The five pillars of eCommerce SEO, rebuilt for the AI-search era

Back in 2018 I wrote a long primer that broke eCommerce SEO into four pillars: technical, authority, relevance and experience. Re-reading it now is a strange experience. The structure is exactly how I still think — I've never found a cleaner way to hold the whole discipline in your head at once — but half the specific advice has aged badly. It talks about AMP. It cites desktop search share. It doesn't mention Core Web Vitals, E-E-A-T, search intent or AI Overviews, because in 2018 most of those either didn't exist or didn't matter.

So this isn't a tidy-up. It's a rebuild on the original foundations. The four pillars stay, because they've earned it. Everything underneath them is current to 2026, and the parts that were quietly wrong have been taken out rather than dressed up.

Why I still think in pillars

SEO is a field that rewards people who can do two opposite things at once: zoom all the way out to strategy, and all the way in to a single misfiring tag. Most people are good at one and bad at the other. The pillars exist so you never lose the map while you're down in the weeds — every task you do belongs to one of five buckets, and if you can't say which, you probably shouldn't be doing it yet.

They aren't separate workstreams. They feed each other constantly: faster pages improve experience, which improves the signals that build authority; sharper relevance makes your technical work actually convert. I treat them separately only because it's the only honest way to explain them. In practice you work all four in parallel, and the art is knowing which one a particular site needs first. A brand-new store, an ageing one carrying years of bad SEO, and a site in a brutal vertical all need a different pillar leading.

The five pillars at a glance

Here's the whole framework in one view before I go deep on each. Four pillars are the originals, refreshed; the fifth — AI search — is new, and I've set it apart below because it behaves differently from the rest. If you take nothing else from this piece, take these five buckets and the idea that none of them is optional.

Pillar 01Technical & on-page

The technical pillar answers one unglamorous question: can a machine reach, render and understand every page you care about? If the answer is no, nothing else in this article matters — you can have the best content in your category and it will sit unseen. In 2018 this mostly meant crawlability and clean tags. In 2026 the "machine" is no longer just Googlebot; it's also the crawlers behind AI Overviews, ChatGPT and Perplexity, and they are less forgiving of a messy site than a human with a back button.

Keyword research is still the most under-invested hour you'll spend

I said this seven years ago and I'll say it louder now: thorough keyword research is the single most ignored, under-funded part of most SEO programmes, and it quietly decides the ceiling of everything else. The difference today is that I no longer research keywords so much as intent and entities. A tyre shop doesn't rank by repeating "tyres" — it ranks by mapping the real questions a buyer has at each stage, and the long-tail terms attached to them: winter tyres for a Volvo XC60, part-worn tyres legal limit, mobile tyre fitting Glasgow. Less competitive, far higher intent, and far more likely to convert.

The tooling has consolidated. The stack I actually use looks like this:

• Ahrefs or Semrush for the addressable-market view — total search demand, competitor gaps, and the long-tail clusters worth owning.
• Google Search Console as the truth source, because it shows the queries you already rank for and the ones you're one nudge away from.
• The SERP itself, in incognito, read like a brief: the actual intent is whatever Google has chosen to rank, not what a volume number implies.
• Screaming Frog for the crawl — the same workhorse it was in 2018, still the fastest way to see your site the way a bot does.

Architecture: still three clicks, now also for agents

The old rule holds — no important page should be more than about three clicks from the homepage, and your structure should scale without bolting on a new layer every time you add a category. A clean hierarchy of category → subcategory → product, with internal links that follow how people actually shop, does double duty now: it helps human navigation and gives an AI crawler a legible map of how your catalogue fits together. Flat, orphaned, or infinitely-faceted URL sprawl is where crawl budget goes to die.

Structured data is no longer optional decoration

This is the biggest single change to the technical pillar. Seven years ago schema markup was a nice-to-have that won you the occasional rich snippet. Today it's the spine of how both Google and AI systems decide what a page is and whether to trust it. For eCommerce, Product markup with price, availability and review data isn't garnish — it's how you stay eligible for the visual, AI-summarised results that increasingly sit above the old blue links.

"@type": "Product",
"name": "...",
"image": ["..."],
"offers": {
  "@type": "Offer",
  "price": "129.00",
  "priceCurrency": "GBP",
  "availability": "https://schema.org/InStock"
}

The one rule that matters more than any field: the markup must match what's visibly on the page. Schema that claims a review score the page doesn't show, or a price that isn't real, is the fastest way to get penalised — and the fastest way to lose an AI system's trust, which it does not give back easily. I went deep on exactly this honesty problem in the 120-point AI readiness framework.

The audit that finds the rot

Everything technical eventually comes back to a crawl-based audit. The 2018 checklist still mostly applies, with the dead items removed:

• Crawl the whole site and fix what surfaces: broken links, redirect chains, orphan pages, missing or duplicate titles and descriptions.
• Check indexation honestly in Search Console — not "are pages indexed" but "are the right pages indexed", and is thin, near-duplicate product sprawl being kept out.
• HTTPS everywhere, as a given. This was advice in 2018; in 2026 it's table stakes and its absence is disqualifying.
• Kill duplicate and missing metadata — still one of the most common faults I find on big catalogues, and still a fast, cheap win.
• Forget AMP. Google dropped its AMP requirement for Top Stories years ago; a fast, well-built responsive page beats it on every axis. If you're still maintaining an AMP variant, retiring it is usually a simplification, not a risk.

Pillar 02Authority

Authority is whether the rest of the web — and Google's quality systems — treat you as a source worth believing. The mechanics are the same as they always were: what you publish on your own site, and what others say about you elsewhere. What's changed is how Google frames it. The old E-A-T — Expertise, Authoritativeness, Trust — gained an extra E for Experience at the end of 2022, and that addition is a genuine signal about intent, not a cosmetic tweak.

"Experience" means first-hand knowledge. For eCommerce that's the difference between a product page that recites the manufacturer's spec sheet and one that reflects having actually handled, sold and supported the thing — sizing notes, what it's compared against, the question support gets asked every week. Google is increasingly good at telling those apart, and AI systems lean on the same signals when they decide whose page to summarise.

Links still matter — but earned, not built

Link building hasn't died; the 2018 version of it has. Volume plays, link networks and "100 backlinks" packages are liabilities now. A handful of genuinely earned links from sites that are themselves authoritative — real publications, industry bodies, suppliers, the occasional .edu or .gov — outweighs hundreds of mediocre ones. The reliable way to earn them in 2026 is digital PR: produce something genuinely worth citing — original data, a useful tool, a strong point of view — and the links follow because people actually want to reference it.

• Be the expert on one or two things first. Depth in a narrow area builds authority faster than shallow coverage of everything, and it's far harder to fake.
• Audit your existing link profile and disavow or distance the toxic, spammy legacy links — old bad SEO is a real anchor on a lot of established stores.
• Treat your own blog as an authority engine, not a keyword dumping ground. Thought-leadership, primers and genuinely useful guides are what get cited; thin SEO filler now actively hurts under Google's helpful-content systems.

Pillar 03Relevance & intent

Relevance is the pillar people most often get wrong, because they confuse it with keywords. Relevance isn't whether your page contains the words someone typed — it's whether the page resolves the intent behind those words. Google has spent the better part of a decade getting good at understanding meaning rather than matching strings, and the sites that win are the ones that answer the real question fully and quickly.

Practically, that means starting every page from intent, not vocabulary:

• Classify the intent behind each target query — is the searcher trying to buy, compare, or just understand? A comparison query met with a product page is a mismatch, and Google reads the bounce.
• Deliver on the promise immediately. Whatever the title or ad promised, the answer should be visible above the fold, not buried under throat-clearing. This is "top-loading", and it matters double for AI extraction.
• Go deep with long-tail and entities. Detailed, specific pages convert better and rank more easily. A 10% conversion on 1,000 well-targeted visitors beats 1% on 5,000 mistargeted ones — that maths was true in 2018 and it's truer now that traffic is harder to win.
• Write meta descriptions as invitations, not crawler bait. They rarely affect ranking directly, but they decide the click — and a click-then-stay is itself a relevance signal.

One genuinely new tool here is AI itself: I use it to pressure-test whether a page actually covers a topic's important sub-questions, and to find the entities a thin page is missing. It's a research assistant, not a writer — the judgement and the first-hand experience still have to be yours, or you're back to the thin filler Google now punishes.

Pillar 04Experience

The experience pillar has changed more than any other since 2018 — not in principle, but in precision. Back then "user experience" was a soft idea you gestured at. Now Google measures it with named, public thresholds: the Core Web Vitals. They're the clearest example of the whole field maturing from vibes to numbers, and they're a direct ranking input.

Three things to know about that row. LCP is how fast your main content appears. INP — interaction to next paint — replaced the old First Input Delay metric in March 2024, and it's stricter: it measures how responsive the page feels across every interaction, not just the first tap. CLS is visual stability — the reason every image needs explicit dimensions so nothing jumps as the page loads. And they're measured on real users in the field, not a friendly lab run, which is why a store that feels fine on your office Mac can quietly fail on the mid-range phones most customers actually hold.

Mobile isn't a variant of the experience pillar; it is the experience pillar. Google indexes the mobile version of your site as the primary one, and the majority of eCommerce sessions happen on a phone, often on a patchy connection. A responsive design that merely shrinks the desktop layout isn't enough — the mobile experience has to be designed: legible type, tap targets you can actually hit, a checkout that doesn't fight your thumbs.

Pillar 05AI search

For two decades I resisted adding a pillar. Four was clean, four was complete, and every new trend I was tempted to promote turned out to be a tactic that belonged inside one of the existing four. AI search is the first thing in twenty years that genuinely doesn't fit — so rather than force it into the technical or relevance pillar, I've given it its own. Here's the argument for why it earns that, because I don't want you to take it on faith.

The other four pillars all answer variations of one question: can you win the classic search result? Can you be crawled, be trusted, be relevant, be fast. AI search changes the question itself, because it changes where the answer is delivered. When AI Overviews, ChatGPT or Perplexity resolve a query inside their own interface, the ranking you fought for can be technically perfect and commercially invisible — the user got their answer and never came to your site. A growing share of searches now end with no click at all. That's not a tactic inside another pillar; that's a different finish line, and a different discipline for reaching it.

The honest tension — and the reason I held off so long — is that this pillar is downstream of the other four. You cannot be cited by an AI if you're not crawlable, not trusted, not relevant and not extractable. So it sits on top rather than alongside. But it's now load-bearing enough, and distinct enough in the work it requires, to be its own pillar rather than a footnote on the others:

• Top-load the answer. AI extractors reward pages that state the conclusion early and clearly, then support it — the same discipline that sharpens human relevance, applied harder.
• Be honest in your structured data. Schema that contradicts the visible page doesn't just risk a penalty; it removes you from the trusted set an AI will quote from, and that trust is slow to win back.
• Be machine-extractable. A clean accessibility tree, stable layout and properly labelled controls — the same things Lighthouse now scores as "agentic browsing" — are what let an AI agent actually read and operate your store.
• Publish genuine first-hand expertise. The "Experience" in E-E-A-T is exactly what these systems try to surface, and exactly what thin, generated filler can't fake.

This pillar is now substantial enough that I've written about its two halves separately. The rubric — how I score whether a site is eligible to be cited, whether its content is worth citing, and whether it can be extracted cleanly — is the 120-point AI readiness framework, the audit I run on every client site. The technical floor underneath it — the fast, stable, agent-readable build that earns a clean Lighthouse "agentic browsing" pass — is exactly what I documented in how I built this site. Treat this section as the strategy, and those two posts as the detailed execution.

A word on B2B

Most of this applies cleanly to both B2C and B2B, but B2B deserves a note because its sales cycle is long and its buyers are patient researchers. They compare suppliers for weeks, route decisions through several stakeholders, and want self-service depth — clear pricing logic, bulk and tiered options, technical documentation, configurators. In B2B, positioning your brand as the credible authority often matters more than squeezing the last drop out of a product page, because the order closes on trust built over many visits, not a single impulsive click. That's the authority and relevance pillars doing the heavy lifting, with technical and experience making sure the long research journey is never frustrating.

Where to actually start

You cannot do all of this at once, and you shouldn't try. The framework's real value is helping you choose what leads. After two decades I've broken my own process into a year-long programme of technical sprints across more than twenty projects — but you don't need that to begin. You need to know which pillar your site is weakest on, and start there.

• A brand-new store? Lead with the technical pillar — get the architecture, indexation and schema right so everything you build later can actually be found.
• An established store that's plateaued? It's usually authority and relevance — years of thin content and a tired link profile holding back genuinely good products.
• Good rankings but poor sales? Almost always experience — Core Web Vitals failing on mobile, or a checkout fighting the customer.
• Ranking well but losing visibility to AI answers? The fifth pillar — start with the AI readiness audit and the schema-honesty and extractability work it surfaces.

That's the framework, rebuilt. Four pillars I trusted in 2018, because the way of thinking has outlasted nearly every tactic underneath it — and a fifth I added only when the era left me no honest choice. The hard part was never naming the buckets. It's the discipline to keep working all five, honestly, while the ground keeps moving.