword has nothing quotable in it. A page carrying a specific number, a comparison, or a fact that is not on twenty other sites gives an AI system a reason to cite you by name. Generic templated prose gives it nothing to work with, so it pulls from whoever does have the specifics.

The upside is that programmatic pages built on real structured data are unusually well suited to this. Clean facts, consistent formatting, one clear answer per page. That is close to an ideal shape for retrieval. This is the same principle behind generative engine optimization, applied at scale.

When Not To Do This At All

Skip programmatic SEO entirely if any of these are true:

• Your data source is public and free, so your competitors have the identical pages already
• Your site has fewer than a few dozen pages and no established authority yet, in which case ten strong pages will outperform a thousand weak ones
• You cannot describe, in one sentence, what a visitor gets from an individual page
• You have no way to keep the data current, and stale generated pages age badly and visibly

That last one gets underestimated. A directory of 8,000 pages is a maintenance commitment, not a launch. Prices move, businesses close, numbers go out of date, and a page confidently stating something untrue is worse than no page at all.

The Short Version

Programmatic SEO is a distribution method, not a content strategy. It multiplies whatever you feed it. Feed it real, differentiated, maintained data and it is one of the highest-leverage things you can build. Feed it a thesaurus and a city list and it will multiply that just as efficiently.

Decide which one you have before you write the template.

Written by Shahzaib Ul Hassan, senior AI SEO consultant and founder of ShazzSEO. Ranking sites since 2009. 500+ websites optimized, 3,000+ students trained.