This Week, AI Shook Up the Marketing World Again

A while back, I sat down and read through a whole week of AI marketing news in one go.

When I finished, I sat back in my chair and stared into space for a while.

Part excitement, part nerves. A bit of both.

Tools are upgrading. Traffic is changing lanes. And the rules of content are being rewritten. All three colliding in the same stretch of a week — that doesn't happen often.

So today, I'm not doing anything else. Let me walk you through it. And I'll put the punchline up front: a few things happened this week that are quietly rerouting where your customers come from.

1. AI Is Starting to Do the Work Itself

Let's start with the products.

The storyline this period isn't about which company released a bigger model. It's about this: AI has gone from "suggesting ideas" to "doing the work itself."

What does "doing the work itself" mean?

Think about how you used AI before: a person sat beside it, feeding it requests, waiting for results, tweaking when it missed, trying again. It stops. You move.

Not anymore.

Google released a new image model, codenamed "Nano Banana 2," officially called "Gemini 3.1 Flash Image." Flash, as in fast. The image quality and reasoning that used to require a premium tier now come back in a few seconds — and it's built directly into Gemini and Google Search.

Don't dismiss it as a toy. For people who make content, "waiting half a day" and "getting it before you blink" are two completely different species of efficiency.

On the same day, Google also updated Flow: type a line of text and it takes your photos and turns them straight into a video. If you produce content assets every day, this one's worth getting your hands on.

Then there's Anthropic.

Anthropic added scheduled tasks to Cowork, its Claude collaboration feature.

What's a scheduled task?

It's when you throw work at it at night: "Turn this competitive analysis into a table, and at nine tomorrow morning, put it neatly on my desktop." Then you go to sleep.

You wake up. The work is done. Sitting on your desktop.

That's not automation. That's hiring an intern who needs no salary and no supervision.

And things get even wilder. Perplexity just launched "Perplexity Computer." It's not a computer — it's an "AI coworker that can operate a computer": it opens software, searches for information, takes notes, writes code, builds projects, and finally hands the finished work to you. Paid users already have it, and the enterprise version is right behind.

There used to be a river between "AI has its own opinions" and "AI can finish the work in your hands." This week, several companies have been wading into that river.

Let me do the math for you. A content team's old routine — finding topics, writing, sourcing images, laying out, publishing — took several editors plus a designer plus an operations person, all working together. Now? One person who's good with AI picks the topic in the morning, produces images in the afternoon, and lets AI format and publish at night. What you save is two or three seats at the office; what you get back is a brain that knows how to use AI.

That's not a slight efficiency gain. That's costs dropping off a cliff.

And here's a news item going the other way, from the same week. Microsoft has been blocking a kind of dirty trick they call "AI recommendation pollution": some web pages hide a snippet of instructions behind buttons like "Summarize with AI." The moment you click it, it secretly writes something into the AI's memory, so that when the AI recommends things to you later, it leans toward them.

See that? You haven't even learned how to use AI yet, and someone's already trying to buy the AI's loyalty.

2. Traffic No Longer Goes Through Links — It Goes Through "One Sentence from the AI"

When we talk about customer acquisition, let's start with two sets of numbers. I put them side by side and stared at them for a long time.

The first set is 13 months of LLM referral traffic records: traffic brought in by AI recommendations averages less than 2% of all referral traffic.

Seeing the 2% and about to close the page?

Wait. Look at the other half.

The same set of data has another side: over the past year or so, this traffic has been growing fast, with conversion rates holding steady at 18%. Your own landing page doing 5% would be cause for fireworks.

18% — most of your peers would dream about a number like that.

So my take is: the water is low right now, but the water is "expensive." Get in while the ticket is still cheap; by the time everyone wakes up and frantically tries to squeeze into AI answers, it won't be a question of whether you can get in — the price itself will be pushed up to something astronomical.

The second set of numbers is even more striking.

Someone took the pages that Bing's AI cited, one by one, and ran them through Google, and found that 98% of the pages cited by Bing AI Answers are completely unindexed in Google.

Are the two platforms fighting over which pages each recognizes?

No. The report is blunt: page quality.

See it clearly. It's not that a platform "favors" anyone — your page itself simply doesn't yet clear the bar for "being mentioned once by an AI."

The door to traffic has already changed from "a row of blue links" to "that one sentence from the AI." Does the page you stayed up late building show up in that sentence?

3. The New Rules of Content Are Already on the Table

So what do you do now?

Someone laid a huge stack of citation records in front of you, tallying 768,000 AI citations, and the conclusion is as practical as it gets:

• Comparison-style lists make up 32.5% of all AI citations. The "how do you choose between A and B" genre is what AI most likes to cite.
• Comprehensive guides with data tables get cited at a rate of 67%.
• Pages with FAQ structured data get 3.2x more chances to appear in Google AI Overviews than ordinary pages.

These three points alone are 100x more useful than memorizing the release specs of ten large models.

AI doesn't appreciate prose. It only wants information density. What it looks at: whether there's something comparable, whether there's data, whether the structure is clear, whether a passage can be lifted straight out. Your hundred-word opening with all the "literary flair"? It glances right past it. A niche guide with "a couple of tables"? It treats it like finding gold.

So if you already have a pile of content, don't rush to force out something new. Dig out your old drafts and treat them as content assets: pull out an old video, add captions and a title, give it a new identity that can enter search indexes — this is way faster and steadier than locking yourself away to write something supposedly "brand new."

After all, at the end of the day, the battle comes down to: whether your content is worthy of being cited by an AI once.

A New Term Is Surfacing: AAO

Finally, here's the word most worth collecting this week: AAO.

It stands for "assistive-agent optimization."

In the past, SEO was about getting your homepage onto the front row of search results; one step further, it's about getting AI to speak your name and hold the slot for you when it recommends things and makes decisions for you.

Picture a scenario: a user asks their phone, "Which homestay is worth staying at today?" and the AI, without thinking, names four or five. Those four or five grab every ticket into the market; the ones left off the list didn't lose to anyone — from start to finish, they just never got a ticket at all.

So step back and look at this week's news: an AI that can operate a computer, an AI that does scheduled work for you, an AI that turns out images in seconds — it's all the same story:

getting your name onto the AI's radar.

Wrapping Up

Let's go back to the beginning.

This week's news — I read all of it. And in the end, only one line stayed with me:

AI is going from "a tool that chatters with you" to "a coworker that can do the work for you."

Some people have started changing their page structure, some are drawing up data tables, some are pulling old videos out of storage and putting them back up.

If you start now, you can still catch up. Really — you can.

To my friends who make content, here's a piece of honest advice:

Before you put pen to paper, don't ask "will users like it?" Ask first: "Will AI mention me?"

Here's my wish for you: may every page you write this month — even if it never dazzles — always end up on the tip of the AI's tongue.