Does programmatic SEO still work?

Why does programmatic SEO feel so much harder now?

Programmatic SEO is harder today because Google has become much less tolerant of disposable pages, while AI search is taking clicks away from some of the long-tail queries that used to make the model so attractive.

The first change came from Google itself. Its scaled content abuse policy now focuses on large numbers of pages created mainly to manipulate rankings while adding little value. Google deliberately made the production method secondary. Human-written pages, AI pages and automatically generated pages can all fall under the same policy when the end result is weak.

The second change is happening after the ranking. Pew Research looked at 68,879 Google searches and found that people clicked a normal search result on 15% of searches without an AI summary, versus 8% when an AI summary appeared. That is almost a halving of the click rate. Only 1% of searches with an AI summary produced a click on one of the sources inside the summary.

Ahrefs found something similar from a different dataset. Its analysis of 300,000 keywords found that the presence of an AI Overview correlated with a 58% lower average click-through rate for the top-ranking result compared with the expected rate without one.

Programmatic SEO has to clear two harder tests now. The page has to be worth indexing, then the underlying query has to be worth winning.

Did Google actually crack down on programmatic SEO?

Google still allows programmatic SEO, but its current rules leave far less room for pages whose only purpose is capturing another keyword variation.

Google’s scaled content abuse policy says the problem is generating many pages primarily to manipulate rankings while providing little or no value. Automation itself remains completely compatible with Search.

Google’s newer guidance for generative AI makes the distinction even clearer. It says AI can help with research and structuring original content, while mass-producing pages without extra value can violate the scaled content policy.

More recently, Google published specific guidance for search in the generative-AI era. One warning is particularly relevant to programmatic SEO: creating separate pages for every possible variation of how somebody might search can become a scaled-content problem. Google also says its systems have become better at understanding relevance even when the page does not contain the exact wording of the query.

That attacks one of the oldest reasons for building pSEO pages. A site no longer needs a separate URL simply because “CRM for architects,” “CRM software for architects” and “best architect CRM” are three different keyword strings.

Google still wants useful pages at scale. It has become much better at seeing through scale created mainly for keyword coverage.

Are programmatic SEO pages still ranking today?

Programmatic SEO pages are still ranking today across software, salaries, travel, property and other valuable searches.

A current spot-check gives us a useful picture of what survived. Search for Google Sheets and Slack integrations and dedicated pages from Zapier and Make still surface. Search for software engineer salaries in San Francisco and Levels.fyi has a page built around current compensation records. Tripadvisor currently exposes more than 4,000 restaurant results on its Phuket restaurant page. Large property sites produce searchable location pages from constantly changing listing databases.

These are all programmatic pages in the traditional sense. A template combines with structured information to create a very large number of URLs.

The interesting part is what the successful examples have in common. When the variable changes, something useful changes with it.

Switch the city on Levels.fyi and the salary distribution changes. Switch the destination on Tripadvisor and the restaurants, reviews and rankings change. Switch one app on Make and a different set of triggers and actions becomes available.

That makes today’s surviving pSEO look increasingly like searchable databases with landing pages attached.

Current page type	Example	What actually changes
Software integrations	Zapier, Make	Triggers, actions and workflows
Salary pages	Levels.fyi	Compensation data, employers and percentiles
Local discovery	Tripadvisor	Businesses, reviews, rankings and availability
Property search	Marketplace sites	Listings, prices, locations and attributes

Which programmatic SEO pages are getting crushed?

Programmatic SEO gets into trouble fastest when hundreds or thousands of URLs answer almost the same question with almost the same information.

The clearest failures have come from scaled publishing where the keyword changed much more than the underlying value.

Causal is an extreme example. The finance software company became famous for an “SEO heist” that used a competitor’s sitemap as the basis for roughly 1,800 AI-generated articles. Ahrefs later estimated that Causal’s organic traffic fell from about 307,000 to 1,485, a 99.5% collapse. The number of pages receiving organic traffic fell by more than 90%.

Causal sits at the aggressive end of scaled AI publishing, so we should avoid pretending every pSEO site faces the same outcome. Still, the example shows how quickly apparent scale can disappear when most of the value lives in the keyword targeting.

Google’s doorway-page rules create another problem for old-school pSEO. City pages become particularly exposed when dozens of locations ultimately provide the same answer or push the visitor toward the same destination.

The dangerous pattern is easy to recognize. The database exists because the SEO strategy needed thousands of URLs. Stronger projects usually work the other way around: thousands of meaningful records or combinations already exist, and search pages make them discoverable.

Why do database pages still work when templated articles struggle?

Database-driven programmatic SEO keeps working because the information itself changes from page to page, even when the design stays almost identical.

Levels.fyi is a clean example. Its software engineer page for San Francisco currently shows median compensation around the high-$200,000 range, percentile distributions, top-paying companies and recent salary submissions. Change the location, level or role and those numbers change.

Make does the same thing with software integrations. Its Google Sheets and Slack page currently exposes 73 modules across triggers, actions and searches. The combination itself determines what someone can automate.

Tripadvisor gets there through inventory and user contributions. Its Phuket restaurant page currently contains more than 4,000 results. Reviews, cuisine, price range, popularity and location create thousands of genuinely different underlying entities.

The template barely needs to change in any of these examples. Consistency can actually make the product easier to use.

So we would spend far less time trying to make every introduction sound unique. The harder and more useful question is whether the page still contains something meaningfully different after we remove the generated prose.

If the answer is yes, the template has a reason to exist. If almost everything distinctive disappears, rewriting the introduction twelve ways will do very little.

Can a small site still win with programmatic SEO?

Smaller sites can still make programmatic SEO work today, although the evidence is much messier than the famous examples from giant domains.

One recent healthcare case study from Uproer is useful here. The agency created thousands of location-and-specialty directory pages for a healthcare software company and reported a 349% increase in clicks alongside a 362% increase in impressions.

Those two numbers are more interesting together than separately. Impressions grew about 4.62 times while clicks grew about 4.49 times. In other words, almost all the growth came from becoming visible across far more relevant searches. The implied click-through rate barely changed.

That looks like genuine long-tail expansion.

Another recent example comes from Linkscope, a new marketplace domain that published roughly 8,000 pages for individual publisher websites. A Google Search Console screenshot in the case study showed 5,210 pages indexed, roughly 65% of the total. The case study also reports a large traffic spike based on Ahrefs estimates, which we would treat more cautiously than the Search Console indexation number.

Nakora published an even more commercially interesting SaaS example. It says a programmatic project generated 4,065 signups and 621 new customers during its first three months, producing $129,000 in revenue. That works out to roughly 15 customers for every 100 signups.

All three examples come from agencies or service providers presenting their own work, so their headline numbers deserve some skepticism. What they do establish is narrower and more useful: new programmatic page sets are still getting indexed, attracting search demand and producing customers.

Big domains have an easier starting position. They no longer have a monopoly on the tactic.

Recent example	Page model	Reported result	What we trust most
Uproer healthcare client	Location × specialty	+349% clicks	Click and impression trajectory
Linkscope	One page per marketplace entity	5,210 of ~8,000 pages indexed	Search Console indexation
Nakora SaaS client	Product capability pages	621 new customers	Commercial intent of the page set

Does publishing 10,000 programmatic pages at once still make sense?

Publishing 10,000 programmatic pages can still make sense, but page count has almost zero value by itself now.

Google’s current crawl documentation openly deals with websites containing millions of URLs. Its advanced crawl-budget guidance is aimed at sites above roughly one million moderately changing pages, sites with more than roughly 10,000 rapidly changing pages, and sites where many URLs remain “Discovered - currently not indexed.”

Large URL inventories clearly remain normal on the web.

The trap comes from confusing the number of possible combinations with the number of combinations Google should index.

Imagine a database with 100 cities and 100 professions. Technically, it can generate 10,000 city-profession pages. Perhaps only 1,400 combinations have enough data, enough search demand and a sufficiently different answer. The remaining 8,600 pages add crawl work while contributing little.

Linkscope gives us a real version of that problem. About 8,000 pages were published and 5,210 were shown as indexed. Even in a case study presented as a success, around one-third of the page set was outside the index at the measured point.

That is why arbitrary publishing rules such as “release only 50 pages per week” are unconvincing. We could publish 20,000 useful inventory pages and have a solid reason for every one. We could also publish 300 nearly empty pages and already have too many.

The data sets the useful ceiling; a magic publishing cadence doesn’t.

Is indexing now the real bottleneck for programmatic SEO?

Indexing has become one of the biggest bottlenecks in programmatic SEO because generating URLs is almost free while Google still chooses which URLs deserve repeated crawling and inclusion.

Google says the web exceeds its ability to explore and index every available URL. After a page gets crawled, Google still evaluates whether it belongs in the index.

Its newer faceted-navigation documentation is even more concrete. Filters can create almost infinite combinations, forcing crawlers to spend resources discovering URLs that eventually turn out to be useless. Google recommends preventing crawling when those filtered URLs have little search value.

That sounds technical, but it changes how we should read an indexing problem.

Suppose a site creates 20,000 pSEO pages and Google indexes 6,000. The instinctive reaction used to be: how do we force Google to index the other 14,000?

We would ask a harder question first: what does Google see in the 6,000 that is missing from the rest?

Maybe they have more inventory. Maybe they receive stronger internal links. Maybe the underlying query exists. Maybe hundreds of the excluded URLs are near-duplicates.

Indexation rate has become a useful diagnostic metric rather than a trophy. A declining rate can tell us that the page generator is expanding faster than the useful part of the database.

Are programmatic location pages still worth building?

Programmatic location pages still work when the location genuinely changes what a visitor can see, buy, compare or decide.

Travel and property searches make this obvious. Tripadvisor can create a Phuket restaurant page because Phuket has thousands of specific restaurants, reviews and rankings. A property marketplace can create a Phuket homes page because the available houses, neighborhoods and prices really are local.

Healthcare directories can work for the same reason. The Uproer project mentioned above combined specialties with locations because different places contained different providers. The 349% click growth came alongside almost identical proportional growth in impressions, which strongly suggests that the pages expanded the number of relevant searches the site could appear for.

A local service company faces a very different situation when it creates “accountant in Bristol,” “accountant in Bath,” “accountant in Cardiff” and 200 more pages while offering essentially the same remote service from one office.

Here is the test we would use: remove the city name and look at what remains.

A strong location page still has local inventory, local prices, local rules, local reviews, local availability or some other meaningful geographic difference. A weak one suddenly looks almost interchangeable with 100 siblings.

That simple check catches a surprising amount of bad local pSEO.

Do integration pages still work for programmatic SEO?

Integration pages remain one of the strongest programmatic SEO formats today because the search query often describes an actual product action.

Zapier currently has a dedicated Google Sheets and Slack page where someone can choose a Sheets trigger and a Slack action. Make exposes 73 modules for the same combination, including watching new rows, updating cells, finding messages and managing Slack actions.

That is unusually good alignment between SEO and product architecture.

Someone searching “Google Sheets Slack integration” has effectively described what the automation platform sells. The landing page can answer the search and move directly into the product.

Scale also comes naturally. If a platform supports thousands of apps, thousands or millions of useful combinations can exist before an SEO team writes a single paragraph.

This is probably the clearest modern pSEO model because deleting the SEO copy barely hurts the usefulness of the page. The triggers, actions, templates and product interface still answer the query.

When a page can survive that test, we are usually looking at something stronger than content arbitrage.

Are “alternatives” and comparison pages getting weaker?

Programmatic “alternatives” and comparison pages have become harder because people want an actual judgment, and public product data alone rarely gives them one.

A page for “[Software A] vs [Software B]” can easily pull feature lists, pricing fields and review scores from a database. The technical part is straightforward.

The difficult part begins when somebody asks which tool is better for a five-person startup, which one becomes expensive at 50 seats, or which product has become frustrating after a recent pricing change.

Those answers need context.

That is why review platforms such as G2 have a structural advantage: the database includes large amounts of user feedback. A software company comparing itself with a competitor can also have useful first-party knowledge, although readers will naturally discount some of its judgment.

Generic comparison farms have a weaker position. AI systems can already combine public pricing pages and feature lists remarkably well. A templated comparison page needs something beyond the same public facts to justify the click.

So comparison pSEO can still work, especially around commercial searches, but we would demand more from every page: proprietary data, current testing, large-scale reviews, product usage or a genuinely useful decision framework.

A table generated from two public pricing pages feels much less defensible these days.

Can AI write programmatic SEO pages without hurting the site?

AI can help produce programmatic SEO pages, but generated prose is a weak foundation for the page’s value.

Google’s current generative-AI guidance explicitly allows AI-assisted content creation when the result is accurate, relevant and useful. The same guidance warns that generating many pages without adding value can fall under scaled content abuse.

The strongest use for AI starts downstream of real information.

A salary site can use AI to explain an unusual compensation gap. A marketplace can summarize structured attributes. A financial database can turn a calculation into plain English. A large directory can use AI to classify records, catch missing fields or explain patterns hidden inside the data.

Starting with 15,000 keywords and asking an LLM to invent enough text for 15,000 pages gives us a much weaker setup.

Google’s latest AI-search guidance makes this especially relevant now. It explicitly warns against creating separate pages for every possible search variation simply to capture rankings or generative-AI responses.

AI has made the text layer incredibly cheap. That pushes the competitive advantage further toward the parts that remain expensive: proprietary data, product functionality, reviews, transactions, expert judgment and continuously updated inventory.

Is programmatic SEO still worth it for customer acquisition?

Programmatic SEO can still be excellent for customer acquisition when the search itself reveals what the visitor wants to buy or do.

The Nakora SaaS case is useful because it goes beyond traffic. Its client reportedly generated 4,065 signups and 621 customers from the program during the first three months, with $129,000 in attributed revenue.

Nakora says roughly 12.8% of visits became signups and around 2% became paying users. Those numbers are unusually strong, and they come from the agency behind the project, so we would avoid treating them as a market benchmark.

The underlying page idea is more convincing than the exact conversion rate. The company’s customer-success team kept receiving recurring questions about whether the product could do specific things. Those questions became high-intent landing pages.

Integration pages operate on the same logic. A person searching for a Slack and Google Sheets integration is already describing the workflow that Zapier or Make can sell.

That is where pSEO looks strongest economically today.

Ten thousand informational visits that disappear when Google answers the question itself may be worth surprisingly little. Five hundred visitors who are actively trying to solve a problem your product handles can be much more valuable.

The metric we care about has shifted from “How much traffic can this template generate?” toward “What percentage of these searches describe a potential customer?”

Are AI Overviews killing the business case for programmatic SEO?

AI Overviews are hurting some programmatic SEO models badly, especially pages that exist to answer simple informational questions, but they are much less threatening to pages people still need to use.

The click data is difficult to dismiss. Pew saw traditional-result clicking fall from 15% to 8% when an AI summary appeared. Ahrefs later estimated a 58% CTR reduction for the top-ranking result on searches showing an AI Overview.

Simple informational pSEO sits directly in that danger zone.

If someone asks a factual long-tail question and Google can produce a satisfactory answer in five sentences, ranking first has become less valuable. Creating another 20,000 variants of those questions becomes harder to justify.

Inventory behaves differently. Google can tell someone that Phuket has a large property market, but a buyer still needs to inspect individual homes. Google can explain how Slack and Google Sheets work together, but somebody may still need Make or Zapier to build the automation. An AI answer can summarize San Francisco software salaries while Levels.fyi still has the underlying distribution, employers and fresh submissions.

Google itself has now started giving site owners dedicated Search Console visibility data for its generative-AI features, and those reports have been rolled out worldwide. That is a fresh clue about where search is heading: websites are increasingly competing for visibility across classic results and AI-generated experiences at the same time.

For programmatic SEO, the best defense is having a page whose usefulness survives the summary.

Programmatic query type	Outlook now	Why
Live inventory	Strong	The user still needs to browse the underlying items
Integrations and tools	Strong	The page helps complete an action
Proprietary datasets	Strong	AI needs an underlying source
Local directories	Good when inventory changes	Geography creates real differences
Product comparisons	Mixed	Generic public facts are easy to summarize
Simple informational questions	Weakening fast	Search can often answer them directly

What should a programmatic SEO project look like now?

A good programmatic SEO project now starts with something genuinely useful that can be expressed across many search states, then limits indexable pages to the combinations that deserve their own URL.

We would start with the entities themselves: companies, properties, jobs, salaries, integrations, products, locations, routes, prices, reviews or whatever the business actually knows about.

Then we would ask where combining those entities changes the answer.

A software marketplace might have enough information for “[tool] alternatives” pages but very little for every possible “[tool A] vs [tool B]” combination. A property database may justify thousands of city and neighborhood pages while producing almost empty pages for individual streets. A salary database can support job × city pages only where enough submissions exist.

That decision should happen before page generation.

Once the pages are live, we would track them by template family. How many are indexed? How many receive impressions? How many receive clicks? How many generate customers? Which combinations repeatedly stay outside the index? Which pages have become stale because their underlying inventory disappeared?

Weak families should shrink. Strong families should expand.

This is where programmatic SEO has become much more interesting technically. The winning system behaves less like an article factory and more like an index that constantly decides which parts of a database are worth exposing to search.

So, does programmatic SEO still work?

Yes, programmatic SEO still works today, and we have enough current evidence to say that confidently. The easy version has deteriorated badly; the data-backed version remains one of the most scalable forms of SEO.

Google still documents how very large sites should manage millions of URLs. Current searches still surface deeply programmatic pages from Levels.fyi, Zapier, Make, Tripadvisor and major marketplaces. Recent smaller case studies show new page sets gaining substantial indexation, clicks, signups and customers.

At the same time, Google has become unusually explicit about the limit. Its newer AI-search guidance warns publishers against creating pages for every possible query variation, while scaled content rules target large amounts of low-value material regardless of whether humans, automation or AI created it.

AI search adds another filter. Pew’s observed click data and Ahrefs’ CTR analysis both show a large reduction in outbound clicking when Google supplies an AI answer. That makes shallow informational long-tail traffic less attractive than it was a few years ago.

If the strategy is “we found 20,000 keywords, so we should generate 20,000 pages,” we would expect a lot of those pages to struggle.

If the business already contains 20,000 useful combinations of inventory, data, functionality or user contributions, exposing those combinations through search can still work extremely well.

Page generation has never been easier. Finding thousands of pages that deserve to exist is now the hard part.