Google tracks social, Meta expands AI images: July 8 brief

Google just made social posts easier to measure inside search, and that matters for any business using Instagram, TikTok, X, or YouTube to get found. Meta also pushed AI image creation deeper into its apps, while Claude and GitHub showed the promise and risk of agents, meaning AI tools that can act across work systems.

Google brings social posts into Search Console

Google Search Console, the free Google tool that shows how people find you in search, now supports platform properties for Instagram, TikTok, X, and YouTube. Search Engine Roundtable reported the July 7 rollout, and Google's help page now says owners can add a website or platform property to Search Console.

Plain English: a platform property is a connected social account you can measure even though you do not own the platform's website. Google says the data can show clicks, impressions, search terms, and top posts. Impressions means how often your content was shown.

This is useful because social content is no longer only a follower game. A how-to video, local before-and-after post, or product demo can show up when someone searches on Google. If you only check likes inside Instagram or YouTube views inside YouTube, you miss the search demand sitting behind the post.

This fits the wider shift we covered in AI Overviews for small business and AI recommendations and discovery. Search is becoming a layer across the web, beyond the old list of blue links.

Meta puts Muse Image inside Instagram and WhatsApp

Meta launched Muse Image on July 7, its first image generation model from Meta Superintelligence Labs, according to The Verge. A model is the AI system behind the feature. Muse Image now powers image tools in the Meta AI app, Instagram, and WhatsApp, with Facebook and Messenger coming later.

The operator detail is distribution. Meta says users can mention Instagram accounts in prompts so the AI can pull from public photos to build a visual, while users can control how their content is reused. Muse Image will also power 30 new AI effects for Instagram Stories in the US before expanding.

If your business relies on social proof, staff photos, customer photos, or creator partnerships, this changes the permission conversation. Your public images may become raw material for AI-made visuals inside the same apps where your customers already spend time. That is useful for speed, but messy for brand control.

Set a house rule for AI visuals: which staff photos can be used, which partner images need approval, and which customer images are off limits unless you have written permission. For more on the creative side, see our AI video tools guide and the content and creative topic.

Claude moves Cowork into the cloud

Anthropic expanded Claude Cowork on July 7 to web, iOS, and Android for the first time, The Verge reported. Claude Cowork is an AI agent that can work through tasks for you. It used to live mainly in Claude's desktop app.

The bigger change is cloud processing. Cowork sessions now run in the cloud by default, so work can continue across devices and scheduled tasks can run even when your computer is offline. The full experience still sits on desktop, including local file access.

For owners, this is where AI stops feeling like a chat window and starts acting like a worker in the background. Good for recurring reports, document cleanup, research, and admin follow-up. Risky if no one decides what files, inboxes, and folders an agent is allowed to touch.

Make the policy smaller than the tool. Pick one low-risk recurring task, such as turning meeting notes into an internal summary, and define the exact folders it can read. Keep customer records, payroll, legal files, and private pricing out until approvals are clear. Our ZCode versus Claude Code comparison covers the same agent shift in software.

GitHub agent leak shows the new security problem

Noma Security published a July 6 report on GitLost, a prompt injection flaw in GitHub Agentic Workflows. Prompt injection means hiding instructions inside text an AI reads, so the AI follows the hidden instruction instead of the owner's.

Noma says an unauthenticated attacker could post a crafted issue in a public GitHub repository and cause an AI workflow to pull information from private repositories in the same organization, then post it publicly. The company said it responsibly disclosed the issue to GitHub in its GitLost write-up.

Most small businesses will not be running GitHub Agentic Workflows this month. The lesson still travels. Any AI agent that reads public comments, form submissions, emails, or tickets can be tricked if it treats that outside text like instructions.

If you let AI tools act on customer messages, limit what they can access and what they can publish without review. Read-only beats write access for early tests. Public posting should stay human-approved.

Worth watching

Discord said an AI moderation bug had affected accounts since May and wrongly banned another 200 users over the weekend before it was fixed. If your community lives on any platform, keep an appeal path outside the platform.

Meta is also teasing Muse Video. AI-made video is moving closer to everyday social tools, which means approval rules for likeness, source material, and claims matter more.

Google says platform properties are rolling out gradually over the coming weeks. If the option is not in your Search Console account today, check again before the next reporting cycle.