AI visibility vs SEO.

Different audiences, different surfaces

The cleanest mental model separates the two by who is reading. SEO targets the human user clicking through a list of blue links. AI visibility targets the language model assembling an answer the human reads instead of the link list. The same brand page may serve both audiences, but the two surfaces ask different things from the page.

A page optimised purely for SEO can be invisible to AI visibility if it strips reasoning to a snippet, hides facts behind JavaScript, or omits source attribution. The same is true in reverse — a long, evidence-rich AI-optimised page can rank below shorter rivals on Google if it doesn't pass the snippet-readability filter. The work is to optimise for both, but never to compress for the human surface at the expense of the AI surface; LLMs need the full reasoning chain to absorb a page.

What carries over from SEO

Most of the infrastructure layer is shared. If your SEO is good, your AI visibility starts from a higher floor:

• Index coverage. ChatGPT search rides on Bing's index. Google AI Overview and Gemini ride on Google's. Perplexity uses a Bing-leaning blend with its own crawler. If your site isn't in the underlying search index, no AI engine can retrieve it. Submit sitemaps to Google Search Console and Bing Webmaster Tools as your first step in either discipline.
• Crawlability. Server-rendered HTML, semantic URLs, fast page load, no broken links. AI crawlers (GPTBot, ClaudeBot, PerplexityBot, OAI-SearchBot) fetch raw HTML and do not execute JavaScript (Vercel 2025 crawler study). Sites built as JavaScript-rendered SPAs are functionally invisible to AI retrieval, even when Googlebot can render them.
• Internal linking and information architecture. Pillar pages, related-content links and clear site hierarchy help both human search and AI traversal. The "topic cluster" model from SEO carries over directly; LLMs use the link graph to disambiguate entities.
• Page-level metadata. Canonical URLs, Open Graph, Twitter cards, dated article:published_time markup. These don't drive AI citation directly but they prevent your pages from being deduplicated incorrectly or stripped of attribution.

Where the two disciplines diverge

Keyword density vs semantic similarity

Traditional SEO has decades of accumulated practice around keyword targeting, density, H-tag hierarchy, exact-match anchor text. AI engines do not read pages this way. The strongest single predictor across published GEO studies is semantic similarity between page content and user query. Write for the actual question, not for a keyword. A page that earns "what is X" citations is one that thoroughly answers what X is in plain language — not one that includes "what is X" 12 times in the body.

FAQ schema as anti-pattern

FAQPage JSON-LD was a high-leverage SEO tactic from 2019 to 2022 in the rich-snippet era. It is now an anti-pattern for AI citation. SE Ranking's 2025 analysis of 129,000 domains and 216,524 pages found FAQ-schema pages averaged 3.6 ChatGPT citations versus 4.2 without. Williams-Cook's controlled DUCKYEA test (Feb 2026) confirmed FAQPage schema confers no extraction advantage over visible Q&A copy. The Aggarwal et al. KDD 2024 paper did not include FAQ format among its nine measured tactics — its absence from the validated list is itself informative.

JSON-LD as a Bing/Copilot bonus, not a universal AI signal

Schema.org JSON-LD remains genuinely useful for SEO (rich-snippet eligibility) and Bing/Copilot (Fabrice Canel publicly confirmed at SMX Munich in March 2025 that Bing uses schema for Copilot LLM grounding). It does not transfer to ChatGPT or Perplexity. SearchVIU's 2025 5-system × 8-scenario × 10-query study found 0 of 5 systems extracted price data placed exclusively in JSON-LD. Williams-Cook's fake-schema test showed ChatGPT and Perplexity tokenise schema as plain text without structural parsing. Keep JSON-LD where it earns rich-snippet placement; do not expand it as your primary AI-citation tactic.

Backlinks vs third-party citations

SEO measures the link graph — how many domains link to yours, what their domain authority is, what anchor text they use. AI visibility measures the citation graph — how often third-party sources name your brand in evidence-bearing language. The two correlate but are not identical. A Reddit thread that mentions your product without linking to your site contributes nothing to PageRank but materially to ChatGPT's brand recognition. Published research finds third-party citation roughly 6.5× more effective than self-citation for AI visibility — a stronger lever than any on-page SEO tactic.

The 83%-from-outside-top-10 problem

A long-standing SEO assumption is that ranking in the top 10 organic results captures most of the available value. For Google AI Overview specifically, this is wrong. Roughly 83% of AI Overview citations come from pages outside Google's top 10 organic results. The AI surface samples a much wider candidate set than the human surface, weighted by evidence density rather than rank. Brands that placed all their bets on top-10 organic ranking have measured this effect as a cliff: their PageRank-validated pages are not the ones AI is citing.

Length sweet spot is different

SEO best practice often advises shorter, scannable content optimised for click-through. AI engines reward depth: 1,000–3,000 words with 10+ headings is the published sweet spot. Low-cited pages average 170 words; high-cited pages average ~2,000 — a more-than-10× gap. Padding for length alone does not help, because each chunk must be independently useful, but truly information-dense long-form material outperforms the shorter SEO ideal in citation rate.

Traffic vs influence — why <1% referral still matters

AI search referral traffic is small by SEO standards. Aggregate measurements in 2026 put AI assistants at less than 1% of total referral traffic globally. This is the number that makes some teams dismiss GEO as a marginal concern. The number that makes them reconsider:

The right framing is brand visibility, not session volume. Most of the value is indirect: brand-search lift after first AI exposure, direct-traffic lift, lower sales-call friction, customers citing AI recommendations during procurement, faster B2B vetting cycles. The AI platform is where buyers first form a view; the conversion happens later — usually on the brand's own search result, direct visit, or sales call. SEO measures the conversion event. AI visibility measures the upstream consideration-set inclusion that makes the conversion event likely.

Should I stop doing SEO?

No. SEO continues to drive the bulk of measurable referral traffic for most brands, and the underlying index coverage on Bing and Google is the gate to AI visibility itself. The pragmatic frame:

• Keep core SEO infrastructure — sitemaps, indexing, semantic URLs, canonicals, page speed. This is shared cost.
• Reduce or stop FAQ-schema work and JSON-LD expansion as primary AI tactics. Maintain JSON-LD where it earns rich-snippet placement; don't add new JSON-LD chasing AI citation.
• Add the AI-specific layer: llms.txt, granular five-bucket robots.txt, Markdown alternates, IndexNow. See AI crawler readiness for the implementation guide.
• Shift content investment from short snippet-optimised pieces toward 1,000–3,000-word reference content. Encyclopedia-style explainer pages outperform news pages by ~3× per citation.
• Invest in third-party citation work — Wikipedia, Reddit, Hacker News, vertical-industry pubs. This is the highest-leverage AI-visibility activity and has no direct SEO equivalent (it carries weak link-graph value if no link is included).

What changes in China

Mainland-Chinese consumers do not use ChatGPT, Claude, Perplexity or Gemini. They use DeepSeek, Qwen and Doubao, with secondary engines including Yuanbao, Kimi and ERNIE Bot. The SEO equivalents (Baidu, Sogou, 360 Search) have weaker leverage on these AI engines than Bing/Google have on Western AI. The third-party citation graph is a different stack entirely: 百度百科 not Wikipedia, 知乎 not Reddit, 小红书 not YouTube, SMZDM not RTINGS.

For Western brands operating in China, the practical consequence is that traditional China SEO (Baidu / Sogou ranking) is not a substitute for China AI visibility work. The two run in parallel. See China AI visibility for global brands, GEO for China, and how to get cited on Chinese AI.