Without proprietary data, AI-powered programmatic SEO tends to look like spam to Google

On March 5, 2024, Google tightened its policy against scaled content abuse, which now applies to pages produced by automation, by people or by a combination of the two. Afterward, the company reported 45% less low-quality, unoriginal content in search results.

On September 7, 2026, John Mueller replied on Bluesky to a site that was generating pages from every combination of domain names, technologies and attributes. In his view, this kind of programmatic SEO usually leads to a spammy, almost spammy or low-quality site, and pages already published can keep weighing on how Google evaluates the whole site.

Both episodes open an opinion piece published by Search Engine Journal on November 10, written by a practitioner who has been building programmatic SEO systems since 2016. The argument: generative AI removed the barrier to publishing at scale, and many brands started calling mass publishing programmatic SEO. Projects like these, the author says, are collapsing or getting hit with manual actions.

What separates one from the other

For the author, programmatic SEO is a publishing method with three parts: a repeating search pattern, a structured database that makes each page unique and useful, and a template that answers the query in every variation. His example is a currency converter, where each currency pair has an up-to-date exchange rate and becomes its own page. You publish only as many pages as the data can support.

In AI mass publishing, the template and the system are still there, but whatever makes each page different comes from a prompt. Swap the city name and you get a new page with nothing new on it. Google's spam policies don't even use the word programmatic. What they describe is scaled content abuse: many pages generated primarily to manipulate rankings, no matter how they are produced. Using generative AI to create lots of pages without adding value is listed as one of the examples.

The proposed test is simple: ask where every number on the page came from and when it was last updated. A cost-of-living page for Lisbon, with rent, transportation and grocery figures pulled from a maintained database, can answer that. One with numbers estimated by the model, with no source or date, and the same sentences as every other city, can't.

"At Wise, we focus on real customer utility: solving a problem users face on their journey to becoming customers. Programmatic SEO can then make that solution relevant and accessible across the many searches and variations that share the same problem. It should never be a shortcut to publishing large volumes of low-quality pages," says Lucia Lecesne, marketing lead at Wise.

The pre-launch checklist

None of the author's criteria ask whether AI wrote the page. Among the items on the checklist he runs before any launch:

• Have a unique data source and be able to name it.
• Define the minimum data completeness a row needs to generate a page, and enforce it in code.
• Publish a test batch of 25 to 50 pages first, not the full set.
• Read ten pages by hand, start to finish.
• Before launch, write the rule for pruning or merging pages that get no impressions, including in AI features.

In AI search

The author recommends measuring by template, not just at the site level. Search Console's generative AI performance report shows how often links to your pages appear in Google AI Overviews and Google AI Mode. His suggestion is to group those pages by template and compare them with clicks in the standard performance report.

For brand teams, the takeaway is to publish fewer pages, each with data nobody else has. An AI answer that needs a source for Lisbon rents can link to the page that holds those numbers, and has no reason to link to one that repeats what dozens of others already say.