# Pulse · October 27, 2026

> Shopify folds AI-agent facts into its admin, Google tells platforms to noindex shared and generated pages by default, and OpenAI halves GPT-6.1 Terra's price.

Canonical: https://brandonlazovic.dev/pulse/2026-10-27/  
Author: Brandon Lazovic  
Published: 2026-10-27

## Shopify Moves Its Merchant Knowledge Base Into The Admin

Status: Confirmed  |  Topics: agentic-commerce, platform-ecommerce

Shopify folded its standalone Knowledge Base app directly into the admin on October 26, reachable under Sales channels, Agentic, Knowledge base. Merchants review, edit, and publish up to 1,000 facts, such as contact details and gift services, that AI shopping agents use to answer questions about the store. Shopify auto-drafts facts from published storefront pages, each linked back to its source, and a fact stays unpublished until a merchant approves it. The standalone app sunsets January 15, 2027.

In plain terms: A 'fact' here is a short, merchant-approved statement about the store, like its return policy, that an AI agent can quote back to a shopper without guessing.

Why it matters: This is the first widely shipped admin screen for curating what AI shopping agents are allowed to say about a store, turning fact approval into a routine merchant task instead of a developer project.

Our take: Expect other commerce platforms to add their own fact-curation layer, since an agent that cites an unapproved page is a liability no platform wants to own. The 1,000-fact cap is the detail worth watching, since stores with large catalogs or many locations will hit it fast.

- [Shopify Changelog: Knowledge base is now available in your admin](https://changelog.shopify.com/posts/knowledge-base-is-now-available-in-your-admin)

## Google Tells Platforms To Noindex Shared And Generated Pages By Default

Status: Confirmed  |  Topics: crawling-indexing-rendering

Google updated its 'Prevent user-generated spam' guide on October 26 to cover tools that let users create shareable pages, generated content, or interactive web apps, not just hosting platforms. It now recommends noindex by default, via a meta robots tag, an X-Robots-Tag header, or a robots.txt block, for pages built for sharing or utility rather than public search discovery. It also tells platforms to noindex new users' posts until they build reputation, and to monitor Search Console for abuse.

In plain terms: Noindex is a tag that tells Google not to list a page in search results at all, which is different from blocking the page from being crawled.

Why it matters: Any product that lets end users generate a public URL, a shared document link, or a widget embed now has a documented Google expectation to keep those pages out of the index by default, flipping the assumption from indexable to excluded.

Our take: Platforms that built sharing features partly to generate search traffic will read this as a loss, but it closes a loophole scaled content abuse has exploited for years: mass-produced pages riding on a trusted domain's authority. Watch for enforcement to follow the policy language within a quarter or two, the way it has after past spam-guideline updates.

- [Google Search Central: Prevent user-generated spam](https://developers.google.com/search/docs/monitor-debug/prevent-abuse)
- [Search Engine Journal](https://www.searchenginejournal.com/google-indexing-advice-shareable-pages-spam-guide/594451/)

## Model & AI releases

### OpenAI Launches GPT-6.1 Terra At Half The Price Of Its Predecessor

Status: Confirmed  |  Topics: llm-models-agents

OpenAI released GPT-6.1 Terra, succeeding GPT-5.6 Terra and focused on visual reasoning, computer use, and everyday work like drafting proposals and filling out forms. API pricing is $1 per million input tokens, $0.05 cached, and $5 per million output tokens, half of what GPT-5.6 Terra and GPT-6.1 Sol charge. OpenAI reports Terra solves 50.3% of Terminal-Bench 4.0 coding tasks at maximum effort, more than double GPT-5.6 Terra's 21.5%. It ships now in ChatGPT Work, Codex, and the API as gpt-6.1-terra.

In plain terms: Terminal-Bench is a public test that scores how well an AI model fixes real coding problems from a command line, so a higher percentage means it solves more of them correctly.

Why it matters: For agent and coding workloads that don't need the top-tier model, Terra cuts cost roughly in half while closing much of the capability gap to GPT-6.1 Sol, making it the model to re-test before paying for a pricier one.

Our take: Price cuts on mid-tier models are becoming the fastest lever for agent-heavy pipelines, since a 50% drop changes the math on running an agent over every item rather than a sample. Worth re-testing any computer-use workflow currently built on GPT-5.6 Terra against this release before the next invoice.

- [OpenAI: Introducing GPT-6.1 Terra](https://openai.com/index/introducing-gpt-6-1-terra/)