Crawled, Not Indexed: Google’s AI-content quality tell

A mass of “crawled - currently not indexed” URLs in Google Search Console is one of the most misread signals in technical SEO — and on July 16, 2026, Google put its clearest framing yet on the record. On the Search Off the Record podcast, John Mueller confirmed that when Google’s systems have serious doubts about a site’s overall quality, they crawl less and index less — and he named undifferentiated AI-generated content as a live example of what triggers those doubts.

The stakes are practical. Site owners routinely treat the Page Indexing report as an error list — resubmitting URLs, hunting for a technical bug, filing the pattern under “Google being slow.” Mueller’s message points the other way: sometimes there is no technical issue to fix, because the report is functioning as a quality verdict on the site as a whole. For teams publishing at volume with AI in the loop, that changes what the report is actually telling you.

This playbook covers what Mueller and Splitt actually said (with verbatim quotes), the mechanical difference between the two not-indexed statuses, why Google’s written documentation and its spoken commentary are two different tiers of guidance, and a four-cause triage tree that puts quality where it belongs in the diagnosis: last, after the cheaper technical causes are ruled out.

The episode — titled “How to read the indexing report” — is a conversation between Martin Splitt and John Mueller of Google Search Relations about what the Search Console Page Indexing report does and does not mean. Splitt set up the central question by distinguishing the two statuses and then asking directly about “crawled - currently not indexed”: “Would you say that that is often or only sometimes a sign of a quality issue?”

Mueller’s answer opens with the word that should anchor every diagnosis built on this episode: “Sometimes.” He then explains the mechanism — and it is a site-level mechanism, not a page-level one.

Two details in that answer matter for diagnosis. First, the unit of judgment is the website, not the page — when quality concerns are serious, Google spends less effort on the whole site, which is why the statuses show up in bulk rather than on one stray URL. Second, the behavior spans both crawling and indexing, which is why quality doubts can surface as either status: “discovered not indexed” when Google declines to fetch, “crawled not indexed” when it fetches and then declines to index.

Just as important is what Mueller said this is not. He explicitly rejected the technical-bug framing: “It's not so much that I would say you should take these situations and try to fix them. Like from a technical point of view. It's not that you need to fix this technical issue that Google is not indexing this page at the moment.” His recommended response instead: “you almost need to take a step back and think about the quality overall” — and to do it “trying to look at it with the eyes of someone who is not directly involved with your website.”

The two statuses get conflated constantly, but they describe decisions made at different points in Google’s pipeline. Google’s own Page Indexing report documentation defines them mechanically — and the mechanical difference is where diagnosis starts.

The practical reading: “discovered - currently not indexed” is a statement about crawl priority — Google decided your URL wasn’t worth fetching yet. “Crawled - currently not indexed” is a statement about evaluation — Google fetched the page, looked at it, and put it back. That’s why Mueller’s quality commentary lands hardest on the second status: it is a post-evaluation decision. But note his own phrasing — serious site-level quality doubts reduce both crawling and indexing, so at scale the two statuses often climb together, and the trend across both is more diagnostic than either count alone.

One more detail from the official definition that most site owners miss: for “crawled - currently not indexed,” Google’s documentation says there is “no need to resubmit this URL for crawling.” Resubmission loops are the single most common response to this status, and the written guidance rules them out before the podcast ever gets to quality.

Here is the distinction most coverage of this story flattens: Google’s guidance on these statuses exists at two different levels, and they say different things. The written Search Central documentation defines the statuses mechanically — what happened in the pipeline — and says nothing about content quality or AI-generated content as a cause of either status. The causal, diagnostic explanation — quality doubts reduce crawl and index investment, undifferentiated AI content can trigger those doubts — comes from Mueller’s spoken commentary on the podcast, not from the help-center reference.

That two-tier structure isn’t a contradiction; it’s a division of labor. The written doc is the stable, evergreen contract: this is what the status means. The podcast is the diagnostic color: this is why it often happens and what to do about it. If you only read the docs, you’ll under-diagnose quality. If you only hear the podcast, you’ll over-diagnose it — because “sometimes” is doing real work in Mueller’s answer, and the mechanical causes in the docs still explain a large share of cases.

The same episode reinforces this by covering a second, entirely technical cause of indexing shortfalls: hosting and CDN bot-protection layers that return HTTP 200 on blocking or deceptive error pages, hiding real fetch failures from log-based diagnosis. Google’s own framing puts a technical cause and a quality cause side by side in one episode — which is exactly how your triage should work too. While you’re in Search Console, it’s worth pairing this diagnosis with the link report workflow — the two reports together give you both sides of the crawl-and-authority picture.

Quality is the least fixable and fastest-to-misdiagnose bucket — so it should be the last conclusion you reach, not the first. The triage tree below sequences the four candidate causes of a “crawled/discovered - not indexed” pattern in the order you should rule them out: the three technical causes first (each has a cheap, falsifiable check), then quality. The technical taxonomy draws on long-standing practitioner work — Search Engine Land’s Dan Taylor laid out the technical-vs-quality split back in May 2025 — while the quality row now carries Google’s own on-record language from July 16, 2026.

Two of the technical rows deserve a note. Canonicalization is the cause that most convincingly mimics a quality problem — the page sits “crawled - not indexed” because Google consolidated its signals to a different URL, for reasons that have nothing to do with content quality. We covered how that consolidation behaves, and why canonical fixes take time to stick, in a separate guide — rule it out before reading anything into the status. And crawl budget, per Google’s own large-site guidance, only becomes a primary concern around roughly one million+ URLs or very high daily publish volumes — which means for the large majority of business sites, a mass not-indexed pattern is far more likely to be architecture or quality than a budget ceiling.

When the technical rows are ruled out and quality is what remains, Mueller’s commentary gives you the most concrete self-audit heuristic Google has offered in this context. His example is a site built mostly from AI-generated content that “worked for a while” — until readers started reacting to it. In his words, people look at the site and think: “Well, I can tell this is AI generated. There's nothing unique or valuable that is available here for me.” And he immediately bounds the claim: “That's not to say that all AI-generated content is bad.”

The variable is differentiation, not authorship method. Google’s systems aren’t described here as detecting AI text; they’re described as responding to a site whose content gives readers — and by extension quality systems — nothing they can’t get anywhere else. Mueller’s closing formulation is the one worth pinning above your content pipeline:

As a self-audit, that sentence converts directly into a page-level question: for each template or content cluster showing the not-indexed pattern, could a competitor have published this page without your data, your client experience, or your point of view? If the honest answer is yes across most of the cluster, you’ve found the quality bucket. Mueller’s instruction to review the site “with the eyes of someone who is not directly involved with your website” is the operational version — and if your archive mixes human and AI drafting at volume, an audit pass over which pages actually read as AI-generated is a practical way to shortlist where that outside-eyes review should start.

Note also what this framing is not: a penalty. Mueller describes reduced crawl and index investment flowing from quality doubts — a resource-allocation decision, not a punitive action with a reconsideration path. That distinction matters for remediation: there is no “request a review” lever here. The lever is making the site worth Google’s investment again, and the indexing report is how you watch that re-evaluation happen over time.

Once the diagnosis lands in the quality bucket, the remediation is editorial, not technical — and it operates on clusters, not URLs. These are the four moves we run, mapped to the specific tell Mueller described rather than generic “improve E-E-A-T” advice.

The sequencing matters as much as the moves. Resubmission loops are explicitly ruled out by Google’s own documentation; mass rewriting before the probe step wastes the effort on pages that may have been stuck for technical reasons; and pruning before consolidating throws away pages that could have strengthened a merged asset. This is the operating model behind our content engine work — AI-assisted production with human value-add as a designed stage, not an afterthought — and it’s the remediation half of the agentic SEO engagements we run when a site arrives with an indexing report that reads like a verdict.

Read as a trend, the July 16 episode is less a policy change than a tell about where Google’s communication is heading. Search Engine Roundtable notes it has covered the quality-to-indexing link many times before — the position is long-standing. What changed is the example: in 2026, when Google Search Relations reaches for an illustration of “a site our systems stop trusting,” the illustration is an AI-generated site that worked for a while and then stopped earning reader trust. That choice of example, on the record, in the indexing-report context, is the signal — the Page Indexing report is being positioned as the surface where AI-content quality problems become visible first, before rankings drama and before any named update.

Projecting forward, the operational implication for publishing teams is that indexation rate becomes a leading quality KPI. A site can hold rankings on its established pages while its newest AI-assisted clusters quietly stop entering the index — and the trend line of indexed-to-known pages will show that months before traffic does. Teams publishing at volume should watch that ratio per template and per content cluster, treat a sustained decline as an editorial alarm rather than a technical ticket, and keep the triage order honest so real technical regressions — a canonicalization slip, a bot-protection layer returning false 200s, an internal-linking gap after a redesign — don’t get misfiled as quality. If you manage multiple properties or newer property types, the same trend-reading applies across Search Console’s newer property structures too.

The honest caveat cuts both ways: because Mueller quantified nothing, nobody can tell you how much indexation improvement a quality pass will buy, or how fast. Google’s documentation says a crawled-not-indexed page “may or may not be indexed in the future.” What the episode does establish is the direction of the lever — sites that give readers something distinct get more crawl and index investment, and sites that read like anyone could have written them get less.