The most common question Lumio gets from incoming SEO operators: “Is AI search optimization a new discipline, or is it just SEO with new vocabulary?” The answer is “both, and the proportions matter.” About 60% of what you do for traditional Google SEO carries over. About 25% has shifted weight enough that the ratio of effort needs to change. About 15% is genuinely net-new — practices that didn’t exist or didn’t matter when SEO was about ranking ten blue links.
This guide is the operator’s onboarding to AI search. What carries over, what shifted, what’s new. It’s the broader companion to the surface-specific guides like Optimizing Shopify products for ChatGPT Shopping — here we talk about the discipline; there we talk about how to execute it on one specific surface.
The four routes
Traditional SEO had one route: a query enters Google, Google ranks ten links, the user clicks one. Optimization meant ranking higher in that ranked list.
AI search has four routes, and most of them end without a click on the product’s website at all:
Each route has different inputs:
- Route 1 (carousels and overviews) rewards structured data and feed completeness more than backlinks
- Route 2 (citations) rewards content authority and named-entity density — closer to traditional SEO than the others
- Route 3 (synthesized) rewards brand recognition across third- party coverage — closer to PR than to SEO
- Route 4 (voice) rewards concise, conversational answer-shaped content with structured data the assistant can read aloud
The discipline of AI search optimization is choosing which routes matter most for your category and weighting effort accordingly.
What carries over from traditional SEO
The 60%. These tactics work the same way they did in 2020:
Content quality
Long-form, well-researched, accurate content still wins. AI agents that synthesize answers (Route 2 and Route 3) cite content with the same characteristics Google has rewarded for years: depth, specificity, named entities, evidence. The thin-content era is deader than ever in AI search.
Internal linking
A site with strong internal links from cornerstone pages to deep pages still ranks better. AI agents follow internal links to understand topical relationships. The content engine playbook treatment of cornerstone-and-cluster posts maps directly to AI search.
Backlinks (with caveats)
Authoritative backlinks still matter for Routes 1 and 2. The shift: the anchor text matters less, and the topical relationship of the linking domain matters more. A backlink from a category-relevant publication is worth more than a category-irrelevant one with the same DR.
Page experience and Core Web Vitals
Google AI Overviews weight Core Web Vitals into product card ranking. This wasn’t true for organic blue-link SEO until 2021; it’s true now and increasingly so for AI surfaces.
Mobile-first design
Same. Both Google’s and the standalone AI surfaces (ChatGPT, Perplexity) serve heavily on mobile.
HTTPS and basic technical SEO
Same. Robots.txt, sitemaps, canonical URLs, hreflang for international — all still required, all still as they were.
What shifted weight
The 25%. These tactics matter, but the relative effort allocation has changed:
Structured data: from “nice-to-have” to “load-bearing”
In 2020, Schema.org markup was a polish item — present it for
rich-result eligibility, leave it incomplete and you’d still rank.
In 2026, structured data is the gate for Routes 1 and 2. A complete
Product schema with gtin13, brand, category, and full
offers is the difference between a product that surfaces in
ChatGPT Shopping’s carousel and one that doesn’t.
The full implementation is in Product schema for Shopify. The shift in weighting: structured data went from ~10% of the optimization effort to ~25%.
Google Merchant Center: from “paid Shopping” to “free organic”
GMC was historically a paid-channel tool. Then “Surfaces across Google” launched, then ChatGPT’s product index integrated GMC, then Microsoft Merchant Center launched parallel infrastructure. GMC is now data infrastructure, not a marketing channel.
Google Merchant Center setup for Shopify is the implementation; the strategic shift is that any catalog without a validating GMC feed is invisible to roughly half of the AI surfacing surface area.
Identifiers (GTIN/MPN): from “for retail buyers” to “for AI agents”
Universal identifiers were historically a wholesale concern — needed
to sell into Amazon, walmart.com, big-box retailers. AI agents now
treat the presence of gtin13 as a credibility signal. Catalogs
without it compete with their own resellers and lose, even when the
brand owns the canonical product page. The principle: register
GTINs through GS1 and populate them on every product, even private
label.
Page authority: still matters, but distributed
Domain Authority and page-level authority still influence Routes 1 and 2. The shift: AI agents weigh authority across multiple surfaces. A brand with low DR but strong presence on Reddit, niche forums, and review sites can outrank a high-DR competitor for Route 3 (synthesized) recommendations.
This means PR and brand-building work that was historically harder to attribute now show up in AI agent recommendations. The work matters more; the measurement is fuzzier.
What’s new
The 15%. Practices that genuinely didn’t exist in traditional SEO:
Multi-modal content: images parsed for content
Multi-modal models — Gemini, GPT-4V, Claude Sonnet with vision — read product images directly and extract attributes from them. Catalogs with weak text data but strong contextual imagery (worn-on-model for apparel, scaled-with-context for home goods, in-use shots for tools) outperform what their text-only signal would predict.
The new optimization vector: every product image is also a structured data source. Alt text, file naming, structured imagery in JSON-LD all matter; the image content itself matters too. This is genuinely new — Google’s image search existed but didn’t parse image content the way multi-modal models do.
Conversational query intent
Traditional SEO targeted keywords. AI search targets intents expressed as conversational queries: “what’s a good travel stroller for European cobblestones under $400?” doesn’t have a clean keyword target. The AI parses the intent — travel-friendly + cobblestone- appropriate + budget-constrained — and matches against products that have those attributes structured.
The new optimization: structure attributes that map to constraint- based queries (durability, weight, size, use case), not just descriptive keywords.
Brand recognition as the main Route 3 lever
Routes 1 and 2 reward markup and content. Route 3 rewards “the brand that AI agents have associated with the category.” This requires sustained presence across third-party sources: published reviews, mentions in industry coverage, citations in best-of lists. Route 3 traffic plausibly converts higher than Routes 1 and 2 because the brand recognition acts as a pre-decision signal, but it’s the route most decoupled from on-site optimization — measure it in your own catalog rather than borrowing benchmarks.
The new discipline: optimize on-site for Routes 1 and 2 simultaneously with PR/content marketing for Route 3.
Multi-surface measurement
Traditional SEO measured Google rankings. AI search optimization measures presence across Google AI Overviews, ChatGPT product index, Perplexity citations, Claude responses, Gemini recommendations, Microsoft Shopping, voice assistants, and the long tail of vertical AI surfaces. No single tool covers all of them yet.
Lumio’s analytics integration measures referral traffic from each surface separately and triangulates surfacing position by querying agents directly. Pre-AI, no equivalent existed. The measurement discipline is genuinely new and still being built.
What no longer moves the needle
A short list. Practices that worked in 2020 SEO and either don’t help or actively hurt in 2026 AI search:
- Keyword stuffing. AI agent embeddings actively penalize this. A 2010s SEO move; obvious in 2026.
- Meta keywords tag. Long deprecated; mentioned only because it still ships on some Shopify themes.
- Anchor-text manipulation in backlinks. Was the basis of many link-building strategies; now neutral or slightly negative.
- Thin doorway pages. Traditional SEO sometimes built per-keyword landing pages; AI agents collapse these into the canonical version or skip entirely. Programmatic content still works if the content is genuinely different per page; thin variants don’t.
- Hidden text or invisible content. Same penalty as before, plus AI agents trained to detect it specifically.
- Excessive schema (FAQ-stuffing). FAQPage schema was over- applied 2020-2023; Google narrowed which pages trigger FAQ rich results in August 2023. Excessive or non-genuine FAQ markup is plausibly read as a manipulation signal by ranking systems. Use FAQPage schema only on pages where the Q&A content is genuine.
The contrarian take
Most SEO content frames AI search as a paradigm shift that requires rebuilding the discipline from scratch. The framing sells consulting; it doesn’t match the data. AI search is the most demanding evolution of SEO yet — the same fundamental work (content quality, structured data, technical SEO, authority signals) extended into more surfaces with higher precision requirements.
Operators who already do SEO well need to reweight effort, learn the new measurement discipline, and add the multi-modal and brand-route work. They don’t need to start over. Operators who don’t do SEO well will struggle in AI search for the same reasons they struggle in SEO — the underlying data quality work is the same.
The reframe: AI search optimization is SEO with a higher floor.
Where the analogy breaks
- Voice commerce. Route 4 (voice assistants) is genuinely different — the optimization is for spoken answers, not display. Conciseness matters more than depth. This route doesn’t fit cleanly in the SEO framework.
- Vertical AI search. Wayfair AI, real-estate AI, travel AI — each vertical builds its own AI search layer with its own optimization patterns. Generic AI search optimization gets you 80% of the way in each vertical; the last 20% is vertical-specific.
- Conversational follow-ups. A user can ask ChatGPT a follow-up question about a product the agent just recommended. The follow-up answer is generated, not retrieved. There’s no “ranking” to optimize for. The brand presence work for Route 3 is what matters.
What to do this quarter
For an operator coming from traditional SEO who’s adding AI search to the discipline:
- Audit structured data first. This is the load-bearing work. Every other optimization compounds on it. Start with Product schema for Shopify or your platform’s equivalent.
- Set up GMC if you haven’t. This is the second-highest-leverage item. See Google Merchant Center setup for Shopify.
- Run the AI readiness audit. The 6 dimensions framework gives you the prioritization order for the catalog work.
- Add measurement. Pick a tool (Lumio, or build internal tracking) that monitors AI agent recommendations across surfaces. The measurement discipline is what most catalogs lack.
- Don’t abandon traditional SEO work. Backlinks, content, technical SEO still matter. The reweighting goes from “70% of effort” to “50% of effort,” not to zero.
A team that does this work over a quarter typically sees AI surfacing catch up to traditional Google SEO performance. The catalogs that treat it as a separate discipline (rather than an SEO extension) tend to over-engineer, miss the carryover, and slow themselves down.
