AI search has a content quality problem, and the SEO industry helped create it.
That sounds harsh, but this week’s reporting made the issue hard to ignore. In Search Engine Journal’s analysis of the “AI slop loop”, Lily Ray lays out what happens when search systems retrieve weak, synthetic, derivative pages and turn them into confident answers. The faster the web fills with generic content, the easier it becomes for AI search tools to cite garbage with a straight face.
At the same time, a New York Times investigation summarized by Oumi put real numbers on the trust problem inside Google AI Overviews. Oumi found that AI Overviews powered by Gemini 3 were correct about 91% of the time, but only 39% were both correct and fully supported by their cited sources. Accuracy improved. Verifiability got worse.
That is the tension marketers need to understand right now. The problem is not simply that AI systems hallucinate. It is that retrieval-based search often pulls from a web already crowded with thin, repetitive, low-evidence pages. When that happens, brands that publish commodity content are no longer just wasting budget. They are polluting the answer layer their future customers rely on.
The good news is that this creates a real opening for brands willing to do the opposite. If your content is specific, evidence-backed, clearly structured, and easy to verify, AI systems have a better chance of using it correctly. This is where answer engine optimization becomes practical. It is less about stuffing pages with a new acronym and more about making your content citable.
The problem is not AI content, it is commodity content
A lot of marketers are framing this debate the wrong way. They keep asking whether AI-written content is bad, as if the deciding factor is who or what drafted the first version.
That misses the point.
The real divide in 2026 is commodity content versus citable content.
Commodity content is the page that could live on any agency blog and nobody would notice. It restates public information, makes broad claims, adds no firsthand evidence, and sounds polished enough to pass a quick skim. It may even rank for a while if the query is weak enough. But it does not help an AI system distinguish signal from noise because it contributes nothing original to the web.
Citable content does something different. It answers a clear question, supports claims with evidence, explains the source of that evidence, and presents information in a way a human or machine can follow. It has a point of view. It says something worth extracting.
That is why the “AI content versus human content” argument has become less useful. A skilled team can use AI in research, outlining, QA, and drafting support without publishing generic sludge. A careless team can publish lifeless copy with or without AI. The output quality still comes down to editorial judgment.
This matters because retrieval systems do not care how sincere your workflow felt. They care what is on the page, how understandable it is, and whether other sources reinforce it.
Why AI search is exposing the mess now
Traditional search tolerated a lot of mediocre content because ranking and clicking are separate events. A weak page could still earn traffic if the keyword match was decent, the domain had authority, and the SERP was soft.
AI search compresses that process.
Instead of showing ten blue links and letting the user sort it out, the system reads across sources, synthesizes a response, and decides which claims deserve to be cited. That puts much more pressure on content quality. If the source material is vague, repetitive, or wrong, the answer layer inherits those weaknesses.
Ray’s piece is valuable because it describes the contamination as a retrieval problem, not just a training-data problem. That distinction matters. Marketers often hear “model collapse” and assume the risk is long-term. Retrieval contamination is immediate. A bad page can be published, crawled, retrieved, and echoed back to users without waiting for the next model release.
The Oumi numbers make that urgency tangible. In the sample covered by its report, correctness improved from Gemini 2 to Gemini 3, yet ungrounded answers increased. In plain English: answers can look more polished while becoming harder to verify against their own citations.
For brand marketers, that means you cannot judge AI visibility on appearance alone. A mention inside an answer is not automatically a win if the surrounding framing is wrong, thin, or attributed to the wrong source. Visibility and trust now have to be measured together.
What weak content looks like in a retrieval system
Most teams do not set out to create unusable content. They just follow incentives that used to be good enough.
Here is what repeatedly breaks inside AI retrieval.
1. The page says nothing new
If your article is a cleaned-up remix of the top five results, it offers no reason to cite you. AI systems already have access to the original sources. A derivative summary without firsthand data or a sharper interpretation is disposable.
2. The answer is buried under throat-clearing
Many SEO pages wait 400 to 700 words before making a real point. That is annoying for readers and risky for retrieval. If the answer only appears after a long setup, the system may pull a partial explanation or miss the strongest line entirely.
3. Claims are unsupported
Pages often use phrases like “studies show” or “many experts agree” without linking to anything. That worked when the page only needed to sound plausible. It fails when AI systems and skeptical readers need to verify the claim.
4. Every section sounds interchangeable
This is a common symptom of scaled content production. The H2s look clean, but each section repeats the same idea with slightly different wording. That makes extraction harder because there is no strong informational anchor.
5. The brand has no reinforcing footprint
Even a good page can struggle if nothing else on the web supports the entity behind it. Reviews, author bios, external mentions, case studies, podcasts, and consistent brand descriptions all help AI systems trust what they retrieve.
If any of those problems sound familiar, this is fixable. But it starts with admitting that some older SEO habits are now liabilities.
What citable content looks like instead
The easiest way to improve AI visibility is to stop writing for a generic SERP model and start writing for a verification model.
That changes how you build the page.
Lead with the answer
State the point in the first paragraph. Do not save the thesis for later. If the article exists to prove that AI search is exposing the content quality gap, say that plainly and support it below.
Use real sources inline
Every external claim should have a real source link in the sentence where it first appears. That is now table stakes. It helps users, it helps editors, and it helps machines connect the claim to supporting evidence.
Add a distinct point of view
A page becomes more citable when it adds interpretation, not just aggregation. Your value is not only collecting facts. It is explaining what they mean for a specific audience. In Emarketed’s case, the useful lens is often: what should agency owners, healthcare marketers, and in-house teams do next?
Make sections self-contained
Each H2 should be able to stand on its own as a useful answer. That improves readability and makes the content easier for AI systems to extract without losing context.
Show your work
Original screenshots, case study numbers, process notes, and firsthand observations all help. The more a page looks like evidence instead of filler, the more valuable it becomes.
This is also where many service pages need attention. A beautifully designed page that makes big promises with vague copy may still convert some humans. It is much harder for an AI system to cite confidently. That is one reason strong AEO services now require editorial discipline, not just technical SEO work.
The healthcare and YMYL problem is even bigger
The content quality issue gets more serious in healthcare, finance, and other high-trust categories.
When an AI system retrieves bad information about shoes, the user gets annoyed. When it retrieves bad information about symptoms, treatment options, or provider credibility, the stakes are much higher. That is why healthcare brands need cleaner evidence chains, stronger authorship signals, and tighter review processes than the average publisher.
This is not theoretical for Emarketed’s audience. Healthcare marketers are already dealing with a visibility split between classic rankings and AI answers. We have written before about the healthcare AI search trust problem, and the retrieval-quality issue makes that problem harder, not easier.
We also see the upside when the content foundation is strong. Seasons in Malibu holds 4,200+ keyword rankings, 814K+ monthly social impressions, and averages 5 patient admits per month driven directly through Emarketed’s marketing. That result depends on more than rankings. It reflects a full-service system where expertise, authority, cited pages, and platform consistency reinforce each other. In an AI search environment, that kind of reinforcement matters.
For healthcare teams, this is the practical takeaway: if your content can survive clinical review, compliance review, and skeptical user review, it is much more likely to survive AI retrieval too.
A practical cleanup plan for brands and agencies
Most teams do not need more content. They need a content cleanup.
Here is the playbook I would use this quarter.
Audit your top 50 pages for citation readiness
Look at the pages most likely to influence revenue: service pages, high-intent blog posts, comparison pages, FAQ hubs, and healthcare education content. Ask four questions.
- Does the page answer the core question in the first 100 words?
- Does it link to every external claim worth checking?
- Does it include a specific example, data point, or firsthand observation?
- Could a competitor swap logos with you and publish basically the same article?
If the answer to the last question is yes, rewrite it.
Consolidate duplicate and near-duplicate content
Many sites have three or four weak posts covering nearly the same topic. In an AI environment, that does not build authority. It fragments it. Merge overlapping pages into a stronger canonical asset with better structure and better evidence.
Tighten authorship and review signals
Add expert reviewers where appropriate, especially in healthcare and other YMYL categories. Make author and reviewer credentials easy to find. If the expertise is real, surface it.
Replace filler with proof
Delete empty lines like “businesses need to adapt to changing consumer behavior” unless you can show how, why, or with what result. Replace generic statements with screenshots, numbers, workflow details, or examples.
Track AI visibility separately from rankings
A page can hold rankings and still disappear from the answer layer. That is why more brands are investing in citation tracking and answer-surface monitoring. If you are not measuring where your brand appears, you are guessing. We covered that in more depth in our post on the AI citation tracking gap.
Treat editing as a growth lever
A lot of AI-search wins will come from improving what you already published, not from cranking out net-new articles. Strong editing is now a distribution strategy.
What agencies should tell clients right now
Clients do not need another abstract lecture about how AI changes everything. They need a clearer explanation of what changed in their actual content strategy.
Here is the short version.
First, rankings still matter. Keep the technical and authority foundations strong. Do not abandon classic SEO.
Second, generic content is losing value faster than most reporting dashboards show. Pages that used to be “good enough” are becoming weak source material for AI answers.
Third, the goal is no longer just to publish a lot of useful-looking pages. The goal is to publish fewer, sharper, evidence-backed assets that can earn both clicks and citations.
Fourth, content QA can no longer stop at spelling, metadata, and on-page basics. Teams need editorial review for distinctiveness, proof, structure, and source quality.
That is the conversation agencies should be having with clients right now, especially in B2B and healthcare. The web is getting noisier. Clean signal is becoming more valuable.
The brands that win will sound more human, not less
There is a strange irony in all of this. As AI search spreads, the winning pages are not the ones that sound the most machine-optimized. They are the ones that feel grounded.
Grounded means the page has a real perspective. It names the tradeoff. It uses actual numbers. It links to the original source. It sounds like someone who knows the topic took the time to make a useful argument.
That is why I am not worried about the volume game in the long run. AI can flood the web with average content. It still struggles to manufacture firsthand proof, precise judgment, and earned authority at scale.
Brands that keep publishing interchangeable pages will train themselves into irrelevance. Brands that publish citable work will become easier to trust, easier to quote, and harder to replace.
If your team is staring at a bloated content library right now, that is actually good news. The fastest win is not writing 100 more articles. It is turning your best existing pages into the clearest source in the room.
If you want help figuring out which pages to fix first, this is the kind of work a serious SEO strategy and AEO program should handle together. Or if you already know your content library needs a reset, work with us and we can help you prioritize the cleanup.
FAQ
Is AI-written content automatically bad for SEO or AEO?
No. The bigger issue is whether the finished page is generic, unsupported, or derivative. AI can help with research and drafting, but pages still need editorial judgment, evidence, and a distinct point of view to be useful in search and AI retrieval.
What does “citable content” mean?
Citable content is content that answers a clear question, supports claims with real sources or firsthand evidence, and is structured so humans and AI systems can easily understand and verify it.
Why are AI Overviews sometimes correct but still untrustworthy?
Because an answer can be factually correct while citing sources that do not fully support the claim. That is what Oumi’s April 2026 analysis highlighted: correctness improved, but grounding against cited sources often lagged behind.
Should brands delete old low-quality blog posts?
Sometimes, yes. Other times the better move is consolidation or a full rewrite. The right decision depends on whether the page has rankings, links, or useful structure worth preserving. The key is to stop leaving weak, duplicative pages untouched.
How do I know if my brand is showing up in AI search?
You need to track citations and answer visibility separately from standard keyword rankings. Manual query testing helps, but brands should also monitor AI surfaces systematically so they can see where their content is being cited, summarized, or ignored.
What is the best first step this week?
Pick your ten highest-intent pages and review them for four things: direct answers near the top, source links for key claims, specific proof points, and a point of view that does not sound interchangeable with a competitor. That audit will surface most of the real problems quickly.
