The list got shorter and the top of it changed. Here is what actually pays now, in order.

Number one, by a distance: the JavaScript rendering gap. Vercel's crawler analysis found GPTBot making 569 million requests a month on its network and ClaudeBot 370 million, about a fifth of Googlebot's volume, and none of them execute JavaScript. They fetch your raw HTML and leave.

Googlebot renders your client-side content fine, which is exactly why the problem hides: rankings look normal while every AI crawler sees an empty shell. It compounds through Bing, since roughly 92 percent of ChatGPT's web-search responses draw on Bing's index and Bingbot's rendering is limited.

The fix is server-side rendering, static generation or prerendering for anything that must be readable: product details, prices, reviews, policies, article bodies. On Shopify, Liquid-rendered themes are mostly safe; app-injected content and headless builds are where to check.

Number two: crawling and indexing hygiene, because AI answers are assembled from the same index. How Google crawls and processes your pages still gates everything above it. That means killing index bloat from faceted and parameter URLs, keeping canonicals coherent so the right version of a page is the one that gets cited, and internal links that give every money page a short path from crawled hubs.

Number three: structured data, upgraded from garnish to interface. Identifiers, ratings, policies and attributes in markup and feeds are what agents compare when they pick products, and validation errors silently remove you from comparison surfaces.

Number four: log-file analysis, which quietly became strategy. Your logs now show a zoo of agents, and OpenAI and Anthropic each run separate bots for training, search indexing and live user fetches. Confirm they get 200s with content-rich HTML.

Then decide deliberately which to allow, since blocking a training bot while permitting the search bot is a real choice most teams have not made. The root files that describe your site to agents are part of the same audit, and worth exactly this much effort and no more.

Number five: speed, demoted but not dead. Fast, complete initial HTML matters doubly now, for Core Web Vitals and because single-fetch AI crawlers take whatever arrives in that first response. Chasing a perfect lab score past "fast enough" is where the ROI ends.

What no longer moves the needle: exotic markup on pages with nothing to mark up, dynamic rendering hacks Google itself calls a workaround, and micro-optimizations on sites whose real problem is thin content, because no technical fix rescues pages nobody would cite.

The one-afternoon audit: curl your top templates and grep the raw HTML for your actual content, then check Search Console's page indexing report for bloat and exclusions.

Run the Rich Results Test on money pages, and grep a week of logs for GPTBot, ClaudeBot, OAI-SearchBot and PerplexityBot.

Those four checks find ninety percent of what technical SEO can still fix in 2026. Everything they surface is boring, fixable in sprints, and worth more than the next redesign, because it decides what every machine reading your store can actually see.