Ranking on Google no longer gets you into the AI answer.

Why your rankings and your citations stopped matching.

For most of the last decade, optimizing for Google was optimizing for everything. If you ranked first, you got the featured snippet, the voice answer, and the traffic. One target, one strategy.

That stopped being true somewhere in 2025. As the PR executive Ronn Torossian put it when the data first landed: when the overlap between Google's top results and AI citations was around 70 percent, optimizing for Google was effectively optimizing for both — at under 20 percent, that thinking is broken.

The mechanical reason is that AI search is not a ranking function at all. It is a retrieval pipeline. When you ask an AI a question, it does not paste your question into a search box. It breaks the question into smaller sub-queries, searches for each one separately, retrieves candidate passages, scores them, and then synthesizes an answer with citations attached. The decision about whether to cite you happens during retrieval, at the passage level — not at the page level, and not against your domain authority.

This is why a page can rank tenth on Google and get cited constantly, and why a page can rank first and never get quoted. What gets retrieved is a self-contained chunk of text that directly answers a specific sub-question, in language that reads like a reference document rather than a sales page.

The Princeton study, and the three things that reliably work

The foundational research on this is a peer-reviewed study out of Princeton, Georgia Tech and IIT Delhi — Aggarwal et al., presented at KDD 2024 — which tested optimization strategies across 10,000 queries in 25 domains and measured the visibility lift for each.

The findings have held up remarkably well, and they are unglamorous. Adding direct quotations from named sources lifted AI visibility by up to 41 percent. Adding statistics lifted it by up to 32 percent. Citing your sources lifted it by up to 30 percent. Improving fluency lifted it by up to 28 percent. Keyword stuffing, unsurprisingly, did nothing.

Read those four together and a pattern emerges: AI engines cite content that behaves like a source. Specific numbers, attributed claims, named references, clean prose. The kind of writing that would survive an editor. "Conversion rates improved 47 percent after the redesign" gets cited; "we deliver outstanding results" does not, and never will, no matter how many times you publish it.

That is the entire reason this page — and every page on this site — is written with sources attached to the numbers. It is not academic habit. It is the format that gets quoted.

Being cited matters even when nobody clicks

Here is the uncomfortable part. An analysis of 68,879 Google searches by Chapekis and Lieb found that only about 1 percent of users who encounter an AI summary click a cited source, against roughly 15 percent who click a traditional result when no AI summary appears. Seer Interactive measured organic click-through falling around 61 percent on queries where an AI Overview appears, across 3,119 informational queries at 42 organizations.

So if you measure AI search purely in sessions, it will look like a losing proposition. That is the wrong frame. Three things are true simultaneously:

• The clicks you do get are extraordinarily valuable. Conversion rates of 10 to 16 percent against organic's 1.76 percent are not a rounding error. The person arriving has been pre-sold by a source they trust.
• Citation influences decisions without a click. If ChatGPT tells someone planning a Kauai trip that three specific operators are the ones to look at, that shortlist is formed whether or not anyone clicks. You are either on it or you are not.
• Citation lifts your traditional results too. BrightEdge found pages cited in AI Overviews earn roughly 35 percent more organic clicks than uncited competitors on the same page. Being in the answer makes your blue link more attractive, not less.

The part that is genuinely hard: third-party mentions

The single most awkward finding in this whole field is that your own website is not the main lever. Ahrefs' 2026 research found external brand mentions correlate with AI Overview appearances at about 0.664 — the strongest signal they measured. AirOps found brands are roughly 6.5 times more likely to be cited through third-party sources than through their own domain.

Reddit alone accounts for something like 46.7 percent of Perplexity's citations. Forums, review platforms, local news, industry publications, and community discussions carry disproportionate weight because an AI engine treats them as independent corroboration in a way it will never treat your About page.

This is where SEO stopped being a first-party game. You can control your own site completely and still be invisible, because the engines are asking what other people say about you. For a Hawaii business with no press coverage and no forum presence, that is a real structural problem — and it is the specific gap our 370-plus domain network is built to close, since we can place a client into genuine category context on properties we already operate.

Freshness is a ranking factor again, and it is aggressive

Traditional SEO taught everyone that a good page could rank for years. AI citation does not work that way. Analyses through 2026 found content updated within the last 30 days earning several times more AI citations than stale content, with the large majority of AI Overview citations drawn from recent material.

Perplexity is the extreme case: it weights freshness at roughly 40 percent of its ranking signal, serves results around 3.3 times fresher than Google, and actively deprioritizes content older than 30 days for medium-velocity topics. A page that has not been touched in a year is, to Perplexity, close to invisible regardless of quality.

The practical implication for a small business is a maintenance rhythm rather than a publishing sprint. A handful of genuinely important pages, reviewed and updated quarterly with current numbers and dates, will outperform forty pages published once and abandoned.

Measurement, and the mistake nearly everyone makes

Analytics does not handle this well. Out of the box, GA4 buckets AI engine referrals into Direct or (none) — which means most businesses running AI search work are measuring approximately none of it, and concluding it does nothing.

The fix is unglamorous but necessary: build referral segments for the AI sources explicitly, track branded search volume as a proxy for shortlist influence, and run a recurring prompt audit — literally asking the engines the questions your customers ask and recording whether you appear. That last one is the metric that matters most, and it is the one the tool above is a starter version of. We run a full version quarterly for clients across ChatGPT, Perplexity, Gemini, and AI Overviews.

The industry term for the underlying metric is Share of Model: how often your brand appears in AI responses for a defined set of prompts, relative to competitors. It is the AI-era replacement for share of voice, and unlike rankings, it cannot be checked with a rank tracker.