On October 15, cybersecurity vendor DataDome released its Q3 AI traffic report. One number stands out: requests from agentic browsers, the AI agents that browse and act on a website on someone's behalf, grew 4.7x between August and September 2026.
And these agents don't just read your pages. They fill in your forms, starting with your demo request or your contact form.
Here is what is happening, what it means for your website and your CRM, and the position we take at if/else agency: don't block these agents on reflex, get ready to welcome them 👇🏼
For two years, agents that "browse for you" were mostly demos and betas for early adopters. In September, they landed in the pockets of the general public.
On September 8, Meta introduced Muse, a personal agent that is free for most uses, rolling out in the US on iOS, Android and the web. According to Meta's announcement, Muse has its own browser and can "open a browser, fill out forms, and negotiate" on its user's behalf. It checks with the person before sensitive actions, such as sending an email or making a purchase.
The impact was immediate. According to PPC Land and Help Net Security, which report on DataDome's findings, agentic browser requests went from 12.5 million in August to 58.2 million in September. Muse already accounts for 33% of that, even though it was only measured for nine days of the month. PPC Land also reports that ChatGPT's cloud browser makes up 47% of the total, and Claude's browser extension 10.5 million requests.
On October 13, Contentsquare announced a new version of its analytics platform that separates human sessions, LLM referrals and agent interactions. When a vendor that says it runs on more than 1.3 million websites builds a product around this, it means its customers are asking.
The most telling detail in the report, again according to PPC Land and Help Net Security, is what these agents do once they reach a site. In a sample of Muse traffic, 8.9% of requests hit sensitive pages: login, payment, account creation, cart and forms. Of those sensitive requests, 81% targeted forms.
In B2B, the form is where business starts. It creates the contact in your CRM and kicks off nurturing. When an agent fills it in, your database gets the result.
A word of caution: these figures come from a vendor that sells bot management, and DataDome doesn't say how many sites its network covers. The trend is clear. The exact volumes still need confirmation.
When automated traffic shows up, the usual reflex is to tighten defenses, with CAPTCHAs everywhere and data center IP addresses blocked. But those tools were built to stop spam, not to sort out legitimate visitors.
An agent filling in your demo request is, in most cases, working for a real buyer. Block it, and you shut the door on the customer who sent it.
The same articles note that 99.3% of the Muse requests in DataDome's sample came from cloud infrastructure. A filter that only trusts residential connections will flag those visits as noise. And since Muse doesn't identify itself, PPC Land adds, you won't find its name in your logs.
Our position is simple. On the pages that generate business, the agent is a visitor to welcome. On the pages that expose data (customer portal, payment), security comes first. Putting the whole site under one rule, either way, is a mistake.
An agent "looking" at your page trips over the same things a screen reader does: fields without labels, placeholders used instead of labels, custom dropdowns, vague error messages. Good news: if you've worked on the accessibility of your forms, you're halfway there. Our article on WCAG compliance levels covers those rules.
On the browser side, Google is also testing WebMCP, a proposed standard that entered origin trials in Chrome 149, according to InfoQ. It lets you add attributes to HTML forms that tell agents what a form is for and how to fill it in. It's still experimental: worth watching, with no rush to implement it.
If an agent visits 15 pages in 40 seconds and then submits a form, your conversion rate and time on page are skewed. Until you separate those sessions, your design and content decisions rest on blended numbers.
A form filled in by an agent lands in HubSpot like any other. The contact triggers the same workflows and earns the same scoring points. The result: MQLs that may not be MQLs at all, and sales reps calling someone who doesn't remember filling in a form.
Here's a typical case. A procurement director asks their agent to "find three web agencies and request a quote." The agent fills in three forms in two minutes. The first agency's form has twelve required fields and a CAPTCHA, so the agent gives up. The other two have short, well-labeled forms, and the request goes through. The first agency will never know it was dropped from the shortlist.
We don't promise to detect every agent. Today, nobody can do that reliably. What we recommend is a flow that works whether a human or an agent fills in the form.
For a newsletter sign-up, HubSpot's double opt-in has the person confirm their marketing consent themselves, which also helps with GDPR. For a demo or quote request, a workflow sends a confirmation email, so the sales rep knows a human is waiting for the call.
The source property relies on likely signals from your bot management tool or your analytics. It flags a lead for review and is not proof.
This work touches both your page design and your CRM setup. That's why we handle them together, in our web design projects as well as our CRM and marketing automation work.
AI agents went from demo to mass market in a month. And among sensitive pages, forms come first: the exact place where your website generates business.
Block them by default and you lose legitimate requests. Let them in without changing anything and you pollute your CRM. The right answer is readable forms and an email confirmation, with the source tracked in your CRM.
To dig into what makes a page convert, read our guide on how to design a website that drives conversions.
Are your forms ready for AI agents? Let's talk.