Subscribe
Share
Share
Embed
Welcome to our podcast, where we dive into everything Go High Levelβfrom mastering the basics to tackling the most complex tasks. I use GHL daily in my business and rely on Google NotebookLM to stay ahead of the curve, keeping up with all the latest GHL features, tools, and innovations. This podcast is powered by AI, fueled by the research and insights I personally curate to bring you the most valuable and up-to-date content.
Copy this link for a free trial of Go High Level - https://www.gohighlevel.com/highlevel-bootcamp?fp_ref=amplifi-technologies12
Welcome everyone to the deep dive. It is, uh, genuinely great to have you with us. Yeah, thanks for tuning in. And before we get into the thick of today's research, I want to give you a massive heads up. We actually have a special offer for you waiting right now in the show notes below. It's a completely free 30-day Go High Level trial. Which is, you know, double the standard trial length you're going to find anywhere else. Exactly. Double the time. So as we go through these concepts today, if you feel that itch to, you know, actually get your hands dirty and build what we're talking about, go click that link and grab your 30 days. It's really the best way to learn, honestly, just getting into the sandbox and, you know, experimenting with it yourself. Absolutely, because, well, let's be real for a second here. When you think about scaling a digital marketing agency, the mental image is usually this, uh, this beautiful, clean, upward trending graph on a whiteboard. Right. Everyone loves the whiteboard graph. You visualize more clients, higher retainers, better margins. Yeah. But the reality of acquiring those clients is, it's significantly messier. Suddenly, that clean whiteboard graph is just covered in red ink. Oh, completely. Why? Because scaling an agency often means you are suddenly drowning in this multi-channel communication bottleneck. Mm-hm. You've got SMS notifications pinging, Facebook Messengers blowing up, Instagram DMs are rolling in, and, uh, webchat windows are demanding instant attention. It creates an almost impossible standard for human operators. I mean, let's face it. You scale your lead generation campaigns, which is great, obviously, but you kind of forget to scale the capacity to actually have meaningful instantaneous conversations with those leads. Right. And the drop-off in conversion rates when you don't respond within like five minutes, it is staggering. Exactly. You're pouring all this money into the top of the funnel, only to just let it leak out the bottom because your team, you know, needs to sleep occasionally. Why not detail? Right. Yeah. So what if you could deploy an employee who never sleeps? Someone who knows every single one of your clients' obscure business policies, responds within seconds, and, uh, actively books meetings while you're literally out to dinner. That is the dream. It is. And today our mission is to unpack exactly how you build that entity. We are pulling from a massive stack of official Go High Level documentation, alongside some really highly specialized expert implementation guides from the team at Consult Evo. Yeah, those guys are gold. They really are. We are going to master the conversation AI setup. The goal here is to give you, the agency owner, the exact step-by-step mechanisms to configure, train, and deploy these AI assistants, without overwhelming your clients, and crucially, without overwhelming yourself. It's a vital mission, really, because this technology is wildly capable, but, um, if you don't structure it correctly, it's not an asset, it's a liability. Totally. It can feel like trying to drink from a fire hose if you don't have a blueprint for the underlying architecture. Okay, let's unpack this starting with that architectural foundation. Yeah. Because before you can expect an AI to perform, you have to give it a highly structured workspace. If you just open a blank prompt window and say like, "Hey, be a good bot." You are asking for trouble. Oh, definitely. You're basically begging for a disaster. Right. So the Go High Level documentation outlines something called the guided form-based setup. It essentially acts as a, uh, a guardrail system. Yeah, which is super necessary. It forces you through these sequential screens to establish the basic details, operational behavior, FAQs, and, you know, routing protocols. I like to think of it as creating a standardized, absolutely foolproof onboarding package for a brand new human hire. That analogy holds up perfectly, actually. Think about onboarding a new front desk receptionist. You wouldn't just point them to a ringing phone, shrug, and walk away. Good luck. Right. You would hand them an operations manual. You'd outline the company turn, the hours of operation, the escalation path for angry customers, the core services, the basics. Exactly. The guided setup in the software forces you to codify all of that abstract business knowledge into concrete parameters. And from an agency perspective, the mechanism here is crucial because it standardizes setups across all your sub accounts. Oh, that makes sense. Yeah. If you manage 50 different client locations, you aren't guessing what settings you applied to which account. The guided form ensures uniformity across the board. I can see how that solves the immediate configuration chaos, for sure. But the Consult Evo guides emphasize that this isn't just one massive monolithic bot handling everything. There's an entire hierarchy to how you deploy this. What's fascinating here is that Go High Level supports a true multi-bot ecosystem. Right. You are building an automated department, not just a single widget. The architecture dictates that you designate one primary bot, and then you have the capacity to deploy multiple non-primary bots. Okay, wait. I need to push back on that a bit, though. Why introduce that level of complexity? Well, I mean, if the AI is smart enough to read the whole company manual, why can't one bot just handle every single message that comes in? It comes down to context and workflow specificity. The primary bot is your generalist. It sits at the front desk. Its job is to monitor all general inbound traffic across your assigned channels, whether that's SMS or Instagram or whatever that isn't already part of an active specific workflow. So, it's fielding the random walk-ins. Exactly. It handles the initial triage, the greetings, the basic FAQs. But non-primary bots are your highly specialized agents. You assign them to specific channels based on explicit workflow. Give me an example of that. Let's say you have a sophisticated campaign designed strictly to re-engage cold leads from a highly technical webinar you hosted like six months ago. Right. So those leads require a very specific conversational context. Precisely. If the primary bot takes that conversation, it might just give a generic, "Hey, how can I help you today?" greeting. Which is terrible for a cold lead. Right. But if you route that workflow to a non-primary bot trained exclusively on the webinar's content and the specific offer you're making to those cold leads, the bot operates with hyper-specific context. It keeps the automated conversation relevant and highly targeted. That makes a lot of sense, actually. You don't want the general receptionist trying to close a highly technical enterprise deal. Exactly. But, um, let me put my skeptical agency owner hat on for a second here. Even with specialized bots, how do we safely roll this out for a new client? That's the big question. Right. Because if I hand over my client's Facebook Messenger to an AI, and it starts hallucinating, promising free services, or, I don't know, arguing with a frustrated customer, I am losing that client's retainer today. The risk feels incredibly high. It is the single biggest barrier to entry for agencies. Trust. You have to mitigate that risk mechanically. The software addresses this directly through three distinct bot statuses: off, suggestive, and autopilot. Off means exactly what you think. And autopilot is the ultimate goal. The bot reads the intent, formulates the response, and fires it off entirely autonomously. I am definitely not starting a new client on autopilot. That sounds terrifying. And you shouldn't. That is where the suggestive status becomes the most critical tool in your rollout strategy. Okay. Suggestive mode. How does that work? In suggestive mode, the AI still does all the heavy analytical lifting. It reads the incoming message, parses the intent, and generates a fully formed draft response. But crucially, it does not send it. Oh. It places that text right into the message composer within the unified inbox. A human operator, whether that's you or your client's in-house staff, has to read that draft, edit it if there's a slight nuance missing, and manually click the send button. So it's essentially acting as a hyper-efficient copywriter for the person managing the inbox. That's a perfect way to look at it. That is an incredibly powerful actionable takeaway. You use the suggestive status as, like, training wheels. Yes, exactly. The client sees the bot formulating these brilliant, accurate replies, but they retain absolute physical control over what goes out the door. Right. They still hold the keys. You let that run for two weeks. Once you and the client review the logs and realize the bot is nailing the response, you know, 99% of the time, the client is usually the one begging you to flip the switch to autopilot. That's to save him the clicks. Exactly. You are engineering trust. You are proving the AI's competence before you ever remove the safety harness. Okay. So, we've got the architectural structure down. We have our primary and non-primary bots, and we have a safe rollout strategy using suggestive mode. But none of that matters if the bot doesn't actually know anything about the business. It's just an empty shell otherwise. Right. And here's where it gets really interesting, the web crawler. Oh, yeah. The web crawler is the ingestion engine. It's how the AI builds its internal knowledge base. And the high-level documentation states you can feed it up to 4,000 URLs. I want to pause on that, because 4,000 URLs is an astronomical amount of data. You could feed it an entire corporate wiki. You could. It essentially allows the bot to ingest an entire website and instantly become a subject matter expert. But the real power isn't just in the volume. It's in the granularity of how you instruct the crawler to parse that data. What do you mean? Well, you aren't just giving it a domain and hoping for the best. You have three distinct targeting options. First, you can use exact URLs, which commands the crawler to scrape only that singular page and nothing else. So, like, perfect for a standalone pricing page. Exactly. Or, um, if a client has a hidden VIP promotional page, or an internal staff directory on the site, I just use exact URLs for the public stuff and avoid the domain scrape altogether. Yes, to keep the private stuff private. Mm. Then you have the specific path option. Let's say the site architecture has a path like domain.com/services. If you use that path, the crawler will grab every single page nested under services, but it will completely ignore domain.com/blog. Okay, that's smart. And finally, you have the entire domain option, which is just the comprehensive sweep of everything. But wait. What if the client's website is a disaster? We've all onboarded that one client, say, like, a local roofing company, whose website hasn't been updated since 2014. Oh, definitely. It's full of outdated pricing, broken links, contradictory information. If I point the crawler at that, the bot is going to be incredibly confused. And it will give terrible answers. This is where a brilliantly simple workaround from the Consult Evo implementation guide comes into play. Tell me. If the source material is garbage, you just don't use the web crawler on the website. Instead, you use a Google Doc. A Google Doc? Yeah. You sit down with the client, extract their current service descriptions, their actual operating hours, their specific routing rules, and you dump it all into a clean, well-formatted Google document. And then you just feed the Google Doc URL to the crawler. Exactly. You feed that single link to the bot, and it treats it as the ultimate source of truth. That is so simple. But the mechanical requirement here, and you must not forget this, is that the Google Doc sharing settings must be set to anyone with the link. Oh, right. It has to be public. Yes. It has to be public so the crawler's parsing engine can actually access and read the text. That is such a massive headache saver. You basically bypass the client's terrible website entirely and build a custom curated brain for the AI. Exactly. But let's transition from what the bot knows to how the bot speaks. Knowing the refund policy is one thing, but delivering it in the right tone is another. I mean, a luxury real estate bot shouldn't sound like a casual streetwear brand. Right. That brings us to prompting and persona configuration. The guided setup requires you to shape the bot's behavioral goals. You select a personality profile, friendly, professional, or formal. What's the difference in practice? Well, friendly utilizes a more casual syntax, maybe throws in some emojis. Professional is your standard direct business communication. And formal is highly structured, you know, appropriate for legal or financial services. Got it. And you also configure the intent, right? You tell it whether its primary goal is resolving queries, meaning it just acts as a support agent, or generating leads, meaning it's actively trying to steer the conversation toward an appointment. Yes. And I noticed this documentation says the additional instructions field has been expanded to allow up to 2,000 characters. That expansion is huge. 2,000 characters gives you the runway to provide deep, nuanced behavioral conditioning. Like what? You can give it explicit negative prompts like, "Never mention our competitors by name," or conditional prompts like, "If the user mentions water damage, escalate immediately with a sense of urgency." Okay, let me test a theory, then. If my intent is set to generating leads, and I want this bot to aggressively book appointments for my agency, I should just use those 2,000 characters to map out my schedule. Right. I could just type, like, "I'm available on Tuesdays at 2:00 PM, Thursdays at 4:00 PM, and Friday mornings," so the bot knows exactly when to slot people in. This raises an important question and it highlights one of the most critical warnings in the official Go High Level documentation. You must never, under any circumstances, hard code calendar slots or static availability into the bot's prompt instructions. Really? Why not? If the goal is to book meetings, shouldn't I tell it when I'm free? Because of how large language models fundamentally operate, they are predictive text engines, not dynamic database query tools. If you hard code available Tuesday at 2:00 PM into the prompt, the AI treats that as static truth. Okay. It has no mechanism to cross-reference your actual Google Calendar in real-time to see if a human just booked that 2:00 PM slot like five minutes ago. Oh, I see. So the bot will confidently offer Tuesday at 2:00 PM to three different leads simultaneously. Or worse, it gets confused by the user's time zone versus your time zone, and it just starts inventing times that sound plausible. Oh, wow. In the AI world, this is called a hallucination. It will hallucinate availability resulting in massive double booking chaos and furious clients. That is a nightmare scenario. So if we can't hard code specific data to guide it, how do we mechanically ensure it's not going to hallucinate other things like, say, a non-existent 50% discount? You employ real-time iterative testing. Inside of the dashboard, directly adjacent to the training configurations, there is a live chat interface. Before you ever attach this bot to a public-facing channel, you sit there and aggressively interrogate it. You play the bad cop. You act like the most difficult, confusing customer imaginable. You try to break it. And if it does break? Let's say I ask a complicated question about a prorated refund, and it gives me a completely wrong, hallucinated answer. What then? The mechanism to fix it is built right into the test window. When the bot generates that bad answer, you just click the thumbs down feedback icon next to the message. And what does that do? That action triggers an immediate override protocol. It automatically opens a custom FAQ entry field based on the exact question you just asked. You then manually type in the perfect, structurally correct answer. So, I'm essentially hard-coding the correct response for that specific semantic query. Exactly. From that moment forward, if a user asks a variation of that question, the system recognizes the intent, bypasses the LLM's generative engine entirely, and delivers your exact, hard-coded FAQ answer. Wow. It prevents the hallucination from ever happening again. You don't have to guess what people will ask and build 100 FAQs from scratch. You just let the bot fail in the sandbox, thumbs down the errors, build the database dynamically. That is brilliant. It's a self-refining system. Mm-hm. But let's look at the ultimate goal here. A bot that provides perfect customer support is great for retention, sure. But for an agency owner, the ultimate metric of success, the thing that justifies your retainer, is driving new revenue. Always. It's getting qualified leads onto a calendar. If we aren't allowed to let the bot negotiate specific time slots in the chat, how does it actually facilitate the booking? Well, think about it. Negotiating time slots via text is a terrible user experience anyway. True. How's Tuesday? Tuesday's bad. What about Wednesday? Right. It's clunky. Instead, the architecture handles appointment booking actions structurally. In the bot settings, you link the bot directly to a specific calendar within that sub account. Okay. When the AI determines based on its conversation that the user is qualified and ready to book, it doesn't negotiate time. It dynamically generates and sends a direct, trackable booking link. Ah, so it just says, "It sounds like we're a great fit. Click here to find a time on our calendar that works for you." Yes. It puts the power in the user's hands and leverages the actual visual calendar UI. Precisely. But the most crucial mechanical setting in this entire sequence is a feature called pause after booking. Pause after booking. Yeah. Once the Go High Level system registers that a meeting has been successfully confirmed via that specific link, it sends a signal to the conversation AI to immediately disable any further automated responses for that specific user's conversation thread. I cannot overstate how important that is from a sales psychology perspective. It's sales 101. Mm-hm. Once the prospect says yes and signs the contract, you stop pitching. Yeah. You shut up. If a lead books a consultation and then sends a follow-up text saying, "Hey, really looking forward to it." You do not want the bot jumping back in with a cheerful generic, "Can I help you with anything else today?" Oh, that's the worst. It completely breaks the illusion. It creates a jarring, robotic post-sale experience. It completely devalues the interaction. Yeah. By toggling pause after booking, the bot achieves its goal and steps back, leaving the floor open for authentic human follow-up. But let's dig into a more complex scenario. What if the prospect doesn't make it to the booking link? What if they hit the bot with a highly technical, multi-layered question about like API integrations that goes way beyond the Google Doc we fed it? If we connect this to the bigger picture, this is where you transition from having a chatbot to having an automated enterprise workforce. Okay, listening. The platform features an action called transfer to employee. When the bot detects an intent or a level of complexity it cannot resolve, it triggers a handoff. But the paradigm shift here is that the employee receiving the transfer does not have to be carbon-based. Wait, really? The bot can seamlessly hand off the conversation to another specialized AI bot. Wait, I want to make sure I understand the mechanics of that. Bot-to-bot handoffs. Yes. Think back to the primary and non-primary architecture. Your primary front desk bot is chatting with a lead. The lead suddenly starts asking deep, technical questions about your highest tier enterprise service package. Right. The primary bot recognizes the intent shift. It executes a transfer protocol, routing the conversation away from itself and over to a non-primary bot that has been exclusively trained on the technical specifications of that enterprise package. That is, that's an invisible digital office floor. The receptionist literally walked the client over to the senior engineer's desk, all within milliseconds without the client ever knowing they changed representatives. It creates a frictionless experience for the user while maintaining deep subject matter expertise. But, and this is important, to maintain that illusion of a seamless, human-like interaction, you have to utilize the advanced pacing settings. Right. Because if it answers instantly. Exactly. You adjust the wait time before responding. Because if the user types a four-paragraph explanation of their business problem and the bot replies with a perfect nuanced solution in .1 seconds, the jig is up. Humans don't read or type that fast. It feels unsettling. Exactly. You program a slight randomized delay. The bot thinks for a few seconds. It makes the cadence of the conversation feel natural. You also implement maximum message limits, which mechanically prevents the bot from getting caught in an endless, expensive loop with a troll or, you know, a spam bot on the other end. Good safeguard. The Consult Evo notes also highlighted a feature called terminology alignment. On the surface, it sounds like a minor UI tweak, but for an agency managing multiple niches, it seems vital. It drastically reduces cognitive friction. Agencies constantly rename standard CRM objects to fit their client's industry. A real estate agent doesn't want to see a column for contacts or opportunities, they want to see buyers and listings. Makes sense. Terminology alignment means the conversation AI automatically ingests and respects those custom object names within its internal logic and UI. So you aren't mentally translating Go High Level's default backend jargon into your client's industry jargon while trying to review bot transcripts. Right. Speaking of transcripts, the new context memory is fully integrated now, too. The bot parses the entire history of the chat automatically. Enabled by default, yeah. If a user states their budget in message two, the bot won't redundantly ask for their budget in message eight. It maintains conversational continuity. Okay, so let's tie the knot on this workflow. The bot navigated the conversation smoothly. It remembered the context. It generated the booking link. The user picked a time, and the bot automatically paused itself. Yep. Is that the end of the line for the automation? If you stop there, you are basically leaving money on the table. The final actionable takeaway for agency owners is mastering the trigger workflow after booking capability. Okay. Getting the meeting on the calendar is only half the battle. Ensuring the prospect actually shows up is the other half. The show rate. So important. Exactly. The millisecond that appointment is confirmed, the conversation AI needs to trigger a secondary Go High Level workflow. That external workflow takes over, immediately firing off a personalized confirmation email, queuing up an SMS reminder for 24 hours prior to the call, and crucially, dropping a notification into your agency's internal Slack channel saying, "Hey, meeting booked. Here is the full AI chat transcript for context." It is the ultimate baton pass from the AI to the human sales team. But from a setup perspective, the documentation is very clear. You have to build that destination workflow first, publish it, and then go back into the bot settings to select it from the dropdown. Correct. You must lay the train tracks before you tell the engine to accelerate. So, what does this all mean for you, the agency owner listening right now? Let's take a step back. We're no longer talking about a generic little chat bubble that pops up on a website and says, "Please leave your email." Not at all. Go High Level's conversation AI is a multi-channel, dynamically trained, multi-bot automated workflow. It reads your exact operational documents. It bypasses hallucinations through real-time FAQ overrides. Yeah. It can be throttled back in suggestive mode to build client trust, or it can be fully unleashed on autopilot to qualify leads and secure revenue while your entire human team is offline. It represents a fundamental mechanical shift in how you scale an agency. You are entirely severing the link between communication capacity and human headcount. It really is incredible. But it also opens up some wild possibilities for where this technology is heading next. It does, and I want to leave everyone with a final thought to really mull over. We spent some time unpacking the bot-to-bot handoff capability. The transfer to employee feature where the AI hands a conversation to another specialized AI. Right. If the underlying architecture to route tasks between specialized non-human agents exists today, how long until we see the emergence of the entirely invisible agency? Oh, wow. I'm talking about an ecosystem where an AI outbound lead gen bot initiates a conversation, hands the qualified prospect off to an AI closing bot, which secures the payment and immediately triggers an AI fulfillment bot to execute the digital service. That's crazy. They collaborate, share context, and execute deliverables entirely behind the scenes. Are we rapidly approaching a reality where the role of the agency owner shifts entirely away from being a manager of people and becomes strictly the architect of an interconnected AI ecosystem? The invisible agency. Right. That is equal parts terrifying and exhilarating. Yeah. But, you know, if that is the future, you have to master the foundational architecture right now. You cannot build the invisible agency if you don't even know how to deploy the receptionist. Exactly. Start with the basics. So, to help you take that first step, let me remind you one final time. The link for the Airva 30-day Go High Level trial is sitting right there in the show notes. Double the standard length. Go get it. That gives you a full month to build your master Google Doc, configure your primary bot, put it in suggestive mode, and watch it draft responses with zero financial risk. Go click that link. Get into the software and start building your automated workflow. Thank you so much for joining us on this deep dive, and we will see you next time.