On September 7, 2026, Search Engine Roundtable reported a Bluesky exchange in which Google’s John Mueller told a site owner that “our systems have possibly lost faith in your site providing good value to users based on the old pages.” Search Engine Journal followed on September 8, noting that Mueller named no ranking system, no page-count threshold, and no recovery timeline.
Within two days the comment was flattened into “Google penalizes programmatic SEO” — not what he said, and not the part that should worry a Michigan catalog site. The part that should is the word site, plus the fact that the judgment stays anchored to pages you may have already deleted.
What did Google actually say about programmatic SEO in September 2026?
He was replying to one person about one site, not announcing policy. He described a setup iterating through “a large list of domain names, technologies, and attributes — finding all possible combinations,” and said that pattern “often leads to a site that’s either spam, borderline spam, or low quality.”
Two details get lost in the summaries. His objection was not to automation but to padding — he used “what does php mean?” as the example of filler bolted onto generated pages. And he qualified the whole thing with “possibly.” Google’s scaled content abuse policy defines the problem as creating many pages “for the primary purpose of manipulating search rankings and not helping users,” and applies it “no matter how it’s created”: purpose first, method second. Mueller’s reply restates that document rather than extending it.
Why doesn’t deleting the pages fix the problem immediately?
Because the assessment attaches to the site, and the site-level picture is rebuilt from whatever sample of your URLs Google has recently crawled — not from the ones you removed this morning.
Here is the mechanism. Page-level signals refresh when a page is recrawled. A site-level impression forms across many pages at once, from the crawl sample Google happens to hold. Crawl frequency is path-dependent: the URLs Google fetches most often this month are largely the ones it fetched most often last month. On a site that ballooned from 900 URLs to 60,000, that habit points the crawler at the generated set. So after cleanup, the sample informing the site-level view is still dominated by pages that no longer exist — and stays that way until crawl budget reallocates toward what survives.
That predicts something the coverage hasn’t addressed. Take two sites that each remove the same 40,000 thin pages. The one with 2,000 surviving URLs flips its crawl sample in weeks, because almost everything left is substantive. The one with 180,000 surviving URLs — most still thin — may see nothing for months, because the ratio barely moved. What governs recovery is not how many pages you deleted, but what share of your remaining crawlable surface is still filler.
Is every generated page now a liability?
No, and a site that reacts by deleting its catalog does more damage than the original problem. The useful distinction: does a page exist because a thing exists, or because a query string exists?
Michigan’s economy makes this concrete. A Warren aftermarket distributor publishing one URL per stocked part number — real price, live availability, torque specs, an OE cross-reference — has a catalog, generated or not. A Sterling Heights seller publishing a page for every year–make–model–part combination, where four in five return an empty result or the same three SKUs, has the pattern Mueller described. Same script. Only one corresponds to inventory.
The same split runs through metro Detroit service businesses: an HVAC company with staffed branches in Livonia, Royal Oak, and Rochester Hills has three location pages worth reading, and if it also runs 84 city pages generated from a radius list, it has 81 pages of filler wrapped around three real ones.
Should you trust the “Google keeps a site quality score” reading?
Not as stated. Mueller said “possibly,” described no stored rating, and gave no duration — Search Engine Journal was right to flag all three gaps. The version circulating on LinkedIn, in which Google keeps a persistent site-quality number that follows you around, is an invention built on a hedge.
The asymmetry still favors acting. Thinning a catalog costs a developer a sprint; being on the wrong side of a site-level read costs a quarter, and the pages you lose are not the ones you were willing to lose. A Livonia distributor with 60,000 generated fitment URLs loses nothing when the thin pages fall — those never converted. It loses the forty category and product pages carrying a $20,000-a-month quote pipeline, which fall with them.
How should a Michigan site audit its generated pages this month?
Seven steps, all executable by a three-person team inside two weeks.
- Count your crawlable surface honestly. Compare your indexable URL count against the number of distinct products, services, or locations you actually have. A ratio above roughly 5:1 is the signal to keep reading.
- Segment by template, not by folder. Group URLs by the script that generated them. One template usually accounts for the bulk of the bloat — and one template is usually the fix.
- Measure unique body content per template. Sample 30 URLs from each and compute what percentage of visible text differs between them. Below about 20% unique, the page is a variable substitution with a heading on it.
- Cross-reference 12 months of Search Console and analytics. Any template whose median URL produced zero clicks and zero assisted conversions in a year has told you what it is worth.
- Consolidate before you delete. Collapse empty year–make–model combinations into the parent fitment page with a faceted filter, and 301 the old URLs there. Consolidation keeps the demand and removes the page; deletion keeps neither.
- Add something only you have to what survives. Stock status, a scanned spec sheet, a cross-reference table, a photo of the part on your own shelf, install notes from counter staff. This is the “time & significant effort” Mueller was pointing at.
- Log the finish date and stop touching it. Re-measure at 30, 60, and 90 days. A site that changes template strategy every three weeks never lets the crawl sample turn over, which is the one thing recovery requires.
One caution on step four: Search Engine Roundtable reported on September 11 that the Search Console page indexing report is missing a large block of June 2026 data. Pull click and impression data, but do not diagnose an indexing collapse from that report until Google confirms it is clean. The rule from our search coverage applies: confirm a drop in two independent sources before acting.
What dates should be on the calendar before Q4?
- August 18, 2026 — the August 2026 spam update began rolling out. If your generated pages lost visibility in that window, the timing matters.
- September 7–8, 2026 — Mueller’s Bluesky reply and the Search Engine Journal analysis. Treat both as commentary on existing policy, not new policy.
- September 30, 2026 — close of Google’s September 1–30 automatic migration window for campaign-level broad match into AI Max, per Search Engine Land. A separate issue landing on the same team this month.
- Mid-October 2026 — the practical freeze date. Catalog changes made after it will not have recrawled into a stable state before peak.
- November 27–30, 2026 — Black Friday through Cyber Monday. Whatever your crawl sample looks like in mid-November is what you sell through.
Does this change how AI Overviews and ChatGPT cite you?
It pushes in the same direction. Google’s guidance on generative AI in Search treats publishing a separate page for every possible query phrasing — specifically to influence rankings or AI responses — as falling under the same scaled content abuse policy.
The economics reinforce it. A generative engine needs a source that resolves a question in one place. A page restating the query and offering three interchangeable SKUs gives it nothing to quote. A consolidated fitment page with a compatibility table, stock status, and an install caveat is a citable object. Thinning a catalog and building pages that answer something are one project seen from two angles — the same argument behind the question pages Michigan manufacturers and trades are publishing for AI Overviews.
Frequently asked questions
Does Google penalize programmatic SEO?
No blanket penalty exists. Google’s scaled content abuse policy targets pages created primarily to manipulate rankings rather than help users, regardless of how they were made. Mueller’s September 7, 2026 comment described one site’s combinatorial pattern. Automation is not the trigger; publishing pages that correspond to nothing a user needs is.
How long does recovery take after removing low-value pages?
Mueller gave no timeline, saying only that it “tends to take time & significant effort” and suggesting spam and core update recoveries as a rough comparison. Those commonly run several months. The governing variable is how quickly your crawl sample shifts toward substantive pages, which depends on what fraction of your surviving URLs are still thin.
Should I noindex generated pages or delete them?
Prefer consolidation with a 301 to the parent page, which preserves link equity and real demand. Reserve noindex for pages you must keep live for internal or user-facing reasons. Outright deletion with a 404 is right only for combinations that never corresponded to a product, service area, or question you can answer.
Do location pages for a multi-city service business count as programmatic SEO?
Only when they are interchangeable. A page for a Royal Oak branch with its own crew, licensing, permit history, and response times is a real page. Forty city pages differing by a place name and a map embed are the pattern Mueller flagged. The test: does a customer in that city learn anything the page next door doesn’t say?
The one thing to take away
Mueller announced no penalty on September 7. He described something more inconvenient: a judgment formed slowly, applied site-wide, lagging your cleanup by however long it takes Google to notice what you left standing. For a Michigan distributor or multi-location service brand, that turns technical debt into a scheduling problem — the work has to start before the quarter you want it to pay off in.
Comments are closed