Building a Store Is Now a Conversation

Shopify introduced Canvas on October 1, 2026, describing it as a new design surface where merchants can use Sidekick, Shopify's AI agent, to build their store. The rollout was taking place over the days that followed. Canvas shows all the store's pages together, letting a merchant view the entire store and then zoom in to edit details. Shopify describes Canvas here.

That changes where the hard work sits. If a conversation can shape the pages, managing templates matters less. The business still has to decide what those pages should say and check that the words match what it sells.

I have built for the web since 1994. I work as a consultant or fractionally on content strategy, web architecture, AI-visibility research and custom agent systems. The shift toward conversation-based building points to the end of the content management system era, but the business's facts still need human attention.

What changes when a store is built by conversation?

With a conversation, a merchant can direct an AI agent to work on a store, rather than begin with code or a template. Shopify director Ben Sehl said an entrepreneur on Shopify can build a fully custom store in twenty minutes. In Canvas, merchants can view pages together and then work on details.

Content management systems were built so people who do not program could change a website without code. When a conversation can build the site, that original reason for a content management system weakens. The shift concerns who or what makes the changes. The business still has to decide what belongs on the site.

What can an AI conversation not know about a business?

A conversation cannot supply the facts: what the business actually offers, its prices and its proof. Someone has to gather and check those details before they go onto the site. Otherwise, an easy way to build a page can make it easy to publish language that the business has not verified.

That work is concrete. A team must decide which offer a page describes, check its price against what the business charges, and identify the proof behind any claim. If the facts are missing, the agent cannot fill that gap with reliable knowledge about the business. A polished page does not change that.

I published my own rubric for scoring how readable a site is to agents. Any firm can use it to score itself and get the same answer.

How do I measure whether AI systems recommend a company?

I measure AI visibility by asking the same buying questions on a fixed schedule and checking whether the answer names the right company. My locked library holds a category's real buying questions. I ask ChatGPT, Claude, Perplexity, Gemini and Grok, check official domains so a same-named company does not count as a win, and make each model explain why it chose what it chose.

I keep each result in one format. That lets me compare week 11 with week 1 without changing the questions or the way I record answers. If the test changes along the way, a difference in results may come from the method instead of a change in visibility.

During one engagement, Soapbox's AI recommendation win rate rose from 6.2% to 36.2% in 11 weeks. Trinzik did the work using the platform I built. This is directional evidence, not a controlled study. In a separate program, publishing into the gaps left the win rate unchanged, so I test again instead of calling it a win.

Common questions