EditionNo. 03September 25–October 1, 2026

AI Search Intelligence Brief

The developments worth paying attention to — and what they tell us about where digital discovery is going.

Hillary Bassett Ross

Hillary Bassett Ross — Founder, Kinetic Pencil

Updated 10/7/26 · 6 min read

Hillary Bassett Ross is an AI search and web optimization strategist with 20 years of experience across content strategy, SEO, UX, analytics and digital optimization for companies including Microsoft, Cisco and Liberty Mutual. She is the founder of Kinetic Pencil, where she helps marketing leaders understand what changes when AI becomes a primary layer between their brand and their customer—and how content, websites and digital ecosystems need to change with it.

This week’s biggest AI search story wasn’t a new model. Or another GEO tactic.

It was the web quietly growing a second layer. One built for machines.

Agents are sticking around and doing ongoing work. Commerce platforms are opening up actions a machine can take. And infrastructure companies have started measuring, and charging for, the value AI creates when it uses information without ever sending a person to the source.

So the question is shifting.

It used to be:How do we rank in AI?

Now it’s closer to:How does an organization become something an AI can understand, trust, and actually use?

01The big one

The web has a second audience now

Cloudflare put it plainly this week: the internet has two audiences. Humans and agents.

1,700%

Daily requests from AI agents across Cloudflare’s network are up more than 1,700% in a year. And for the first time, more than half of internet traffic is non-human.

AI Search IntelligenceKinetic Pencil

(Cloudflare is careful to separate old-school bots from agents acting on someone’s behalf. That distinction matters.)

But the traffic number isn’t the interesting part.

The interesting part is what Cloudflare built for that new audience.

  • Pay Per Use lets publishers charge AI companies when their content actually gets used, not just crawled. A citation. Evidence pulled into a research report. A fact that helps an agent make a recommendation.

  • Monetization Gateway goes further. Websites, datasets, APIs and MCP tools can charge agents per request.

Which means a whole new sequence is forming:

  1. identity
  2. permission
  3. use
  4. payment

That’s not SEO infrastructure. That’s the plumbing for an agent economy.

AI Search IntelligenceKinetic Pencil

And tucked inside it is a distinction I think we’re about to hear a lot more about:

  1. Crawl is not use.
  2. Use is not citation.
  3. Citation is not traffic.
  4. Traffic is not influence.

For most of the web’s history, nobody needed to pull those apart.

Now we do.

Sources:Cloudflare: The Internet has a second audience · Cloudflare: Pay Per Use · Cloudflare: Monetization Gateway

02The signal

From machine-readable content to machine-operable organizations

For years, machine readability has mostly meant helping machines understand. Schema. Clean architecture. Consistent entities.

This week showed the next step.

Shopify extended WebMCP through checkout. Browser agents can now browse products, manage carts, edit checkout and submit an order once the buyer confirms.

  1. browse products
  2. manage carts
  3. edit checkout
  4. submit an order

Here’s the line I keep coming back to:

Schema describes a thing. An agent interface lets a machine do something with it.

AI Search IntelligenceKinetic Pencil

The machine isn’t just learning that a product exists. It’s getting a defined way to act.

Bloomberg added another piece with its Enterprise MCP offering. Their point: handing AI the data isn’t enough. The machine also needs to know what a number means, where it came from, and when it’s appropriate to use.

So machine-ready information starts looking less like a fact and more like a fact plus its meaning, its source, and when it applies.

  1. a fact
  2. its meaning
  3. its source
  4. when it applies

Layer on persistent agents, which hold context and carry work across systems instead of answering one prompt and vanishing, and the thing we’re optimizing gets a lot bigger.

We’re not just optimizing pages for retrieval anymore.

We’re making organizations legible, and operable, to machines.

Sources:Shopify: WebMCP support for checkout · Bloomberg: Enterprise MCP announcement · OpenAI: Introducing Dots

03The data point

AI-first shopping discovery: up 200%

Salesforce’s latest State of Commerce research draws on behavior from more than 1.5 billion shoppers in 37 countries.

  1. 200%

    Starting a shopping journey with agentic search grew 200% year over year.

  2. 150–428%

    Traffic from AI chats to commerce sites grew 150% to 428% year over year, every quarter measured. Far faster than commerce traffic overall.

AI Search IntelligenceKinetic Pencil

Here’s why that matters. AI doesn’t have to finish the purchase to have commercial power. It gets that power much earlier, by shaping one question:

What makes the shortlist at all?

Then Attentive’s 2026 research adds a wrinkle.

  1. 71%

    Among AI-assisted shoppers, 71% said AI meaningfully shaped what they considered or chose at least half the time.

  2. 62%

    62% bought a different product or brand than usual because of it.

  3. 93%

    But 93% still went looking for evidence outside AI before deciding.

AI Search IntelligenceKinetic Pencil

So the journey isn’t AI → purchase.

It’s more like

  1. AI discovery
  2. shortlist
  3. human double-check
  4. decision

Which means brands have to win twice.

First, be clear enough to a machine to make the shortlist.

Then, have enough real evidence to survive the human checking up on it.

Sources:Salesforce: Shopping’s New First Step · Attentive: 2026 State of AI in Retail

04Connecting the dots

Traffic and influence are coming apart

This might be the idea from this week that matters most for measurement.

37,561

A randomized field experiment with 37,561 real Washington Post readers compared regular archive search against an experience that added AI answers with citations.

AI Search IntelligenceKinetic Pencil

The AI answers changed what people learned.

But a lot of that learning happened inside the answer itself. Not on the article it came from.

Now set that next to everything else this week.

  1. Cloudflare is separating AI use from a visit.
  2. Salesforce shows AI moving upstream into discovery.
  3. Attentive shows AI reshaping the shortlist before anyone reaches a brand’s site.

Put together, it’s hard to ignore:

A source can create real value without ever getting the visit we’ve always used to measure that value.

AI Search IntelligenceKinetic Pencil

I’m calling this Invisible Influence.

And it exposes a gap in how AI visibility gets measured right now. Referral traffic tells you something. Citation counts tell you something. Neither tells you how much influence a source actually had.

The real chain is much longer:

  1. Availability
  2. Retrieval
  3. Selection
  4. Use
  5. Representation
  6. Attribution
  7. Exposure
  8. Influence
  9. Outcome

Most AI visibility tools only see a couple of those links.

That’s going to matter a lot once organizations start trying to prove ROI.

05On my radar

Ranking and AI source selection aren’t the same thing

A big Washington University analysis of Google AI Overviews resurfaced this week. The research itself is older than this window, but it’s worth keeping close.

30%

Across 55,000+ searches and 61,000+ citations, nearly 30% of domains cited in AI Overviews weren’t on Google’s regular first page.

AI Search IntelligenceKinetic Pencil

Ranking

≠

getting selected.

The researchers also checked whether AI claims were actually supported by the sources cited next to them. Which raises a second split:

Being a credible source

≠

being represented accurately.

Getting picked is one problem. Getting described correctly is another. And neither one is getting clicked.

That’s why I don’t think a single “AI visibility score” holds up. The better question is where in the chain an organization is winning, and where it’s quietly losing.

Source:Measuring Google AI Overviews (arXiv)

06The through line

The machine side of the web is starting to look like a real operating environment.

It has an audience, an identity layer, actions, permissions, transactions, and now an economy.

  1. An audience
  2. An identity layer
  3. Actions
  4. Permissions
  5. Transactions
  6. And now an economy

At the same time, the human evidence says something just as important.

AI isn’t replacing the journey.
It’s redistributing the thinking inside it.

AI finds, compares and narrows.

People verify, judge and say yes. Especially when the stakes go up.

So this is a much bigger job than getting cited by ChatGPT.

The next generation of web optimization is about making an organization clear, trustworthy and usable enough that a machine can understand it, represent it accurately, and act on it.

And pass its value along to people.

Even when those people never visit.

— HillaryKinetic Pencil