ONE TEMPLATE, HUNDREDS OF PAGES, ONE SITE-WIDE JUDGMENT (ILLUSTRATIVE)
WHAT YOU GENERATED
“[Service] in [City]” × 400 cities
“[Tool] vs [Tool]” × 250 pairs
WHAT GOOGLE MAY CONCLUDE
Most of these pages add little.
So the whole site gets less benefit of the doubt.
Decorative illustration only. Page counts are invented for the picture, not real data.

Programmatic SEO means using a template and a data set to publish pages in bulk, such as one page per city, per product pairing or per integration. Google’s John Mueller said on September 7, 2026 that this approach often ends up as spam, borderline spam or low quality, and that Google’s systems can lose faith in a site based on its older pages. If your team spun up a few hundred of those pages in the last couple of years, it’s worth a hard look.

This isn’t a panic post. Mueller was answering one site owner on Bluesky, not announcing a penalty. But the comment lines up with what Google has put in writing for years, and the same trust problem can follow you into AI answers. Below we separate what he actually said from what the headlines added, show you how to sort the pages worth keeping from the ones that aren’t, and lay out a cleanup order. It’s one piece of the larger SEO work we do, and it builds on our guide to auditing and pruning content for AI search.

45%
Reduction in low-quality, unoriginal results Google reported after its March 2024 update
Google, updated April 26, 2024
Sep 7
Date Mueller said programmatic SEO often leads to spam, borderline spam or low quality
Search Engine Roundtable, 2026
0
Recovery timelines Mueller gave. He said only that it takes time and significant effort
Search Engine Roundtable, Sept 7, 2026

What did Google actually say about programmatic SEO?

Mueller said programmatic SEO “often leads to a site that’s either spam, borderline spam, or low quality,” and that Google’s systems “possibly lost faith” in the site’s value to users based on its old pages. That’s the core of it, and the wording is hedged on purpose.

Per Search Engine Roundtable (Barry Schwartz, September 7, 2026), he was replying to a site owner whose project generated pages from big lists of domain names, technologies and attributes. He said code generators make it easy to produce lots of pages that each have “some” value, and that stacking on more generated text doesn’t make the site more useful. His line on the economics: it’s easy to spin up many pages and hard to give users real value.

Here’s what he did not say. He said it was not to say the site in question was spam. He didn’t name a ranking system, a penalty or a manual action. And he gave no timeline for recovery, only that fixing it tends to take time and significant effort. The phrase “many months” you may have seen in coverage is the reporter’s reading, not Mueller’s wording. TechWyse (September 9, 2026) reported the same limits.

Is programmatic SEO against Google’s rules?

Not by itself. Templates and data-driven pages are fine. What Google’s spam policies prohibit is scaled content abuse, which means producing pages in bulk mainly to rank, without adding value for people.

Google’s Search Central spam policies list the patterns: using generative AI tools to generate many pages without adding value for users, scraping feeds or search results to build pages, stitching together content from other pages without adding anything, and pages that make little sense to a reader but are full of keywords. The policy says it applies however the content gets made. In its March 2024 announcement, Google said it was strengthening the policy against content produced at scale to manipulate rankings, whether by automation, humans or a mix of both.

So the test isn’t “did a script build this?” It’s “did a person get something from this page that they couldn’t get elsewhere?”

Google’s March 2024 low-quality content target vs. reported result (percent reduction)
Expected by Google (March 2024)

40%

Reported after rollout (April 26, 2024 update)

45%

Source: Google Search blog, “New ways we’re tackling spammy, low-quality content on Search,” March 2024, with Google’s April 26, 2024 update (rollout completed April 19). Measures the reduction in low-quality, unoriginal content in results, not programmatic pages specifically.

Can a pile of thin pages hurt the rest of my site?

That’s what Mueller’s comment points to, though he used the word “possibly.” Google has documented that it uses site-wide signals alongside page-level ones. It has not said that any single set of programmatic pages triggers a site-wide drop.

Google’s own documentation says site-wide signals and classifiers contribute to how it understands pages, and that having some good site-wide signals does not mean all content from a site will always rank well. The helpful content system, which used to be a separate thing, was folded into Google’s core ranking systems in March 2024, per the Search Central page. We can’t tell you from public information how much weight a block of thin pages carries against the rest of a site. Nobody outside Google can. What we can say is that treating those pages as harmless clutter is a gamble.

For a business, the practical risk is this: the money pages, the ones that sell, share a domain with the template pages. If trust is judged at the domain level, the weak ones can cost you on the strong ones.

How do I tell a useful templated page from a thin one?

Ask whether the page would still earn its place if the template were invisible. A useful templated page has information specific to that page that a reader can act on. A thin one swaps a noun into the same paragraphs.

Google’s helpful content guidance gives you the questions to ask. It warns about content made mainly to draw search visits, about covering many topics hoping some will rank, about using automation across lots of topics, and about leaving readers to search again for something better. We turned that into a five-part check you can run on a sample of 20 pages:

  1. Swap test. Replace the city or product name with another. If the page reads the same, it’s a template with a label.
  2. Unique data. Does the page carry something only you have, like pricing, availability, local results, or real customer outcomes?
  3. Next step. Can a visitor do something useful here, or do they have to go back to the results and try again?
  4. Traffic and links. Over the last 6 to 12 months, did the page get visits, conversions or links from anywhere? Zero on all three is a flag, not proof.
  5. Overlap. Are 30 near-identical pages competing for the same intent?

These are our working criteria, not Google’s official scoring. Use them to sort, and use judgment on the edge cases.

What should I do with the pages that fail?

Improve what can be improved, merge what overlaps, and remove what can’t be saved. Don’t delete hundreds of URLs in one afternoon and hope for the best.

  1. Export and sort. Pull every templated URL with its traffic, conversions, links and indexed status. Group by template.
  2. Keep and strengthen. Pages with real demand get real content: local detail, original data, an expert voice, a clear next step.
  3. Merge. Combine near-duplicates into one stronger page and 301 redirect the old URLs to it.
  4. Remove. For pages with no traffic, no links and no purpose, remove them or redirect to the closest useful page. Keep a backup before you do it.
  5. Stop the faucet. Put a quality gate in front of the generator so new pages need a human review and a reason to exist.
  6. Watch, then wait. Track indexing and rankings for the whole site, not just the pruned section. Mueller was clear that it takes time.

If you’ve also used AI to write bylines or author bios across those pages, deal with that at the same time. Google’s guidance calls fake creator profiles deceptive and a sign of low quality, and we cover it in our look at fake author profiles and what Google does about them.

Does this matter for AI search citations too?

Probably, but we can’t cite a study that proves the link. AI answer features lean on Google’s index and ranking, so a site that Google trusts less is likely to have fewer chances to show up. That’s our inference, not a published finding.

What we can say with more confidence is that a bulk-page strategy isn’t a good fit for how people use AI search now. Answer engines favor pages with a clear, specific answer and a reason to trust the source. One strong page on a topic beats forty clones. If you want the full picture on what makes a page quotable, our guide to writing citable content for AI is a good next read.

One thing to do this week: pick your biggest templated section, open ten pages at random, and run the swap test. If you can’t tell them apart once the name is hidden, you already know what the audit will find.

Key takeaway

Mueller didn’t announce a penalty, and he gave no recovery timeline. But he did say programmatic pages often end up as spam or low quality and can cost a whole site Google’s trust. Keep the templated pages that give people something real, merge or remove the rest in stages, and put a quality gate on whatever generates new ones.

Built a lot of templated pages and not sure what they’re doing to your site?

We’ll sample your template sections, sort what to keep, merge or retire, and show you what it means for search and AI visibility.

Claim my free audit →

Sources

  • Barry Schwartz, “Google Can Lose Faith In Sites Based On Low Value Programmatic SEO,” Search Engine Roundtable, September 7, 2026 (quoting John Mueller on Bluesky).
  • TechWyse, “Google’s John Mueller Warns Programmatic SEO Sites Can Lose Google’s Trust,” September 9, 2026.
  • Google Search Central, Spam policies for Google web search (scaled content abuse section), current documentation.
  • Google Search blog, “New ways we’re tackling spammy, low-quality content on Search,” March 2024, updated April 26, 2024 (45% reduction reported; rollout completed April 19, 2024).
  • Google Search Central, helpful content system documentation (retired as a standalone system; part of core ranking systems since March 2024).
  • Google Search Central, “Creating helpful, reliable, people-first content,” current documentation.
  • The section questions above are common query patterns for this topic; we did not verify search volume for each one. The five-part page check is C3’s own method, not a Google standard.