Page 2 is dead. For LLMs, it never existed.

For years, the SEO conversation centred on climbing the ladder. Position 5 to position 3. Position 3 to position 1. Every incremental rank gain meant a measurable click increase, and you could plot it on a nice, smooth curve. The gradient was predictable. It rewarded persistence.

LLM citation doesn't care about your gradient.

When we analysed 3,323 prompts across ChatGPT, Perplexity, Gemini, and Google AI Overviews, citation frequency held steady across positions 1 through 8, then dropped off a ledge somewhere around position 8 to 10. Pages at position 12? Functionally invisible. Position 15? Might as well not exist.

The mechanics behind the cliff

Retrieval-dominant models like Perplexity and Google AI Mode run a live search when you ask a question. They pull from whatever the index surfaces in real time, usually the top results. Seer Interactive's analysis of 500+ citations found that 87% of SearchGPT citations match Bing's top 10 organic results. Perplexity operates on its own proprietary index of over 200 billion URLs, but the logic is similar - it retrieves from a pool, and that pool has a hard boundary.

Training-dominant models like Claude and ChatGPT (in its non-search mode) work differently. They've already ingested their training corpus, which was itself assembled from the most visible, most-linked, highest-ranking content on the web. If your page never cracked the top results during the training window, it isn't in the model's memory at all.

Pages below that threshold are excluded from citation pools entirely.

Ahrefs' latest study across 863,000 keywords and 4 million AI Overview URLs complicates this further. In July 2025, 76% of AI Overview citations came from pages ranking in the top 10 organic results. By early 2026, that number had dropped to 38%. Google's switch to Gemini 3 as the default AI Overview engine introduced a query fan-out process that splits your original search into multiple sub-queries, then cites pages that surface frequently across those sub-results. Citation pools now draw from top results across clusters of related queries.

Topical authority across clusters is where citation eligibility now lives. Rank in the top 8 to 10 for a spread of related terms, and you're in the pool. Sit at position 14 across all of them, and you're still invisible.

Marginal gains stop working

Traditional SEO rewards marginal gains. Moving from position 6 to position 4 on a high-volume keyword could mean thousands of extra clicks per month. The ROI on that improvement was always clear and measurable.

For LLM citation, that same jump from position 6 to position 4 does almost nothing. Both positions are inside the pool. Both get cited at roughly the same frequency. The effort you'd spend pushing from 6 to 4 would be far better spent getting your position-11 pages above the threshold.

Striking distance redefines itself here. In traditional SEO, striking distance meant positions 4 to 20 and the goal was to improve rank for more clicks. For LLM citation, the striking-distance window that matters is positions 8 to 12 - pages right on the edge of the cliff. A small rank improvement there has an outsized impact on whether an LLM ever sees your content.

A single position shift at 9 to 11 moves you from invisible to cited. Pages stuck at position 14: the effort to move them six or more positions might not be worth it unless the topic has strong LLM query volume. Pages comfortably at position 2 or 3 can wait - your time is better spent elsewhere.

The practical playbook

Pull your Google Search Console data and find every page ranking between positions 8 and 14. That's your LLM striking-distance list - pages that are close to the pool but not quite in it. Refresh, update, and strengthen those pages before you touch anything else.

Think in topical clusters rather than individual keywords. With Gemini 3's query fan-out pulling from sub-query results, a single page ranking position 6 for one keyword might not be enough. You want to rank in the top results across a cluster of related queries so your domain shows up repeatedly in the fan-out process. That repeated visibility is what gets you cited.

For LLM citation, position 1 and position 5 are functionally equivalent. If your strategy is built around LLM visibility, deploy resources on breadth - getting more pages above the threshold - and on building topical coverage across clusters. For traditional click-through rates, yes, position 1 still matters enormously, but the ceiling on LLM citation gains from pushing position 3 to position 1 is low. That effort compounds better elsewhere.

Watch Perplexity separately. Seer Interactive found only 11% of domains are cited by both ChatGPT and Perplexity. Perplexity runs real-time retrieval across its own 200+ billion URL index, which means your Perplexity visibility is driven by entirely different signals to your Google rankings. Track both.

The threshold shifts too. The Ahrefs data showing the drop from 76% to 38% top-10 overlap happened in under seven months. Google's underlying model changes, Perplexity's index grows, and ChatGPT's training data gets refreshed. The cliff might move. Building topical authority gives you resilience against those shifts, because broad cluster coverage holds its position across model updates in ways that optimising a single keyword never could.

What this means for the content operation

The biggest practical implication is how you allocate effort. When you're running a lean operation, you can't do everything.

The traditional priority order was: find high-volume keywords, create content, optimise for position 1, build links. The LLM-aware priority order looks more like: map topical clusters, create comprehensive coverage across each cluster, get every piece above the position 8 to 10 threshold, and refresh anything that's slipping below it.

For teams of one or two people, getting this right means the difference between showing up in AI-generated answers - where an increasing share of your audience is now looking for information - and being invisible.

FlipAEO's analysis found that 30% of ChatGPT citations come from sites that don't appear in Google's top results at all (methodology: citation tracking across a sample of ChatGPT responses mapped against Google SERP data - worth verifying directly at flipaeo.com). So while the cliff effect is real for retrieval-dominant models, there's a parallel game happening with training-dominant models where brand authority, third-party mentions, and content format all play their own role in whether you get cited. The cliff is the starting point for thinking about this, not the whole picture.

If you're running a small content operation and need one actionable framework to start with: find your pages between positions 8 and 14, and get them above the cliff.