If you’ve been pouring budget into clawing back to position #1 on Google for your money keywords, here’s the punch in the gut: 83% of AI Overview citations come from pages that are NOT in the traditional top 10 organic results. That number, pulled from large-scale 2025 citation analyses, broke a lot of assumptions in the industry — and most agencies still haven’t adjusted their playbooks for it.
I’ve been doing search optimization since 1996. The rule used to be simple: rank, get clicks, repeat. AI Overviews flipped that. The cited source isn’t necessarily the top-ranking page — it’s the page that answers the specific sub-question better than anyone else. And usually that page is sitting somewhere between position 11 and position 60, getting almost no organic traffic, written by someone smaller than the brands in the top 10.
That gap is the entire opportunity for boutique sites and operators right now.
The mechanic — why deep pages outrank brand pages in citations
AI Overviews don’t behave like a traditional ranking algorithm. They behave like a retriever stitching together fragments. When a user asks Google or Gemini something like “how does schema markup interact with entity disambiguation for B2B SaaS,” the system fans out into a dozen related sub-queries, pulls the cleanest 40–60 word answer for each fragment, and assembles a synthesized response. The cited URL beneath each fragment isn’t the page Google would rank #1 for the parent query. It’s the page that most directly answers that one fragment.
That’s why long-tail beats head terms in AI citations. Big brand pages tend to be wide and shallow — written for the parent keyword, optimized for click-through, padded with intro fluff. Smaller sites, when they’re written by an actual practitioner, tend to be narrow and deep — one specific question, answered crisply, with named tools, real numbers, and an opinion. That’s exactly what retrieval systems extract.
There’s also a citation diversity preference baked into how AI Overviews are tuned. Google has explicitly said it tries to surface multiple sources per answer rather than collapsing everything onto one domain. That means even on queries where a Wikipedia-class brand owns the top three slots, Overviews will deliberately reach further down to find a second or third citation from somewhere different. Page 5 of the SERP is no longer dead real estate — it’s where Overviews quietly shop.
What to do this week
1. Map your “almost-ranking” pages. Pull the Search Console queries where you sit between positions 8 and 30. Filter for question-shaped queries (starts with how, why, what, when, can, should, vs.). These are your AI Overview candidates. They were never going to pull meaningful clicks at position 14 — but they can absolutely earn a citation.
2. Rewrite the top 30% of those pages, not the bottom. 44.2% of all LLM citations come from the first 30% of a page’s text. If your “almost-ranking” page buries the answer below 600 words of context-setting, you’re invisible to retrieval. Move the crisp 40–60 word answer above the fold. Use a question as the H2. Put the data point in the first sentence under it.
3. Build out one canonical sub-topic page per cluster. Instead of trying to expand your pillar page to cover everything, spin off a dedicated 1,200–2,000 word page on the narrowest version of the question you care about. “Best CRM for X” pillar plus six “Best CRM for X if you do Y” sub-pages will out-cite the pillar by a wide margin. Long content earns more citations — pages above 20,000 characters get 4.3× more AI citations than shorter ones — but only when the depth maps to a specific entity or scenario.
4. Add quotations from named operators. Direct quotes lift AI visibility roughly 37%. AI engines preferentially extract attributed statements. Get a sentence from a customer, a partner, or you-the-founder, and slot it into the page. “According to [name], [statement]” is one of the highest-leverage formats you can publish in 2026.
Want this done for you? Paris Roussos runs a flat-rate AI SEO service ($500–$1,500/mo per client, white-label for agencies) covering audits, schema and entity work, AI-visibility tracking, and content engineered to be cited by LLMs. Reach him at parisroussos@gmail.com.
The brands sitting at position one for years aren’t worried yet — they should be.