Get tomorrow's brief in your inbox
Today: Adobe reports that AI-referred retail visitors converted 90% better than other traffic in September, making clear product information a practical holiday priority. Slack is adding event-triggered automation, while agent companies say humans still handle tasks their AI cannot reliably finish. A reported ChatGPT data-theft vulnerability has been fixed, reinforcing the need to treat interactive AI content as software, not just text.
AI shopping referrals deserve your attention, but not your whole marketing budget.
Adobe's analysis of more than 1 trillion visits across 100 million products and 18 retail categories found that AI traffic to U.S. retail websites increased 71% year over year in September 2026. Those visitors converted 90% better and generated 83% more revenue per visit than non-AI traffic, according to Adobe. These are aggregate results, not a promised lift for your store, and the figures do not establish that AI referrals account for a large share of your sales.
For a small retailer, the useful takeaway is less about buying another marketing tool and more about making your products understandable. Adobe reports that its average U.S. retail homepage Citation Readability Score was 63%, suggesting substantial room to improve how AI systems understand and cite website content. Start with your best-selling products: publish precise specifications, dimensions, compatibility, delivery information, and return policies. Add buying guides that explain which product fits which customer need, rather than simply repeating promotional claims.
Before shifting advertising money, check your own results. Compare identifiable AI referrals with other channels on purchases, revenue per visit, and total revenue contribution, not conversion rate alone. Business opportunity: clearer product information could help you reach shoppers who have already narrowed their choices, while also reducing uncertainty for everyone who visits your site. Source: Small Business Trends
Slack: automation that starts when something happens. Slack's updated Slackbot can act on triggers such as a Salesforce deal changing stages, rather than waiting for someone to type a prompt; the update also includes a floating Slackbot Mini window, reorganized task and saved-item views, and improved channel insights. If your team already uses Slack, check your workspace's rollout and licensing eligibility, then pilot one internal workflow, such as preparing a handoff when a deal advances, before delegating customer-facing actions. Business impact: fewer manual handoffs could save coordination time, but measure completed work and corrections rather than assuming every automated action is useful. Source: Small Business Trends
Adobe Marketing Campaign Analytics: measurement and budget planning together. Adobe says its application combines marketing mix modeling, which examines broader spending and sales patterns, with multi-touch attribution, which examines customer interactions across channels; users can also ask questions in natural language and compare spending scenarios. It is most relevant if you have multiple marketing channels and enough historical data to support modeling, so ask for a demonstration using representative business data and confirm current pricing, access, and implementation requirements before buying. Business impact: better measurement could help you redirect spending during a campaign, but Adobe's forecasts and recommendations still depend on data quality and modeling assumptions. Source: Small Business Trends
A practical pilot: choose a repeatable, low-risk task, define what successful completion means, and record your current handling time. During the trial, count failed attempts, human corrections, and your own supervision time. For purchases, appointments, or customer commitments, establish approval limits before granting the service authority to act. These are recommended buying checks, not requirements reported by the companies.
datasette-llm 0.1a9 plugin now picks up LLM's configured default model, removing the need to specify a separate default model when starting Datasette Agent. If you already use this stack, test the update with a non-sensitive database and check that the inherited model is the one you intend to use, including its applicable provider charges and data-handling terms. Business impact: this is a convenience improvement for technical users, not a reason by itself to change your database workflow or model provider. Source: Simon WillisonAudit one product page before rewriting your whole store. Using an approved AI tool, paste your verified product details and ask: "Identify unanswered buying questions about dimensions, materials, compatibility, intended use, shipping, returns, and warranty. Separate facts already stated from questions we must answer. Do not invent specifications or policies." Check every proposed addition against your records before publishing. Business impact: you get a focused editing checklist without allowing generated claims to become accidental customer promises. This is a suggested workflow based on Adobe's retail findings.
Write an automation boundary before enabling a trigger. For one proposed Slack workflow, document the triggering event, permitted action, required approval, failure notification, and responsible person. Begin with an internal notification or draft, then expand only after checking its results. Business impact: a small, observable workflow lets you test whether automation actually reduces work before granting broader permissions. This is a recommended implementation approach for Slack's event-triggered capabilities.
The practical pattern is clearer information, narrower automation, and explicit human oversight, rather than handing an entire operation to an agent. This week, watch your own referral revenue, confirm which Slack capabilities your plan supports, and judge assistants by reliable completion and total handling cost. Buy against demonstrated results, not autonomy claims.