Google warns: programmatic SEO can sink your whole site

In the news : On 7 September 2026, Google Search Advocate John Mueller replied on Bluesky to a publisher puzzled by his site losing visibility: “Programmatic SEO like this often leads to a site that’s either spam, borderline spam, or low quality”, adding that “our systems have possibly lost faith in your site providing good value to users based on the old pages”. The loss of trust does not target the offending pages: it targets the entire site (Bluesky, 7 September 2026).

The Mueller sentence that matters is not the one about spam. It is the one about the site.

We have long known that valueless generated pages might simply fail to rank. What Mueller describes on 7 September is different in kind: a judgement passed on the entire domain, one that outlives the pages that triggered it. For any business that has deployed pages by city, by trade or by use case, the question is no longer “do these pages rank”, but “what did they cost the rest of the site”.

A quick definitional note, because the term covers two very different practices. Programmatic SEO means generating pages automatically from a structured source, a database or a catalogue, against a shared template. The method itself is neutral: a rail comparison site generating one page per route delivers genuine utility. What Mueller targets is the usage where the template replaces the content instead of carrying it.

What Mueller actually said

The exchange took place on Bluesky, in reply to a publisher describing a programmatic approach to building a site. Four sentences deserve to be read literally:

• “Programmatic SEO like this often leads to a site that’s either spam, borderline spam, or low quality.”
• “Our systems have possibly lost faith in your site providing good value to users based on the old pages.”
• “Resolving this tends to take time & significant effort to show the value.”
• “It’s easy to spin something up with many pages, it’s hard to provide real value to users.”

Two caveats matter, so we do not over-read this. Mueller says “often”, not “always”. And he says “possibly”, which keeps this an expert account of observed system behaviour rather than the publication of a ranking factor. This is not a product announcement, it is a clarification. Its worth lies in its precision, not in its status.

Why “the whole site” changes the maths

Most teams reason in reversible cost: ship a thousand pages, see what sticks, delete the rest. That reasoning assumes the penalty is local. Mueller’s statement contradicts it on two counts.

First: the signal attaches to the domain. If the systems have “lost faith in your site providing good value”, then your solid pages, the ones that predate the operation and have nothing to answer for, inherit the judgement. That is the difference between a fine and a loss of reputation.

Second: the exit is asymmetric. Mueller puts it plainly: resolving this “tends to take time & significant effort”. Pages can be deleted in an afternoon. The trust signal rebuilds over months. So you can undo the action without undoing the consequence, which is the worst possible shape for a budget decision.

This site-level trust logic echoes what Google formalises elsewhere, notably in its trust signals aimed at AI systems: the unit being assessed is no longer just the URL.

The framework already existed: “scaled content abuse”

Mueller is not inventing anything. Google’s web search spam policies already define scaled content abuse as generating many pages “for the primary purpose of manipulating search rankings and not helping users”. The documentation explicitly names using generative AI tools “to generate many pages without adding value for users”.

Google’s criterion is therefore intent and value, never the production method. Generating pages from a database is not an infringement. Producing three hundred interchangeable variants of the same text is. The whole distinction sits in what the page gives to someone who reads it.

Where the line sits, concretely

We build programmatic pages for our clients, so we may as well be precise about our own guardrails rather than commenting from a distance. Three internal rules, which are production choices we own, not Google thresholds:

1. At least half the content specific to the page, excluding the shared template. If the only thing that changes between two pages is a city name slotted into the same sentences, the page does not really exist.
2. Progressive deployment, in batches of twenty to fifty pages, with measurement between batches. Publishing a thousand pages in one day saves no time: it simply removes the chance to learn before everything is committed.
3. One real, sourced data point per page. A price, a lead time, a local constraint, a verifiable figure. Without it you write a page that discusses a topic instead of answering a question.

The fastest check remains the mask test: take two pages from the same template, hide the variable, and ask whether anything is left. It needs no tooling and settles the question in thirty seconds.

What to do this week

1. Measure your served-URL ratio. In Search Console, compare the number of generated pages against those getting at least one impression over ninety days. A template where the vast majority of URLs are never served is already producing the signal Mueller describes.
2. Do not mass-deindex in a panic. Removing hundreds of URLs at once does not erase the judgement passed on the domain, and you lose the few pages that were working. Sort before you cut.
3. Upgrade rather than delete, where you can. A weak page given its own data, a local example and a direct answer becomes a normal page. It is slower than deletion, and it demonstrates exactly the value Mueller says you need to demonstrate.
4. Log the deployment date of each batch. Without it you can never correlate a drop with a release, or with a spam update like the August 2026 one. Five minutes of logging saves six months of guesswork.

The limits of this information

This is a social media post, not a documentation update nor an announced algorithm change. Nothing indicates a rollout happened on 7 September: Mueller is describing existing system behaviour, in reply to one specific case whose domain, scale and history we do not know.

We also have no data to size the phenomenon: not the share of sites affected, not the average recovery time, not the page volume at which the signal tips. Anyone quoting you a precise number on those three points made it up. What this statement establishes solidly is the level at which the penalty applies. Not its intensity.

The Cicero take

Programmatic SEO is not dead, and it was never banned. What dies is the calculation that made it attractive: produce a lot for cheap, betting that failure would be consequence-free. That bet was already poor, and Mueller has now said out loud that it is settled against the entire domain. The real constraint on content at scale is not production capacity, it is having something worth saying a thousand times.

Method note and disclosure. John Mueller’s quotes are taken from his public Bluesky post of 7 September 2026 and from Search Engine Roundtable’s coverage the same day. The definition of scaled content abuse comes directly from Google Search Central’s spam documentation. The three production rules cited in this article are internal Cicéro choices, in no way thresholds validated by Google. Cicéro has no commercial relationship with Google and no link in this article is sponsored or affiliated.

Frequently asked questions

Alexis Dollé

CEO & Founder

Growth and SEO & GEO content strategist, I founded Cicéro to help businesses build lasting organic visibility : on Google and in AI-generated answers alike. Every piece of content we produce is designed to convert, not just to exist.

LinkedIn