Social Commerce

Only 2% of 2,053 organizations allow revocable AI-agent access

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 itPexels / pixabay

ReadyIndex’s scan of 2,053 organizations found only 2% offer revocable AI-agent authorization, while Meta reported 10 million-plus Muse setups in six weeks.

By Tom Whitfield

4 min read

At a glance

  • ReadyIndex scanned 2,053 organizations and found an average agent-readiness score of 2.5 out of 100.
  • Only 2% supported customer authorization with limited, revocable AI-agent access; none reached top-level readiness.
  • Only 7% could connect to an assistant, 2% published a machine-readable catalog and 7% stated whether agents were welcome.
  • Just 23% published a file identifying current content and permitted agent uses.
  • Meta reported 10 million-plus Muse setups in the United States and Canada within six weeks; 25% used it daily and average purchases exceeded $200.

Only 2% of the 2,053 organizations scanned by ReadyIndex let customers authorize an AI shopping agent with access they can limit and withdraw. The study put average agent readiness at 2.5 out of 100, and no organization reached the top tier covering web, interface and voice.

Meta supplied the strongest consumer-side signal. On its latest earnings call, according to the analysis, the company said more than 10 million people in the United States and Canada had set up Muse agents within six weeks. A quarter used Muse daily, and the average purchase exceeded $200.

The analysis gives no date for that call or independent measurement of completed agent conversions. It describes tens of millions of Americans as trusting personal assistants such as Meta’s Muse or Instinct to book, buy and arrange parts of their lives. Those remain broad claims, not a verified market total.

After more than two decades of filtering bots from the web, marketers now face traffic that the analysis calls customers in another form. Websites are reporting record visitor counts as assistants move at machine speed, although no traffic dataset or named site examples accompany that claim.

What does blocking AI agents cost marketers?

Blocking can sacrifice demand at two points. The assistant does not complain; it moves to the next site that will complete its task, very likely a competitor, and the customer never learns the blocked business was available.

The longer effect runs through model training and context. Feedback from agent interactions shapes what the next generation of models knows about the web. A site the assistant cannot use may be remembered as not worth trying.

The analysis provides no cost-per-click, acquisition-cost or campaign-budget forecast. The $200 average comes from Meta and cannot support a budget forecast for every agent-led purchase. The financial exposures are concrete: a block may transfer a purchase, while poor agent access may weaken a site’s future position. Teams need transaction-level conversion data before assigning spend.

An assistant arrives with intent: a direct, specific request for what the customer wants, delivered before the customer reaches the site. That differs from a referrer or bounce rate, and a working conversation can reveal the person’s actual goal.

Only 7% of scanned organizations could connect to an assistant at all, which the analysis says may be pulling them ahead. Search, social, display and email were built over 30 years around a person viewing a page. That model no longer holds reliably. The emerging audience includes software acting for customers, so competition is moving toward agent attention, task completion and ease of use rather than human eyes alone.

From access to task completion

The analysis draws a deliberate distinction between admitting agents and supporting them. ReadyIndex states: “Letting agents in is the floor rather than the goal.” Companies must decide what an agent may do after entry, where a human must approve a step and when a person must take over.

Only 2% published a machine-readable product catalog, and 7% published any position on whether assistants were welcome. Just 23% had a file identifying current content and permitted uses.

For the rest, an agent could return an offer that expired two years earlier, even if the page had been removed from navigation. A hidden page remains as findable as the homepage because agents read available pages rather than navigate menus.

Software businesses led the scan at 57.7 on average, compared with 38.8 across all organizations. ReadyIndex attributes that advantage to products already built for machine-to-machine use, not superior creative work.

The practical requirements are straightforward:

  • Publish a product catalog that machines can read.
  • Publish a file explaining what content is current and what assistants may use.
  • Give customers narrow, revocable authorization for specific tasks.
  • State in writing whether AI assistants are welcome.
  • Track why agents arrived, what they did and how many became customers.

The analysis compares this moment with the mobile shift, when marketers debated responsive design while competitors built for phones. It says positions are already being allocated according to which businesses an assistant can finish a task with, and that the shift is further along than it appears.

ReadyIndex concluded: “The companies winning it are not winning with better creative. They are winning because an assistant could read them.” The near-term measure is whether authorization and readiness scores begin to rise as more organizations publish machine-readable information and track agent conversions.

Source: fastcompany.com

Tom Whitfield

Show full bio

News editor covering regulation, policy and their impact on Social platform news and policy at Reach Brief.

More from Tom Whitfield →

Related articles

« Previous articleNext article »