twimco.Let’s talk ↗︎
Menu
Field notes / Commerce

Cloudflare now shows which personal agents visit your store. Follow one to checkout before you block any

Some of the automated traffic hitting your product pages is a customer who asked an AI assistant to shop for them. Cloudflare's new Personal Assistants view names those agents and shows where they go. Spend an hour following one before you set a policy.

Editorial engraving of a bright boutique where the owner traces a route across a printed store plan while an assistant stacks folded linen by the windows.

Since November 4, 2026, Cloudflare's Security Analytics shows a Personal Assistants category and a widget listing which personal agents visit your site, with the paths, status codes and networks for each. Its renamed AI Crawler and Agent access policies let you choose Allow, Block on pages with ads, or Block for each supported agent. For a store, filter on one agent, follow it from product page to checkout, fix the pages where it fails, and only then decide which agents to allow.

What Cloudflare added on November 4

Personal agents are the AI assistants people ask to do things online for them: compare a few products, find a returns policy, buy the one with the best reviews. To your store, all of that looks like automated traffic, mixed in with scrapers, crawlers and the occasional attacker. Most stores have been left to guess which is which.

On November 4, Cloudflare added three things that make the guess less necessary:

  • A new category in Security Analytics. The Bot Analysis chart now separates Personal Assistants from verified bots, likely humans and other automated traffic.
  • A Personal Assistants widget in Top statistics. It lists which agents visit and how many requests each sends. Filter on one, and the rest of the dashboard shows only that agent's paths, status codes and source networks.
  • Per-agent access policies. The setting is now called AI Crawler and Agent access policies. You set an overall policy for agents, then a different one for each supported personal agent: Allow, Block on pages with ads, or Block. Your Search and Training policies don't change.

For sites using Account Abuse Protection, which is in early access, Cloudflare also shows which user account each agent acts for, flags logins and sign-ups made by an agent, and lets a fraud team block a single account with a rule on its hashed user ID instead of blocking the agent for everyone.

Why a store should look before it blocks

Blocking every agent feels safe. Cloudflare's own post makes the point that it can also mean losing customers. The person behind the agent wanted to buy something. If the agent can't reach your checkout, it will finish the job somewhere else, and you'll never see the abandoned cart.

An agent that gets a 403 at checkout isn't a bot problem. It's a customer you turned away without seeing their face.

That doesn't mean allowing everything either. Some traffic labeled as a personal agent may behave badly, overload search, or hammer sign-in pages. The point of the new view is that you can now tell the difference by looking at where a specific agent goes and what happens when it gets there.

How the work moves

Following one agent through a store

  1. Pick one

    In Top statistics, filter the Personal Assistants widget to a single agent.

  2. Read the paths

    Product pages, search, cart, sign-in, checkout: where does it go, and where does it stop?

  3. Read the codes

    Any 403s, 404s or 429s on the pages a buyer must reach?

  4. Fix, then set policy

    Repair the pages it fails on, then choose Allow or Block for that agent.

Illustrative walkthrough using the filters Cloudflare described on November 4, 2026; not a client screenshot.
An agent that gets a 403 at checkout isn't a bot problem. It's a customer you turned away without seeing their face.

Follow one agent from product page to checkout

Pick the personal agent sending you the most requests and spend an hour with it.

Start with the paths. Is it reading product pages, using site search, adding to cart, reaching checkout? An agent that reads ten product pages and stops is behaving differently from one that loops on the same search.

Then the status codes. Look for errors on pages a buyer must reach. A 403 on the cart might mean your own security rules are blocking it. A 404 on product URLs might mean it's following stale links from somewhere. A pile of 429s means rate limits are cutting it off mid-task.

Then the networks. Traffic from the networks you'd expect for that agent's operator is reassuring. Traffic for the same agent name from somewhere unexpected deserves a second look.

Finally, the account view, if you have it. An agent that a customer has used on their account for weeks is very different from one that appeared on fifty accounts this morning.

Then repeat the exercise with the second-busiest agent. Two agents are often enough to show whether a problem is about one operator or about your own pages.

Write down what you find for each agent. It usually takes less time than you'd expect, and it makes the policy decision much easier.

Fix what it fails on first

Most of what you'll find isn't about agents at all. It's about pages that were always a bit fragile, now visited by something less forgiving than a person.

  • Rules that block too much. A security rule written for scrapers may be catching legitimate agents at the cart or checkout.
  • Pages that need a script to show the price. If the price, stock or delivery estimate appears only after a script runs, some agents won't see it.
  • Checkout steps built for humans only. Pop-ups, forced account creation and unlabeled buttons slow agents as much as they annoy people.
  • Answers the page doesn't give. Shipping costs, return windows and sizing that live only in a help article.

Our note on blocking AI crawlers explains why closing the door often hands the answer to a competitor, and who inside your company may spend when your own agents hit HTTP 402 looks at the same shift from the buyer's side.

Decision guide

Choosing a policy per agent

ConsiderationLean toward AllowLean toward Block
Where it goesProduct pages, cart and checkout, like a buyerSign-in pages over and over, or paths no buyer uses
What it gets backMostly 200s, a few errors you can fixErrors that suggest probing, or load you can't serve
Who it acts forAccounts that have used it for weeksA new agent appearing on many accounts at once
What your pages earn fromSalesAd impressions an agent never sees (use Block on pages with ads)
Decision framework for a store; Cloudflare's three options as described November 4, 2026.

Then set the policy, one agent at a time

With an hour of evidence, the choice for each agent is usually clear. Allow the ones that behave like buyers. Block the ones that don't. If you also run content pages funded by ads, such as buying guides, Block on pages with ads lets an agent shop your products without reading pages it would never see the ads on.

If you're adding a new site to Cloudflare, the onboarding now asks whether you monetize pages with ads and suggests policies based on the answer. Treat that suggestion as a starting point, not a decision. A store that sells products and also publishes ad-supported guides has both kinds of page, and the right setting depends on which pages an agent actually visits.

Timing matters too. Holiday traffic is the worst moment to discover that a rule written in the spring is turning away agents at checkout. If you can, do the walkthrough before your peak weeks, and agree in advance who may change a policy during a sale if something looks wrong.

Review it monthly. New agents will appear, and the ones you allowed will change how they behave. Cloudflare says its next goal is helping sites carry user context through agents: whether the person is an existing customer, what they authorized, and who is accountable when something goes wrong. That's worth watching. For now, knowing which agent did what is already a big step up from guessing.

If your store needs help reading this traffic, fixing the pages agents trip on, or getting checkout ready for buyers who never see it, see our ecommerce work or talk to us.

Related work: Paddle8

twimco.

We build alongside the people doing the work.
Bring us the problem you’re thinking about.

Related work

Paddle8 ↗︎