Product descriptions should adapt to natural language. Attribution and ROI measurement for these chat-based sales will be a major challenge.
@picoko these questions are exactly the right ones to be asking and most of the answers have become clearer over the past 9 months.
On whether SEO transfers:
Partially. Structured data, schema markup and backlink authority all carry weight in AI discovery but the weighting differs. Traditional SEO rewards the page that best matches a keyword. AI discovery rewards the brand that has the most coherent, consistent signal across multiple sources. A brand with 5 strong editorial citations in relevant publications will often outperform a brand with better on-page SEO but no third-party footprint.
On new best practices:
Yes, and they’re distinct from traditional SEO. The things that matter most for conversational discovery are citation density from authoritative sources, consistent brand positioning across everything AI can read, FAQ schema that mirrors how buyers actually ask questions, and files like llms.txt and agents.md that tell AI platforms directly how to describe and categorise your store. Product descriptions written for natural language queries rather than keyword density also perform better in conversational results.
On attribution:
This has improved significantly. Shopify now has a native agentic section in admin that shows revenue and sessions from ChatGPT, Google AI Mode, Copilot and others as separate line items. Before that existed, UTM parameters were the only reliable method and even those had gaps because AI-referred traffic doesn’t always pass through cleanly.
The broader point worth making: agentic commerce isn’t one channel. ChatGPT, Perplexity, Gemini, Claude and Grok all have different discovery mechanisms and weight signals differently. Optimising for one and assuming it transfers is the mistake most merchants will make at the start.
We’ve been tracking this across Shopify and WordPress stores and wrote about where this is heading - specifically why the brands that win AI discovery over the next 18 months will be the ones tracking the full landscape rather than optimising for a single platform.
One thing worth knowing that most llms.txt/agents.md advice misses: templates/agents.md.liquid isn’t static text, it’s a real Liquid template - Shopify’s docs show it supports loops and standard objects ({% for version in agents.ucp_versions %}, {{ shop.name }}, etc.). Most implementations I’ve seen (including my own app’s, until I looked into this) just write a plain-text snapshot at publish time, which goes stale until you manually refresh it. Building it with actual Liquid against your live catalog means it never goes stale - worth doing if you’re setting this up rather than pasting a static file.
(Disclosure: I build AEO Pro.)