National Today's /us/ directory is a useful losing case because the same content block appears to have lost visibility across classic Google search, Google AI features, and most of the ChatGPT prompts Glenn Gabe tested after a scaled-content manual action.

TL;DR

Glenn Gabe documented a sharp collapse in visibility for National Today's AI-generated local-news directory after Google applied what he identified as a scaled-content manual action in April 2026. His screenshots showed the directory dropping in Google's standard results, AI Overviews, and AI Mode, then disappearing from most of the tested ChatGPT answers that had cited it before.

The important lesson is not "Google controls ChatGPT." The lesson is narrower and more useful: if a section loses eligibility, indexing, or trust in Google Search, that same section can also become much harder to surface in answer engines that rely on live web retrieval and citations.

Why is National Today's /us/ directory a useful AEO case?

It is a useful AEO case because it shows a downside path, not another cherry-picked win. Gabe's April 20, 2026 post focuses on NationalToday.com's /us/ directory, which he says contained more than 850,000 AI-generated local-news URLs and then vanished from Google's index after a manual action for scaled content abuse.

That makes the case more interesting than a generic "AI search is changing" article. We get a specific directory, a specific policy category, screenshots of visibility trending, and before-and-after citation examples from ChatGPT. In other words, the subject is concrete enough to evaluate.

It also answers a question AEO teams actually care about: if you try to manufacture answer-engine visibility with scaled content, can the downside spread beyond blue links? In this case, the answer looks like yes.

What actually happened after the manual action?

The observed pattern was a multi-surface drop, not a single-channel dip. Gabe writes that the /us/ directory had been ranking in traditional Google surfaces and that, after the manual action, search visibility for that section "came crashing down." He then shows accompanying Brand Radar screenshots for declines in AI Overviews and AI Mode.

The evidence matters because Google's own documentation makes the dependency explicit. Google says pages shown as supporting links in AI Overviews or AI Mode must be indexed and eligible to appear in Google Search with a snippet. If a section loses that baseline eligibility, AI feature visibility is at risk too.

This is the cleanest part of the case. We do not need to infer much there. Google's documentation already states the rule, and Gabe's screenshots line up with it:

Surface Evidence Gabe published Why it matters
Google search Visibility trend collapsed after the action The section lost classic search exposure first
AI Overviews Brand Radar trend dropped Google AI visibility depends on index eligibility
AI Mode Brand Radar trend dropped Same dependency, different Google AI surface

For AEO, that is a reminder that answer-engine optimization still sits on top of search eligibility. If the content cannot stay in the trusted retrieval set, the fancy AI layer does not save it.

Why would Google AI features fall with the indexation loss?

They would fall because Google says AI Overviews and AI Mode still depend on indexed, snippet-eligible pages. In its "AI features and your website" documentation, Google says there are no special AI requirements, but it also says a page must be indexed and eligible to appear in Google Search with a snippet to be shown as a supporting link.

That matters more than any "AI optimization checklist." Teams often talk as if AI Overviews sit on a separate ranking system with separate tactics. Google's documentation suggests a less glamorous reality: existing SEO fundamentals, policy compliance, crawlability, and visible text availability still gate entry.

The National Today case fits that exactly. If Google judged the /us/ section to violate scaled-content policy and removed or restricted it, then losing AI Overviews and AI Mode visibility is not surprising. It is the expected downstream effect.

Google's spam policies reinforce the same point. The documentation says Google detects policy violations through automated systems and human review that can result in a manual action, and that sites violating the policies may rank lower or not appear in results at all. It also defines scaled content abuse as generating many pages primarily to manipulate rankings rather than help users, including using generative AI to create many pages without adding value.

That description is why this case is more than gossip. Gabe's interpretation maps to a named Google policy, and the observed collapse matches the consequence Google documents.

What does the ChatGPT evidence actually show?

The ChatGPT evidence shows a strong correlation, not a proven mechanism. Gabe says he tested prompts where National Today's /us/ pages had appeared in ChatGPT answers before the manual action, then found that for "almost all" of the prompts he checked, those citations were gone afterward.

He also includes several before-and-after examples in the article:

  • One ChatGPT answer cited a National Today page before the action, then no longer cited it afterward.
  • Another prompt previously cited and mentioned the directory, then lost both the citation and mention.
  • A third example showed the same pattern for AI-generated local-news content.
  • Gabe notes that a few random citations still remained, which is an important caveat rather than an inconvenience.

That last point improves the case. If every citation vanished instantly, we would suspect sampling or tooling artifacts. Instead, Gabe reports a mostly-gone pattern with a few survivors, which is consistent with a probabilistic retrieval system that can still surface remnants through alternate query paths.

OpenAI's own documentation gives us the bounded mechanism we can safely claim. ChatGPT Search uses the web, can automatically search when current information would help, and includes inline citations or a Sources panel when search is used. OpenAI also says inclusion depends in part on allowing OAI-Searchbot to crawl the site. That is enough to support a practical interpretation: live web eligibility and retrievability affect citation chances.

Why is this not proof that Google directly controls ChatGPT citations?

It is not proof because the case does not isolate ChatGPT's exact source-selection stack. Gabe argues that ChatGPT can leverage Google's index when grounding answers, but the fetched evidence here does not let us prove how often that happens or whether it drove every lost citation in this case.

We should keep the claim narrower. The case shows that after the /us/ directory was removed from Google's visible search ecosystem, most tested ChatGPT citations to that directory also disappeared. That could happen because the pages became harder to discover, because other sources replaced them in retrieval, because query rewrites changed the candidate set, or because ChatGPT's web-search layer responded to the same trust signals indirectly.

This limitation does not make the case weak. It makes it usable. AEO teams do not need perfect visibility into ChatGPT's ranking stack to act on the operational lesson: if your content loses crawl access, policy eligibility, or search trust, expect answer-engine visibility to become less stable too.

What is replicable here, and what is circumstantial?

The replicable part is the monitoring workflow. Gabe tracked one section across classic rankings, Google AI surfaces, and ChatGPT citations before and after a visible policy event. Any serious AEO team can do the same for its own content.

For example, you can build a simple watchlist around:

1. One directory or template family. 2. Ten to twenty prompts that already cite or mention that content. 3. Google indexation checks. 4. AI Overview and AI Mode presence checks. 5. ChatGPT Search citation checks.

The circumstantial part is the specific site and content model. National Today's /us/ directory was an extreme case: very large volume, clearly template-heavy, and described by Gabe as fully AI-generated local news. You should not assume a small informational site will see the same speed or severity of collapse.

The other circumstantial factor is prompt sampling. Gabe says he still wanted to test more prompts and that a few citations remained. That means we should treat the ChatGPT portion as a strong practitioner observation, not a controlled prompt-universe measurement.

What should publishers copy from this case, and what should they avoid?

Publishers should copy the risk model, not the content model. The correct takeaway is that AEO cannot be separated from policy-safe publishing and search fundamentals.

What teams should avoid is the common shortcut logic: "If answer engines reward coverage breadth, we can mass-produce pages until one sticks." Google's spam policy is explicit that large-scale low-value generation is a risk area, and the National Today example suggests the downside can spill into AI surfaces too.

What teams should copy instead is the retrieval checklist hidden inside Google's and OpenAI's documentation:

  • Keep important content in visible text.
  • Make sure pages remain crawlable and indexable.
  • Keep structured data aligned with visible content, not as a substitute for it.
  • Allow the relevant crawlers when you want inclusion.
  • Monitor citations at the URL level, not just at the domain level.

If you need a minimal technical check for ChatGPT eligibility, the OpenAI documentation points to crawler access, which can be reviewed in robots.txt:

User-agent: OAI-Searchbot
Allow: /

That snippet does not guarantee citations. It just removes one avoidable blocker. The National Today case is about a bigger blocker: if the underlying content loses trust and eligibility, crawler access alone will not rescue it.

What is the counter-evidence or alternate explanation?

The counter-evidence is that Gabe did not run a controlled experiment proving the manual action alone caused each ChatGPT citation loss. ChatGPT answers are variable, prompts can be rewritten, and citation sets can change for reasons unrelated to a Google action on one site section.

There is also a residual-citation caveat in the case itself. Gabe found at least one example where the /us/ directory was still cited after the manual action, which suggests the section was not erased from every retrieval path immediately.

Those caveats matter, but they do not erase the main pattern. Google's own documentation explains why AI Overviews and AI Mode would fall when a page is no longer eligible in Search, and Gabe's before-and-after ChatGPT examples are consistent enough to treat the spillover risk as credible.

The honest reading is: this case is strong enough to change operating behavior, but not strong enough to support sweeping claims about ChatGPT's entire ranking system.

What to do Monday morning

1. Pull one high-volume directory, template, or programmatic content set and review whether it would survive a manual human quality review. 2. Check whether the pages are still indexed, still eligible for snippets, and still appearing in any prompts where they used to be cited. 3. Build a small prompt panel for ChatGPT, AI Overviews, and AI Mode so you can catch citation loss before traffic reports make it obvious. 4. Review robots.txt, CDN rules, and bot controls to confirm you are not accidentally blocking Googlebot or OAI-Searchbot. 5. Stop treating AEO as a separate loophole from SEO policy compliance; it is downstream from the same trust and eligibility layer. 6. If a page family is thin, templated, or mass-generated, fix the content model before you try to scale distribution.

Sources