Agentic Marketing

How to Improve Website Engagement With AI Agents: 3 Changes That Nearly Doubled Ours

Lauren McHugh
November 16, 2026
Summarize using
SHARE

Last updated: November 16, 2026

To improve website engagement with AI, move the agent out of the corner chat widget and into the content buyers are reading. On docket.io, building the agent into the page lifted the share of visitors who talked to it from 8.31% to 15.62%. Agents that speak up beside relevant content saw similar lifts.

Most B2B websites still run AI Marketing Agents the way they ran live chat in 2019: a bubble in the bottom-right corner that waits for someone to click it. Buyers don't. They read the page, form a question, and leave with it unanswered. Between August and November 2026 we shipped three changes to where Docket's agent shows up on a page, and measured each one against a matched window before it.

Why don't website visitors use chatbots?

Website visitors ignore chatbots because the widget sits outside what they're reading and asks them to stop and start a new task. A greeting like "How can I help?" has no connection to the paragraph the buyer is on, so it reads as an interruption, not an answer.

The model behind the widget usually isn't the problem. Before our first change, docket.io already ran an AI agent that could answer product, pricing and integration questions. Only 8.31% of tracked visitors talked to it. Same agent, same knowledge, wrong place on the page.

If your website traffic isn't turning into conversations, start with placement before you rewrite prompts or swap vendors. Our guide to why B2B website traffic isn't converting into pipeline covers the other four places it leaks.

What happens when you build the AI agent into the page?

When we built the agent into the page, the share of visitors who talked to it rose 88%, from 8.31% to 15.62%. That's 7.31 percentage points, measured on docket.io across four matched weeks on each side of the August 13 launch (Jul 16 to Aug 12 vs. Aug 13 to Sep 9).

The change was Frameless, Docket's in-page conversational agent experience. Frameless puts the agent inside the homepage layout instead of a floating widget, so asking a question is part of reading the page. Visitors could still open the agent from the launcher, and those conversations count in the 15.62% too. Introducing Docket Frameless shows how it looks on the page.

Do proactive chat prompts increase engagement?

Proactive prompts increase engagement when they're tied to what the visitor is looking at. After we rolled out Contextual Callouts broadly, callouts went from starting 23.6% of all agent conversations on customer sites to 43.9%. That's a 20.3-point gain across two full weeks on either side of the rollout (Aug 3 to 16 vs. Aug 24 to Sep 6), with the rollout week left out.

Contextual Callouts are short prompts Docket's agent shows next to a specific piece of page content, phrased around that content rather than a generic greeting. They also produced more leads. The share of callout-started conversations flagged as a lead went from 0.20% to 1.32% across the same windows. The starting base was small (four leads), so treat the rate change as a strong early signal, not a benchmark.

Can an AI agent point buyers to the right part of a page?

Yes. Spotlight, a Docket agent feature, highlights the sentence on a page that matters most to each visitor and opens the conversation from there. After the October 27 update, the share of docket.io visitors who saw a Spotlight highlight went from 11.4% to 18.7%, a 7.3-point gain in equal windows around the update.

The share of visitors who opened a highlight rose 76%, from 0.60% to 1.05%. Opens are still a small slice of traffic, so reach is the stronger number for now. Introducing Docket Spotlight explains how the agent picks the sentence.

How do the three placement changes compare?

The closer the agent sits to what the buyer is reading, the more conversations it starts. Each change below is measured against a matched window before it, starting from the corner chat widget.

ChangeWhat we measuredBefore → afterLift
In-page agent (Frameless)Replaced the corner chat widgetVisitors who talked to the agentdocket.io, Jul 16 to Aug 12 vs. Aug 13 to Sep 98.31% → 15.62%Before: corner widget+7.31 pts+88% relative
Contextual CalloutsPrompts beside specific page contentConversations started by a calloutCustomer sites, Aug 3 to 16 vs. Aug 24 to Sep 623.6% → 43.9%+20.3 pts+86% relative
SpotlightHighlights the sentence most relevant to the visitorVisitors who saw a highlightdocket.io, equal windows around Oct 2711.4% → 18.7%+7.3 pts+64% relative

A corner widget is still a reasonable choice for support-heavy sites, where visitors arrive with a problem and go looking for help. For B2B buyers who are still evaluating, the agent has to meet them in the content.

Which metrics show website engagement is improving?

The metric that matters is the share of distinct visitors who have a conversation with the agent, not page views or time on site. We track three numbers for every change:

  1. Conversation rate: distinct visitors who talked to the agent, divided by distinct tracked visitors.
  2. Entry point share: which surface started each conversation (launcher, in-page, callout, highlight), so you can see what's doing the work.
  3. Lead capture rate by entry point: the share of conversations from each surface that turn into a lead.

Conversation volume only matters if it reaches sales. Docket, The Demand Platform, qualifies those conversations into AQLs (Agent Qualified Leads): leads qualified by what the buyer said to the agent, not by a form score. Our post on AI marketing agent metrics that predict pipeline covers the numbers after the conversation, and chat-to-demo conversion rate benchmarks covers the step from conversation to meeting.

FAQ

What is a good chatbot engagement rate for a B2B website?

It depends on traffic mix, so a single industry number is misleading. On docket.io, 8.31% of tracked visitors talked to our agent when it lived in a corner widget, and 15.62% after we built it into the page. Measure your own baseline first, then judge each change against it.

Is an in-page AI agent just a chatbot in a different spot?

No. A rule-based chatbot follows scripted flows, while an AI agent answers open questions from your product knowledge and qualifies the buyer as it goes. Placement decides whether the conversation starts at all. Our comparison of chatbots vs AI agents covers the rest.

Do proactive prompts annoy website visitors?

Generic pop-ups can. Prompts written about the content a visitor is already reading performed well for us: callout-started conversations nearly doubled their share of all conversations after broad rollout. Keep each prompt specific to the page section it sits beside.

How long does it take to deploy an in-page AI agent?

Docket deploys in 7 to 14 days, including Frameless and Contextual Callouts on your existing pages.

What does Docket do?

Docket is The Demand Platform. Its agents engage visitors on your website, answer their questions, and turn the conversations into AQLs your sales team can act on.

Want to see what an in-page agent would do on your site? Book a demo.

The Trust Gap

New research fielded by Heinz Marketing on behalf of Docket, based on 192 marketing, demand gen and RevOps practitioners. Three handoffs, one number almost nobody believes, and standard any new lead source will have to meet.
Check it out