AI SEO for ecommerce: the 2026 playbook

What changed for ecommerce SEO

Three shifts pushed AI surfaces from curiosity to line item. ChatGPT folded shopping affordances into the default consumer experience, so more product queries resolve inside the chat. Google AI Overviews expanded category coverage and names products when the source page carries clean schema. And Perplexity's shopping surface established its own citation logic — typically 3–7 sources per answer, with review media often outranking PDPs. As of mid-2026, AI-mediated discovery is large enough that brands see revenue swings from a single PDP rewrite. The umbrella discipline is GEO; this guide is the ecommerce-specific sequence. The measurement unit throughout is the SKU — the case for that is the product AI visibility pillar.

Move 1 — Inventory the catalog against AI engines

Start with a one-time audit: for each SKU, which of the six engines cite it on its primary intent queries? Intent queries are what a shopper types — not "Acme Coffee Company" but "best medium-roast beans for a moka pot." Expect uncomfortable findings: hero SKUs with strong Google rankings and zero ChatGPT citations, and long-tail SKUs that quietly outperform the hero in one engine. That misalignment is where the work starts. The free SKU visibility grader samples five products in about 90 seconds; SKU-level tracking does the full catalog on a schedule.

Move 2 — Fix Product JSON-LD gaps

Most Shopify themes emit partial Product schema. The recurring blanks in eCommerce Insights audits: gtin13, mpn, additionalProperty for material and color, and priceValidUntil on the offer. Each is a confidence signal an engine reaches for before citing a product. This is the single highest-leverage technical fix — cheap, one-time, and it closes the distance between "the site has the data" and "an AI can read the data." Field-by-field instructions and a complete working block are in schema for AI search; the product schema generator drafts one from any PDP URL.

The PDP that ranks in Google isn't automatically the one that wins in ChatGPT. Different surfaces reward different completeness.

Move 3 — Verify crawler access

Check robots.txt and any bot-protection layer against the AI crawler set: GPTBot, OAI-SearchBot, ChatGPT-User, PerplexityBot, ClaudeBot, Google-Extended. OpenAI documents its crawlers and IP ranges publicly at platform.openai.com/docs/bots. Merchants block these by accident more often than by policy — usually a security update that tightened bot rules. If crawler hits are zero four weeks after publishing a product, something is blocking.

Move 4 — Seed review-source coverage

Engines cite third-party review pages at roughly the rate they cite owned pages, sometimes higher. A brand with zero Reddit threads, zero category-site reviews, and zero teardown videos is invisible on the most citable surface. The work is old-school D2C: pitch one category review site per quarter, seed one detailed teardown video per flagship SKU, answer a few relevant Reddit threads monthly with disclosed affiliation. The novelty is that AI engines now turn this into citations that ride into every buyer's chat window.

Move 5 — Publish or upgrade llms.txt

An llms.txt gives AI crawlers a curated, annotated map of the store — homepage, top collections, flagship PDPs, policies. It is a community convention rather than a standard, but crawler-log observations show the major bots fetching it where published. Shopify needs a hosting workaround; the full walkthrough with a working template is llms.txt for Shopify, and the generator produces a first draft in under a minute.

Move 6 — Rewrite the thinnest PDPs

Every catalog has ten to fifty PDPs written in a rush — thin descriptions, no material spec, no use-case copy. Engines cannot cite a page that does not answer the shopper's question. A rewrite needs the product's material, dimensions, use cases, compatibility, care, and one specific reason to pick this SKU over the next — stated in the first 150 words, in sentences an engine can lift verbatim. Ten PDPs a week with a senior editor reviewing moves a hundred-SKU catalog in two months. Patterns and before/afters are in optimize content for AI search. Do not auto-publish; tone drift compounds fast — which is why eCommerce Insights ships diffs a human approves.

Move 7 — Monitor weekly, per product

Build a battery of 20–50 intent queries per product line and run them weekly against the engines. Track citations gained and lost, competitors appearing alongside, and week-over-week deltas — per SKU and per engine, never as a single average. This is the standing rhythm that turns the previous six moves from a project into a program; the cadence and reporting structure are covered in GEO strategy for D2C brands.

What is still unsettled

Honest caveats, as of mid-2026: the relative weight of reviews versus body copy shifts quarter to quarter; per-engine source preferences move with product releases; llms.txt reading behavior is observed but not documented by any engine. Run the seven moves, measure what changed, and discard tactics that stop working — the full settled-versus-forming inventory is in AI content optimization.