Look, I love AI. I love automation. I love when bots do stuff I don’t want to do.
But if you’ve ever tried setting up an AI chatbot in HighLevel and thought, “Wait… did this thing just lie about my pricing?” — you’re not alone.
We recently hosted a super actionable Zoom session with Mario Aldayuz (our guy heading up Partnerships at HL Pro Tools), where we pulled back the curtain on the real-world challenges and fixes for using AI in HighLevel.
Here’s what we discovered — and how you can make your bot actually sound smart instead of just… confidently wrong. 😅
Problem #1: “My AI is hallucinating.”
Yeah, it’s a thing.
AI hallucinations happen when your bot just makes stuff up. This usually happens when the AI gets confused by unclear training data or overly complex documents.
🔥 The Fix: Use clean, simple, focused docs.
Instead of dumping 20 Google Docs and hoping for the best, try this:
- ✅ Create one clean FAQ document
- ✅ Use clear headers + short answers
- ✅ Leave out anything you don’t want the bot to say (especially pricing)
🛑 Pro Tip: Don’t train your AI on pricing unless you want it quoting it.
Just take it out entirely unless it’s absolutely necessary. One user had bots randomly quoting $297/month when that wasn’t even an option. Why? Because it was buried in an old doc somewhere. Don’t be that guy.
Training AI is like onboarding an intern
This hit home during the session:
Your AI bot is basically a really eager, very literal intern.
It doesn’t know what you meant — it only knows what you gave it. So if your docs are messy or contradictory, guess what? It’s gonna mess up.
Mario shared this example:
“If you’re training a bot for a local dog groomer, give it one page with super clear info. Not five tabs with every Yelp review, service detail, and outdated promo code from 2021.”
The cleaner the doc, the better the bot.
Bonus Tips from the Session
Here’s some extra gold that came out of our convo with folks from Progreda, GHL Engine, and the Syndicate:
- 🧠 Use specific prompt language. Don’t just say “help the user.” Say, “If the user asks about hours, respond with X.”
- 🧹 Declutter your training inputs. The more concise your info, the less room there is for the AI to get confused.
- 👥 Test your bot like a customer would. Don’t assume it works — actually try it.
The Big Idea
AI in HighLevel isn’t plug-and-play — it’s train-and-test.
When you treat your bot like a team member and give it the right resources, it’ll perform like a champ. But if you just toss in messy docs and hope for the best? You’re setting yourself up for frustration.
If you want help dialing this stuff in (plus automation tools that actually work), go check out what we’ve built at hlprotools.com — it’s like having a HighLevel cheat code.
Cool Free Thing: Get My Testimonial Workflow 👇
Before you can help businesses with your service, you’ve gotta build trust.
And nothing builds trust faster than testimonials. ⭐️
In the spirit of free stuff, I want to give you my team’s Testimonial Workflow — the exact process we use to gather, edit, and organize testimonials for clients.
🚀 It’s fast, simple, and just works.
