Here’s a number that should be reshaping every SEO team’s plan for this quarter: 38%.
That’s the share of AI Overview citations now coming from pages that rank in Google’s top 10 for the same query, as per Ahrefs new study. Seven months ago, that number was 76%.
Ranking and citation just decoupled, fast. You can rank #1 and still lose. And, you can rank #47 and still win.
About 31% of AI Overview citations now come from pages outside Google’s top 100 entirely, often from YouTube, Reddit, or LinkedIn or through fan-out process.
If you’ve been in SEO for any length of time, this is the moment something breaks in your brain. Because every instinct you’ve built up over a decade tells you the formula is supposed to be: rank → get cited → win. But AI Overviews don’t follow that formula.
The reason isn’t mysterious. It’s just been buried under a pile of hot takes about “content quality,” “E-E-A-T,” and “publish more” that don’t actually explain what’s happening.
After running citation diagnostics on dozens of client sites in 2025-2026, I’d reduce why your content doesn’t appear in AI Overviews down to six failure modes. Most of them are invisible if you’re still measuring success through traditional SEO metrics.
This post breaks down all six, with the data behind each, plus a 5-step diagnostic you can run on your own pages today.
The biggest question: “I rank, so why am I not cited?”
Here’s the assumption that breaks first.
For two decades, ranking was the proxy for everything. If you were on page one, you got traffic. If you were #1, you got most of the traffic.
SEOs learned to treat ranking as the outcome, because it was tightly correlated with the actual outcome (clicks).
That correlation just collapsed.
The Ahrefs study isn’t an outlier.
- BrightEdge’s parallel analysis using a different dataset and methodology found the top-10 overlap was even lower, around 17%.
- Seer Interactive measured a 61% drop in organic CTR for queries that trigger AI Overviews (1.76% to 0.61%).
- Discovered Labs went deeper and traced the network calls: AI Overviews load from a separate /async/folsrch endpoint, with citation selection happening in under 200 milliseconds against a passage retrieval index that operates independently from the main organic ranker.
In simple terms, AI Overviews aren’t picking the top 10 organic results and summarizing them. They’re running a different retrieval system on a different unit of content (passages, not pages) with different selection criteria.
The page that ranks #1 isn’t given any preferential treatment over the page at #45 in this second system.
How AI Overviews Actually Pick What To Cite
Most diagnoses go wrong here, at the start, because teams assume the AI Overview is reading the same signals as the organic ranker. Well, it isn’t.
When a user types a query that triggers an AI Overview, Google’s AI Mode (powered by the Gemini family in 2026) doesn’t pass that query to the same traditional ranker that orders blue links but to the AI Mode/AIO ranker. Google runs a four-stage process in under a second:
- Query fan-out: The user’s query gets decomposed into 8-12 sub-queries covering different angles, follow-up intents, and adjacent entities. Each sub-query is retrieved separately.
- Passage retrieval: For each sub-query, Google runs vector-based retrieval against a passage index, not a page index. Your content is parsed into 40-200 word chunks and scored on semantic similarity. About 31% of final citations end up coming from pages outside the top 100 organic results entirely.
- Credibility filtering: Retrieved passages pass through an E-E-A-T-style filter that weighs author signals, site authority, and third-party brand mentions. Roughly 96% of cited sources clear this gate.
- Generation and citation: Surviving passages get fed to Gemini, which composes the answer and chooses which to attribute. The model favors passages with named-source data, dated stats, and self-contained claims it can quote without context.
Two implications almost nobody internalizes.
First, the retriever reaches outside the organic index. Profound’s research, highlighted by Aleyda Solís, shows 52% of AI Overview citations point to pages outside Google’s top 50 organic results.
Second, your content is judged as a parts catalog, not a document. Each chunk scores on its own. A perfect page with one weak passage can lose to a mediocre page with one excellent passage on the same topic.
Every failure mode below is a place where this four-stage process trips your content up.
Three differences from classic SEO are worth pinning down before we get to the failure modes…
- Retrieval and ranking are decoupled. The page that ranks #1 isn’t given preferential treatment over the page at #45 in this second system.
- The reading unit is the passage, not the page. Your content is parsed in 40-200 word chunks, scored independently, and cited independently.
- Trust signals weigh heavier than they did in classic SEO. Roughly 96% of cited sources pass an E-E-A-T credibility filter before relevance scoring even begins.
Now let’s jump to the main topic.
6 Reasons Why Your Content Doesn’t Appear in AI Overviews
Reason #1: Google can’t crawl, render, or use your content for AI Overviews
The first failure happens before retrieval is even a question. Your content isn’t getting into the candidate pool because Google either can’t read your page properly, or you’ve explicitly told Google it can’t be used in AI training and generation.
Three things to check today:
- Your Google-Extended token in robots.txt: This is the specific opt-out signal Google introduced for AI training and generation. If you blocked Google-Extended in 2023 or 2024 (a lot of publishers did, defensively, when it launched) you’re now blocking your own content from being used in AI Overviews. Allow it back.
- Your JavaScript rendering: Googlebot does render JavaScript, but render budget is finite, and pages that load critical content client-side without SSR routinely get indexed with the body content missing. Open your page in Google’s URL Inspection tool and check the rendered HTML, not the source. If your answer to the query isn’t in the rendered output, the passage retriever doesn’t have it either.
- Your hosting firewall and WAF rules: Some providers (Cloudflare, AWS WAF, others) ship with rules that throttle or challenge Googlebot when traffic spikes look suspicious. Check your access logs for 4xx and 5xx responses to Googlebot user agents over the last 30 days. Spikes here directly correlate with crawl drops.
This is the most boring failure mode and the most common. Roughly 1 in 3 client sites I audit has at least one of these issues actively blocking AI Overview eligibility on at least some pages. It costs you nothing to fix and gets ignored 90% of the time.
Reason #2: You’re filtered out at the E-E-A-T credibility gate
Now suppose your content is crawlable. The next gate is credibility, and AI Overviews apply it more aggressively than the traditional organic ranker does.
Roughly 96% of cited sources pass an E-E-A-T credibility filter before relevance even gets scored.
That means if your page has no clear author, no expert credentials anywhere on the site, no real publishing history, and no third-party brand mentions you can point to, you’re getting filtered out before passage retrieval even runs.
What does the credibility gate actually look at?
- Author byline + bio: Is there a real human attached to the content with a credible bio? Anonymous content gets devalued hard.
- Author’s external footprint: Does the author have a verifiable presence outside this domain (LinkedIn, conference talks, other publications, podcast appearances)?
- Site-level authority signals: About page, contact info, editorial standards, schema markup with author and organization entities, consistent publication metadata.
- Publication history on the topic: Does the site have a track record on this subject, or is this a one-off page that suddenly appeared targeting a money keyword?
If you’re publishing AI-generated content under no byline, on a thin domain, with no organizational schema, you can rank for low-competition keywords. You will not be cited in AI Overviews for high-stakes ones.
The credibility gate is doing exactly what Google says it’s doing: filtering out content that doesn’t carry trust signals.
Reason #3: Your domain lacks entity weight in the candidate pool
The next issue is, even if your content is crawlable and credible, your domain might lack the entity weight to make the candidate pool for high-competition queries.
Entity weight is the distributed signal of your brand’s existence across the open web like brand mentions across third-party platforms, YouTube videos featuring your name or product, Reddit threads that quote you, LinkedIn posts by your founder, G2 and Capterra reviews, podcast appearances, news coverage, and more.
The retriever reads these as a single distributed signal that says: this brand is real, has an opinion, and gets discussed.
Profound’s data shows 52% of AI Overview citations point to pages that aren’t even in Google’s top 50 organic results. The retriever is reaching outside the organic index, and what pulls it toward you is your entity signal.
Ahrefs’ Brand Radar data makes this concrete:
- YouTube is now the most-cited domain in Google AI Overviews, growing 34% in six months.
- Among AIO citations from URLs that don’t rank in the top 100, 18.2% are YouTube videos.
- Reddit holds roughly 21% of AIO citations.
- LinkedIn is the most-cited domain for professional and B2B queries across major AI search platforms, per Profound’s March 2026 dataset.
Last year I watched this pattern play out with a client: their blog ranked top 3 for two dozen keywords and almost never got cited in the AI Overviews for those queries. Their CEO started posting twice a week on LinkedIn about the same topics in his own voice, with proprietary data from their platform. Six weeks in, the LinkedIn account began getting cited for those exact queries.
The blog’s citation rate didn’t change. The brand’s overall AI visibility roughly doubled.
You can’t backlink-build your way out of this one. Backlinks still matter, but they’re table stakes now, not a differentiator. What moves the needle is presence on the surfaces where AI retrievers go looking for trust signals.
Reason #4: Your page is structurally unparseable for passage retrieval
Suppose you clear the entity gate. Your domain’s in the candidate pool. Your page still isn’t getting cited. This is the problem I see most often on audits.
Three patterns are worth tattooing on your forehead (not really, lol!)
- Position on the page beats position in the SERP: Studies of citation behavior across both Google AI Overviews and ChatGPT consistently find that the first 30% of a page produces a disproportionate share of citations (around 44% for ChatGPT). If your direct answer to the query lives in your conclusion, after a 600-word setup, you need to pay serious attention there.
- The model reads your page as a parts catalog, not a document: Each chunk has to make sense in isolation. A passage that opens with “this is why” makes no sense to an extractor that doesn’t have the previous paragraph. A sentence like “the second method works better” references something the AI doesn’t have access to when it grabs that single sentence. Self-contained chunks aren’t a stylistic preference anymore. They’re a precondition for being quotable.
- Stat density beats prose density: AIO extractors prefer “X grew 47% in 2025 according to Y” over “X has been growing rapidly recently.” Pages with at least one named-source citation in the body get cited about 2.1x more often. Pages with three or more original data points (proprietary surveys, internal benchmarks, exclusive case studies) get cited up to 4x more than pages without, based on practitioner audits I’ve run.
Try this on your top-performing post right now. Pull one paragraph at random. Cover the rest of the page with your hand. Can a reader understand the claim, the evidence, and the source from that paragraph alone? If not, the AI extractor’s having the same problem.
Ryan Law, Director of Content Marketing at Ahrefs, has been right about this. He’s warned the SEO community against turning content into spammy markdown nobody actually wants to read just to chase citations. Well, he’s right. The fix isn’t to chunk your post into oblivion or stuff every section with subheadings. The fix is to write each section so it earns its place both inside a narrative and as a standalone unit. That’s harder than it sounds, and almost nobody’s doing it well.
Reason #5: You’re answering the user’s query, not Google’s query fan-out
This is the subtlest issue, and the one that shows up even in well-structured, top-of-page, citation-friendly content.
When someone triggers an AI Overview, Google doesn’t just retrieve that one query. It runs a query fan-out, decomposing the user’s question into multiple sub-queries that explore different angles, follow-up intents, and adjacent entities.
Mike King at iPullRank, who’s been the loudest voice on this through 2025-2026, calls the resulting work “relevance engineering“: building content that satisfies the fan-out, not just the surface query.
Here’s why this trips up smart SEOs. Say you write a post titled “best LLM tracking tools” and target that exact phrase. You rank well. But when a user actually triggers an AI Overview on it, Google fans out into queries like:
- What features matter most for a best LLM tracking tool?
- Best LLM tracking tools pricing comparison 2026
- Best LLM tracking tools for small businesses
If your page is monomaniacally about the literal parent phrase, it gets cited (maybe) for the parent query, while your competitor’s deeper coverage gets cited five more times across the fan-out. That’s the visibility gap most teams don’t see, because they’re not looking for it.
Reason #6: You picked the wrong content format for the query
This last one gets ignored the most, and it’s costing brands real visibility.
Different query types pull from different content formats. AI Overviews don’t treat a blog post as inherently better than a video, a Reddit thread, or a LinkedIn post. They treat each format as roughly interchangeable, with the format that best matches the query winning the citation.
Look at the pattern from current data:
- For “how-to” queries with visual components, YouTube wins about 18% of out-of-top-100 citations.
- For “best X” listicles, the top-cited source is often a G2 or Capterra page, not a brand blog.
- For B2B and professional queries, LinkedIn is the most-cited domain across major AI search platforms.
- For experiential and “honest review” queries, Reddit accounts for around 21% of all AIO citations.
- For health-related queries in some markets, YouTube is the most-cited domain, outranking even official medical sources.
If you wrote a 2,000-word blog post for a query that the retriever wants a video for, you can be the deepest, best-written page on the topic and still lose.
The fix isn’t to rewrite the post. The fix is to add the format the query wants alongside it.
Here’s the format-to-query map I use on audits:
- How-to + visual component: Add a YouTube video.
- Best X listicles/software comparisons: Get on G2/Capterra; pitch into existing high-authority listicles.
- B2B/professional queries: Original content on a founder LinkedIn.
- “Honest review”/experiential queries: Active presence in relevant subreddits.
- Definitional/informational queries: Long-form blog post still wins.
- News/breaking topics: News coverage + LinkedIn commentary.
The brands winning AI Overviews in 2026 aren’t the ones with the best blogs. They’re the ones with diversified content footprints across the formats AI Overviews actually pull from.
The Freshness Paradox Almost Nobody Talks About
A common piece of advice in 2026: AI Overviews favor recency, so update everything every 90 days.
Half right, mostly wrong.
Citation pattern studies through Q1 2026 consistently find that the median age of a cited page is around 14 months.
Recency isn’t the lever it’s marketed to be. What the data actually shows is more subtle: the model devalues undated content and content that contradicts newer dated sources. So a 14-month-old page with a clear “Updated: March 2026” stamp and dated stats inside it can outperform a page published yesterday with no dates anywhere.
The takeaway: don’t churn-update your evergreen pages every quarter for a productivity dopamine hit. Date your data points, version your content visibly, and let your durable pages stay durable.
How To Check if Your Content Is Eligible for AI Overview Citations
Here’s a 5-step audit that can save enormous amounts of work.
- Open the AI Overview for your target query: Is your domain in any cited source, including a Reddit thread quoting you, a YouTube video about you, or a LinkedIn post by your founder? If not, you have a retrieval problem, not a content problem.
- Read your page’s first 200 words: Do they contain the direct, citable answer to the query? Or do they warm the reader up with context first? If the answer is buried, your structural problem is bigger than your content problem.
- Run query fan-out on your target keyword using Qforia or a similar tool: Map your page’s sections against the sub-queries. How many gaps?
- Pull one paragraph at random: Cover the rest of the page with your hand. Can a reader understand the claim, the evidence, and the source from that paragraph alone? If not, the AI extractor’s having the same problem.
- Check your off-page entity strength: Search your brand name in Google. Do you see third-party mentions on YouTube, LinkedIn, Reddit, G2, podcasts, news? The retriever reads these as trust signals, and for newer brands, this is usually the bottleneck.
If you’re failing on all five, don’t start with content. Start with the entity signals (#5) and the page structure (#2 and #4). Those are the highest-impact fixes by a wide margin.
How To Scale This Audit Beyond 5 Pages
The 5-step diagnostic above works fine for the 3-5 pages your business actually depends on. It falls apart at 50, 200, or 500 pages, which is what most content teams are managing in 2026.
This is where dedicated AI visibility tracking earns its keep.
Rankshift is built specifically for this: it tracks your brand’s citation presence across Google AI Overviews, ChatGPT, Perplexity, Gemini, and Copilot, surfaces which queries you’re winning and losing, shows which competitors are taking the citation slots you’re missing, and flags new mentions as they appear. Instead of running the diagnostic manually one page at a time, you get the diagnosis for your full keyword universe in one view.
Worth flagging: tracking is the piece almost nobody does well right now. Search Console doesn’t show this data. GA doesn’t show this data. Conventional rank trackers don’t show this data either. If you’re serious about AI visibility, the first move is getting a baseline. Everything else this post talked about comes after that. Why not give an AI Overviews rank tracker a try?
What This Means for Your 2026 Content Strategy
I’ll close with the awkward conversation most heads of content are having internally: how much of next year’s budget should move from traditional content production to AI search visibility?
My honest answer, after these audits: probably more than you think, and in different directions than you’d guess.
The pages you already have that rank well are still earning their keep, even if their citation rates are lower than you’d like. Don’t decommission them. Restructure the top 20% for passage-level citation, then leave them alone.
What I’d reallocate: roughly 30-40% of net-new content production budget into off-page entity work. Founder LinkedIn content. YouTube explainers built around your strongest topics. Podcast tours. Original research you can pitch to reporters. G2/Capterra profile depth. These were nice-to-haves under classic SEO. Under AI Overviews, they’re the candidate pool gate.
That math will look different for every team. But “publish another 20 blog posts” almost never tops the priority list anymore.
Try #1 AI Overview Tracker – Rankshift
If your content isn’t appearing in AI Overviews, it’s almost never because the content is bad. It’s because you’re writing for a reader that no longer exists in this part of the web. The new reader (the retriever) reads in passages, weighs entities heavier than backlinks, and runs five sub-queries for every query you targeted.
Build for that reader. Your rankings won’t suffer. And your AI visibility will start moving on the timeline that matters: weeks, not months.
One last thing. If you’d rather see your own citation gaps before you start, book a demo with Rankshift. You’ll see which queries your competitors are winning AI Overviews on, where you’re losing citation slots, and which fixes from this post will move the needle for your specific keywords. Knowing what you’re aiming at makes the work easier.
