We spent a decade reverse-engineering the feed. Now the feed wants a prompt. Facebook is rolling out plain-language controls for what people see in Feed and Reels. A reminder for social teams: your feed is a sample of one.

Meta just bought AI analytics startup Sundial. AI can make data easier to query. But in social listening, what counts as a "mention": a post, a reply, or a review? Change the unit, and the trend can change with it. Before trusting the chart, check what it counted.

X says its retrieval model is trained to surface posts people finish reading. Yet many brand dashboards still open with impressions. That's an instrumentation mismatch. Views tell you a post traveled. Replies can reveal what people questioned, challenged, or wanted next.

AI search is getting a byline. Google is testing clickable creator names in AI Mode video results. One click leads to a profile of their broader work. For brands, that could shift the question from “Who gets reach?” to “Whose work can we stand behind?”

The AI buildout has two conversations. The industry talks about compute and scale. Nearby communities talk about water, electricity costs, noise and jobs. If you only listen to the first group, you miss the objections that can shape whether a project moves forward.

AI shopping has a new checkout question: Did the agent buy what you meant, or what it inferred? Amex just published a framework for preserving a customer's authorized intent before an agent transacts. The next conversion bottleneck may be permission, not recommendation.

One viral post can fill a dashboard. It can also distract from the useful signal. When the spike fades, which question keeps coming back in the comments? AI can draft the next 100 posts. Social listening helps you decide which one is worth publishing.

AI does not give marketing teams memory. It gives them speed. Without a record of what customers said, which claim failed, or why sentiment shifted, automation repeats old mistakes faster. The next marketing moat is not more content. It is learning that survives the campaign.

Translation is not understanding. About 3.4B people speak languages underrepresented in today’s AI. A system can translate words and still miss sarcasm, slang, or buying intent. Global AI cannot flatten markets into English. Language is access. Context is intelligence.

AI has an averages problem. An agent can score 77.4% across repeated runs, yet succeed every time on only 53% of tasks. Social data has the same trap: one sentiment score can hide platform splits and recurring objections. The average looks clean. The variance tells the truth.