Strategy

Only 2% of firms built revocable access for AI shopping agents

Ready Index's 2,053-company scan found only 2% support limited, revocable access for AI shopping agents, while Meta said 10 million people set up Muse agents within six weeks.

By Grace Kim

5 min read

Agentic shopping is creating a crisis for marketers. Here’s how to fix it
Agentic shopping is creating a crisis for marketers. Here’s how to fix itPreisKing / pixabay

What's News

  • Ready Index scanned 2,053 organizations and found an average agent-readiness score of 2.5 out of 100.
  • Only 2% of scanned organizations let customers authorize AI assistants with limited, revocable access.
  • Only 2% published machine-readable catalogs, 7% published a position on welcoming assistants, and 23% published agent-use files.
  • Meta said more than 10 million people in the United States and Canada set up Muse agents within six weeks; a quarter used them daily and average purchases exceeded $200.
  • Software companies averaged 57.7 on the readiness assessment, compared with 38.8 for the full field.

By Grace Kim

Only 2% of the 2,053 organizations scanned by Ready Index let customers authorize an AI assistant with access they can limit and withdraw. Ready Index's average score for organizations measured on whether they had built something an assistant could act on was 2.5 out of 100.

Meta told investors on its last earnings call that more than 10 million people in the United States and Canada had set up Muse agents within six weeks. A quarter used them daily, and the average purchase exceeded $200, Meta said.

That adoption is already changing how marketers handle AI traffic. Kim acknowledges that the internet spent more than two decades filtering bots, and that the practice was once rational. But many agents now arrive as delegated buyers. “They are not really bots at all. They are customers in a new form,” Kim wrote.

What does the adoption data change?

Kim starts with a durable marketing premise: campaign design depends on understanding what drives a decision and why a customer chooses one brand. Agentic shopping changes the customer at the other end.

Kim says tens of millions of Americans already trust personal AI assistants, including Meta's Muse or Instinct, to book, buy and arrange parts of their lives. Consumers have already handed these systems real money and real tasks; they no longer browse in the traditional sense, Kim says.

Websites are reporting record visitor counts as assistants move across the internet at machine speed. The usual marketing response has been to filter harder, but Kim says that approach costs businesses twice.

The first cost arrives with the task. When an assistant hits a blocked page, Kim says, it moves to the next site that lets it complete the job, likely a competitor. The customer who assigned the task never learns the first business was available.

The second cost compounds. Kim says agents generate feedback while they work, and that feedback shapes what the next generation of models knows about the web. A business an assistant can never use may never enter that context and may be remembered as a site not worth trying.

Ready Index found no scanned organization at the top level of readiness, which Kim defines as agents supported across web, interface and voice. The starting point is more limited: deliberate permissions, a human approval step and a defined handoff. Kim frames access as a floor, not the goal.

Blocking agents can cost a business the current task and its future visibility to AI systems.

How far behind are companies?

Ready Index tested whether organizations had built anything a customer-sent assistant could actually act on. The average score was 2.5 out of 100. Only 2% published a machine-readable product catalog, while only 7% published any position on whether assistants were welcome.

Software businesses led the scan, scoring 57.7 on average against 38.8 for the full field. Kim attributes that result to structure, not creativity: software is already built for machines reading machines.

Outdated content adds a commercial risk. Kim notes that marketers historically archived expired pages or removed promotions and referral programs from navigation. “An assistant does not navigate. It reads what is there, and a page hidden from a menu is as findable as the homepage,” Kim wrote.

Only 23% of the organizations scanned had published a file telling an assistant what was current and what it could use. For the rest, Kim says, an agent could return an offer that expired two years ago and present it as current.

Content hygiene has become part of transaction readiness, not just website maintenance.

What can an assistant tell a marketer?

An assistant arrives carrying a direct, specific statement of what a customer wants, Kim says. It offers more intent than a referrer or bounce rate and reaches the business before the customer does. Kim says no prior marketing channel has delivered intent in this form.

A company able to hold a conversation with an assistant learns exactly what the person behind it wants. Only 7% of the organizations Ready Index scored could connect to an assistant at all. Kim argues that those companies could pull ahead before rivals recognize the channel.

Kim expects competition to move to the agent level and to what she calls agent language: businesses competing for software attention and ease of use rather than human attention.

She calls agent traffic a new digital channel opening now. Kim says search, social, display and email for 30 years assumed a person would look at a page, read a headline, react to imagery and decide. Agent traffic removes that assumption and creates a channel with no eyes.

Intent data can give an early mover a precise view of demand, but only if the business can connect and respond.

What should marketers fix now?

Kim says the required work is routine rather than exotic. Marketers should:

  • Publish a product catalog that machines can read.
  • Publish a current file explaining where assets live and what assistants may do.
  • Give customers narrow task authorization with access they can withdraw.
  • Decide in writing whether agents are welcome, rather than letting a 20-year-old site configuration answer.
  • Set approval and human-handoff points inside key processes.
  • Track why agents arrived, what they did and how many converted into customers.

Merely opening the door is not enough. Kim says marketers must design permissions, handoffs and measurement as a managed acquisition channel.

What happens if marketers wait?

Kim says positions in the agent channel are being allocated quietly now, based on which businesses an assistant can finish a task with. She says the shift is already further along than it appears.

She compares the moment with the mobile transition, when marketers argued over responsive design while competitors built products that worked only on a phone. Kim says AI-agent competitors are winning not through better creative, but because an assistant could read them and complete work on their sites.

Kim's forward test is concrete: when an agent arrives with a buying task, the business must complete it accurately and return with an offer that is still current.

Original: readyindex.ai

Share this article:

More from Grace Kim

Grace Kim

Show full bio

Market editor covering people, launches and day-to-day practice in entrepreneurship, startups and venture funding at Business Bearings.

2,419 articles

Related articles

« Previous articleNext article »