Thanks to generative AI, publishing thousands of pages in a few hours has never been easier, and many brands have tried programmatic SEO (pSEO). According to the author of an op-ed published by Search Engine Journal, who claims to have been building such systems since 2016, these aggressive programs are now collapsing or facing manual penalties because no one checks if their pages deserve to exist.
Google has already set the framework. On March 5, 2024, the company strengthened its large-scale content abuse policy, which applies whether pages are produced by automation, by humans, or by a combination of both. It then stated it had observed 45% fewer low-quality, unoriginal contents in its results.
On September 7, 2026, John Mueller responded on Bluesky to a site generating pages from all possible combinations of domain names, technologies, and attributes. According to him, this type of pSEO often leads to a spammy site, borderline spam, or low-quality site, and already published pages can continue to weigh on how Google judges the entire site.
For the author, pSEO is above all a publishing method: a repeatable search pattern, a structured dataset that makes each page unique, and a template that answers the query for each variant. The case of a currency converter illustrates this well. The pattern is “convert X to Y,” the data is the real-time exchange rate for each pair, and the template transforms each line into a page. Lucia Lecesne, marketing manager at Wise, which runs this type of program, insists on the real utility for the customer: “pSEO can make this solution relevant and accessible through the many queries that share the same problem, by adapting it to each customer’s search context. It should never be a shortcut to publishing large volumes of low-quality pages.”
Good and Bad pSEO
Viewed from the outside, the two practices look similar, which perpetuates confusion. Mass publishing retains the template and publishing system, but what differentiates each page comes from a simple instruction. Just changing a city’s name yields a ‘new’ page that adds nothing new. The author proposes a test: imagine two pages on the cost of living in Lisbon, built on the same template. The first displays rents, transport fares, and food prices from an up-to-date database. The second appears just as comprehensive, but its figures are an estimate from the language model, without source or date. To tell them apart, simply ask where each figure comes from and when it was updated.
The author clarifies that the use of AI is not the criterion. A page built on real data and drafted with AI’s help can perform well, while a page written manually without solid data has no more legitimacy to rank well. Eli Schwartz draws the same line: “Creating pages just because you can is not SEO, it’s programmatic spam.” Danny Sullivan, then Google’s Search Liaison, proposed a test in 2025: was the intent to “capture that traffic” rather than to answer a user who genuinely expected that content?
Measuring in the Age of AI Search
The context has also changed. According to SparkToro’s analysis based on Similarweb data, 68% of Google searches in the United States ended without a click during the first four months of 2026, compared to about 60% in 2024 and about 45% in 2016. SparkToro attributes this shift to AI Overviews, which would now appear in over 20% of searches. Ahrefs, for its part, estimates that these summaries have reduced clicks on the first result by 58%.
In this context, traffic is no longer enough to judge a page, as the click may never come. The author recommends building fewer pages, each containing data that no one else has, defining the query pattern before the template, and measuring performance by template rather than by site. He concludes with a simple question: before evolving anything, pick a page and ask yourself where every figure it contains comes from.
