The one thing
This week Google started showing its work: inside AI Overviews and AI Mode, it began surfacing your “Preferred Sources” and stamping a “Highly Cited” badge on the reporting other articles lean on.
Read that as the discovery layer turning inside-out. For fifteen years the game was click through rate; the game now is being a source cited in the answer.
GEO is quickly becoming the future of web search.
If you depend on Google search to be found by customers, readers, etc., then the question is no longer “what’s my ranking” but “what does the model say when someone asks about me, and who does it cite?”
If you’re trying to find anything on Google, then the job is no longer scanning links to do your own research; it’s auditing the summary to see if you can actually trust the information it just handed you.
Which begs the real question: How much do you think companies will pay to have their sources injected directly into these summaries?
Signals
Google AI Overview
Google brings Preferred Sources and a “Highly Cited” badge into AI answers
Google is now labeling, inside the AI answer, which sources it trusts and which reporting everyone else cites. Translation: “rank on Google” is becoming “get cited by Google’s AI.” With AI Overviews already cutting clicks ~38% on the queries they touch, the move that ages well is to check how you show up inside the answer the way you used to check your ranking before your traffic tells you to.
Google’s own numbers show a visible 'Preferred' badge doubles your click-through rate inside the AI answer. But the system requires users to explicitly trust you first. It's a flawed design: as of right now, established content giants will become more entrenched, pulling the ladder up on anyone new.
The Human Edge
The one to read slowly this week. When everyone can generate the words, your edge is knowing when not to and Mollick’s point is that staying in the loop and learning is a habit you keep on purpose, not a default you drift into and out of when you don’t feel like working. File it next to anything you publish under your own name. Your personal brand is still the greatest competitive differentiator you can build today.
AI Video Localization
Adobe Firefly now dubs a video into dozens of languages, with lip-sync to match
The lip-syncing piece is rolling out in early access on enterprise plans, but the direction is set: translate-and-dub is now automated and better in every way. And this new feature doesn't just translate words; it uses voice cloning to lock down the original speaker's tone, cadence, and acoustic qualities across languages. AI will continue to eat into post-production editing for the foreseeable future.
AI Video Character Consistency
Reference-image character consistency is now the standard workflow, not a hack
AI video’s core problem is what the current model generation was built to fix: tools like Runway’s Gen-4 and Veo’s reference inputs anchor a character to reference images so it holds across shots. Define the character once, reuse it everywhere, instead of re-rolling the dice each prompt. If you bounced off AI video because it couldn’t hold a face, solutions to that problem are rapidly improving.
Noise
Skip the valuation theater
Anthropic closed a ~$65B round at a ~$965B valuation this week. A real number that changes nothing about your Tuesday. The enterprise-deployment race (Big Four rollouts, OpenAI’s new consulting arm) has been loud for two weeks; it’s an industry story, not your story so let it go. Skip the data-center-grid doom-posting too: real for the grid, noise for your inbox.
And yes, there’s a new Claude (Opus 4.8)
A solid point-upgrade, but no jaw-dropping changes so mostly hype; the bit worth filing away is that early testers say it flags its own uncertainty more readily, handy if you lean on it for anything you can’t personally fact-check. Let the people who tweet benchmark charts carry the hype.
I’ll be testing this model thoroughly during the week as well. So stay tuned on my X for updates.
Think I filed the wrong story under Noise? Hit reply — I read every one.