AI Chat lead capture — when to ask for email, and especially when not to

The most common mistake people make with an AI chat: a modal pops up on page load and asks for an email before the visitor sees anything. That is 2018 newsletter-popup logic in a new outfit. It does not work in 2026 either.

The anti-pattern

The base situation is familiar: a new visitor lands, 1.2 seconds later a chat panel pops up saying "Hey! Before you look around, give us your email so we can send you a personalized offer!" 91% of visitors X it instantly. Of the remaining 9%, about 2% give a real email — the rest type aaa@aaa.aaa.

We measured it: this pattern hurts the storefront's visitor experience so much that the cart abandonment rate goes up by 3.4% compared to the no-popup state. And the captured emails do not bring more revenue long-term, because 78% of submitted emails never reply to the follow-up.

The good pattern

Three simple rules that Nortinia AI Chat's default config follows:

1. Wait for the 2nd interaction

The visitor sent a question, got an answer. That is one interaction. Now they send another — anything. After that, if the context warrants it, we can ask for email. Why? Because by then there is proven intent. The visitor did not just click into something by accident.

2. Or wait until an async request

There is a situation where asking for email is natural and expected: the visitor asks for something that cannot be delivered immediately. "Send me the detailed spec as a PDF", "Notify me when product X is back in stock", "I want a quote on the Pro plan for 50 users." Those are async requests. Without an email they cannot be fulfilled — the visitor offers it voluntarily if we ask well.

The chat's wording matters. Bad: "Give me your email!". Good: "Where should I send it? If you share your email, the PDF will be in your inbox within 2 minutes."

3. Do not ask if already logged in

Trivial, but surprisingly many systems get this wrong. If the visitor is known (cookie, JWT, account session), there is already an email on the backend. A duplicate request is annoying and erodes trust.

The A/B experiment results

Three tenants, four-week A/B run:

• Variant A (anti-pattern): immediate email prompt on page load. 8,231 sessions, 412 emails submitted, 5.0% capture rate. Of 412, 38 replied to a later follow-up — quality rate: 9.2%.
• Variant B (wait for 2nd interaction): only after an interaction. 8,118 sessions, 1,668 emails submitted, 20.5% capture rate. Of 1,668, 567 replied to a later follow-up — quality rate: 34.0%.

Variant B captures 4.1× more emails, and the emails are 3.7× higher quality (by reply rate). In other words: less harassment → more trust → more and better leads.

The technical setup

Nortinia AI Chat controls this with a per-tenant policy object. The default:

{
  "leadCapture": {
    "strategy": "on-async-request",
    "minTurnsBeforePrompt": 2,
    "skipIfAuthenticated": true,
    "copy": {
      "hu": "Hova küldjem el?",
      "en": "Where should I send it?"
    }
  }
}

The tenant can override (e.g. on some B2B surfaces a flat "ask for email after 4 minutes if still here" strategy also works — but decide that from strict measurement, not instinct).

The lesson

Lead capture is not the chat panel's function. Lead capture is the chat panel's byproduct if you communicate well with the visitor. If you flip that, the chat panel becomes another barrier to purchase — not a helper.