Chatbots that don’t kill your conversion rate

Half the AI chatbots we audit on US founder-led sites are net-negative on conversion. The button-pop, the friendly intro message, the LLM that confidently misquotes your pricing — it all costs money. Here’s how to deploy chat that actually helps.

The default chatbot deploy is bad

You install Intercom or Drift, drop in a generic widget, point an LLM at your help docs, and ship. What you’ve actually done is added an interruption to a page that was working. The proactive pop-up draws attention away from your CTA. The bot answers a question the buyer wasn’t asking. Conversion drops 5 to 15% and nobody connects the dots.

The four rules for chat that helps

• No proactive pop-ups on landing pages.Let the buyer get to the CTA first. If they want chat, they’ll click it. If they don’t, you haven’t cost yourself the conversion.
• Chat lives on support pages, not sales pages.Your pricing page should not have a bot. Your help center should. Match the channel to the intent.
• Don’t pretend the bot is human.Buyers smell it instantly and trust drops. "I’m a bot — I can answer X, Y, Z, or pull in a human" is the highest-converting intro we’ve tested.
• Hand off fast.If the bot can’t answer in two turns, get a human in. The cost of a slow handoff is way higher than the cost of an early one.

Where AI chat genuinely wins

Post-purchase support, status questions, shipping, returns, documentation lookups. Anything where the answer is in your data and the buyer is already a customer. Conversion is no longer the metric — resolution time and CSAT are. Bots are great at this. Use them here.

Where AI chat usually loses

• Top-of-funnel discovery. Buyers want to read your page, not interrogate a bot.
• Pricing questions.If your bot quotes the wrong price, that’s a refund, a Twitter post, and a churn risk.
• Anything with a regulatory edge.Healthcare, fintech, legal — an LLM hallucinating here is an actual liability, not just bad UX.

The grounding problem

Most bots fail because they’re grounded in stale or contradictory docs. Your help center says one thing, your pricing page says another, the founder said a third on a podcast. The LLM averages them and lies confidently. Fix the docs first, then deploy the bot. Or accept that you’re shipping a bot that occasionally embarrasses you in front of a SF buyer at 2am.

How to measure whether it’s working

A/B test with the bot on and off. Look at conversion rate on the pages where it lives. If conversion is up, keep it. If conversion is flat or down, the bot is a feature for you, not your buyers. Be honest about which one is true.

How we help at The Nerdish Mic

We deploy AI chat for US founder-led brands — the kind that helps buyers and doesn’t cost you conversion. Grounding, copy, handoff, measurement, all in. If your current bot is embarrassing, we’ll fix or remove it.