The New SEO Playbook for AI Search: 8 Practices, Backed by 1 Billion Data Points

AI search rewrote the SEO rules. Eight practices that replace the old playbook — from 'optimize each AI surface separately' to 'stop adding schema for AI' — each backed by 2026 data, and honestly graded.

By Andrej Ruckij · · 8 min read

The New SEO Playbook for AI Search: 8 Practices, Backed by 1 Billion Data Points

By Andrej Ruckij · June 9, 2026

TL;DR: AI search didn’t tweak SEO — it changed what works. Across 14 large-scale studies and more than a billion data points in 2026, a new playbook is visible: optimize each AI surface separately (a #1 Google ranking no longer means you’re cited), put YouTube and earned mentions ahead of backlinks, stop adding schema markup for AI (the best-designed study found it does nothing for citations), get into “Best X” lists, and write for the citation, not the crawl. Below are the eight practices that replace the old rules — each with the data behind it, and an honest note on how solid that data is. The short version of the caveat: most of these numbers come from one vendor (Ahrefs) and are correlational. Use the direction with confidence; quote the decimals with care.

Search “AI SEO tips 2026” and you’ll get a hundred listicles repeating the same advice — much of it left over from the pre-AI playbook. This is the version that starts from the data and is willing to tell you which of that advice is now actively wrong.

First, the stakes. An AI Overview now cuts clicks to the #1 organic result by 58% — up from 34.5% just eight months earlier. The traffic is leaving, and it’s leaving faster. The response isn’t to optimize harder for a shrinking pie; it’s to change what you optimize for. Here’s how.

1. Optimize each AI surface separately — ranking ≠ being cited

Old rule: rank #1 on Google and you win everywhere. What the data says: the AI surfaces barely overlap. Google’s AI Mode and AI Overviews reach the same answer 86% of the time but share only 13.7% of their citations. Only 38% of AI Overview citations come from pages in Google’s top 10 — and 28% of ChatGPT’s most-cited pages rank for zero Google keywords at all. ChatGPT runs on a partly separate index. New rule: treat ChatGPT, AI Mode, and AI Overviews as three distinct targets with their own winners. A win on one does not transfer. This is the single biggest break from classic SEO — and the reason “just rank higher” is no longer a strategy.

Old rule: backlinks and Domain Authority drive everything. What the data says: ranked against AI visibility, the strongest correlate isn’t a link metric — it’s YouTube mentions (0.737), followed by unlinked brand mentions across the web (0.66–0.71). Domain Rating limps in around 0.3, backlinks weaker still, and raw page count correlates with essentially nothing (0.19). New rule: invest in being talked about — especially on YouTube — before you invest in another link-building sprint or content-volume push. PR is now SEO. (Honest flag: this is correlation, and big brands are on YouTube and get cited, so some of that 0.737 is confounded. But the direction — earned, third-party presence beats the link graph — has independent experimental support.)

3. Stop adding schema markup for AI. It does nothing.

Old rule: schema markup is “non-negotiable” for AI visibility. What the data says: this is the one finding backed by a real experiment, not just a correlation. Measuring 1,885 pages that added schema against 4,000 matched controls, the effect on AI citations was nil — AI Overviews −4.6%, AI Mode +2.4%, ChatGPT +2.2%, all at or near statistical noise. AI crawlers appear to read your visible HTML and largely ignore the JSON-LD. New rule: keep schema for what it actually does — traditional rich results, and author/publisher markup for entity verification (a different mechanism). Stop budgeting it as a GEO lever. This is the rare case where the data says do less.

4. Get into “Best X” lists — and rank high within them

Old rule: publish more blog posts to build authority. What the data says: the single most-cited content format in ChatGPT is the “Best [category]” listicle — 43.8% of its cited page types. AI strongly prefers third-party comparative judgment over your own marketing copy, and brands in the top third of a list are likelier to be the one it recommends. Freshness matters: most cited lists were updated within the last year. New rule: own a genuinely useful “Best X” in your category and run outreach to be placed — high — in the third-party lists AI already cites (G2, Clutch, niche review sites). This is digital PR, not content marketing. (Yes, this very article is a best-practices list. That’s not an accident.)

5. Write for the citation, not the crawl

Old rule: stuff the title with keywords; nobody reads URLs. What the data says: being retrieved by an AI engine isn’t being cited — ChatGPT cites only about half the pages it fetches; the rest feed the answer as uncredited background. What predicts which half gets the citation? How well your title matches the user’s actual question (cited pages scored markedly higher on semantic match) and clean, readable URL slugs (cited ~90% of the time vs ~81% for opaque URLs). New rule: write titles the way users actually ask, and use human-readable URLs. These are small, cheap changes that move the needle on the gate that now matters — citation, not just retrieval.

6. Stop chasing word count

Old rule: longer, “comprehensive” content ranks better. What the data says: the correlation between word count and AI citation is 0.04 — effectively zero. Over half of cited pages are under 1,000 words. New rule: answer the question directly and stop padding. Length is not a quality signal to AI; clarity and a direct answer are. The 2,500-word “ultimate guide” reflex is a cost with no measured payoff in citations.

7. Measure topic-level share of voice, not single rankings

Old rule: track your keyword positions. What the data says: AI answers are volatile on the surface, stable in meaning. An AI Overview changes every 2.15 days on average, with 70% of the content differing between checks and nearly half its citations churning — yet the meaning stays 95% consistent. Chasing a citation on one query is chasing noise. New rule: measure your presence at the topic level — your share of voice across a category of questions over time — not your appearance on any single query on any single day. (This is sometimes called “share of model.”)

8. Defend your money pages — and accept the informational hit

Old rule: informational top-funnel content is your traffic engine. What the data says: the AI-Overview hit is concentrated, not uniform. A striking 99.9% of AI Overviews appear on informational queries; shopping triggers one only 3.2% of the time, transactional 2.1%, navigational under 1%. The collapse is landing squarely on “what is / how to” content. New rule: expect your informational top-funnel traffic to keep migrating into AI answers — and lean into the bottom funnel, which still clicks. Transactional and commercial pages remain comparatively protected. Don’t defend the informational traffic that’s structurally leaving; defend and grow the pages where the click still happens.

How solid is all this? (read before you quote a number)

The honest part most playbooks skip. These findings come from one vendor (Ahrefs), are self-published, and are mostly correlational — built on its search index and Google Search Console aggregates. That’s stronger than the usual blog guesswork (real, disclosed sample sizes in the millions) but it isn’t peer-reviewed, and it hasn’t been independently replicated.

Two anchors keep it grounded:

  • The schema null (Practice 3) is the strongest result — the only one from a controlled, before-after experiment. Trust it most.
  • The CTR-collapse direction has institutional backing. Beyond the vendor’s “58%,” the Pew Research Center measured real browsing behavior and found AI summaries roughly halve clicks to traditional results — independent corroboration of the trend, if not the exact decimal.

So: act on the directions with confidence — they’re consistent across studies and, where checkable, with outside research. Quote the specific percentages with a caveat. And distrust the weakest figure in the set — the claim that “ChatGPT is 12% of Google’s volume,” which hinges on a search-intent classification the original researchers would put far lower.

The playbook in one line

Stop optimizing a single page to rank on a single engine. Start building earned, third-party presence (especially on YouTube and in “Best X” lists), write for the citation rather than the crawl, drop the tactics that don’t move it (schema-for-AI, word count), and measure topic-level share of voice across the separate AI surfaces — while defending the bottom-funnel pages where clicks still live.

Sources