Latest / AI Fire Daily / #601 Neil: Gemini AI Prompts To Build Your Smart Business Assistant

Transcript

#601 Neil: Gemini AI Prompts To Build Your Smart Business Assistant · AI Fire Daily

• 0:00
  There is a strange thing that happens when people
• 0:02
  use AI. Two -sec silence. They open a blank chat
• 0:07
  window for their business. They type in a single
• 0:10
  simple sentence, and then they get frustrated
• 0:12
  by the incredibly generic response. Yeah, they
• 0:16
  really do. People say Gemini just gives totally
• 0:18
  boring, average answers. But maybe the problem
• 0:21
  isn't actually the machine. It usually isn't.
• 0:24
  Maybe the problem is that we expect to read our
• 0:26
  minds. We treat it like a search engine instead
• 0:29
  of a reasoning engine. We don't give it the right
• 0:31
  context first. Yeah, exactly. I mean, we hand
• 0:33
  it a completely blank slate. We provide absolutely
• 0:36
  zero background about who we are. Then we act
• 0:39
  totally surprised when the response is robotic.
• 0:42
  Welcome to today's Deep Dive, everyone. Glad
• 0:44
  to be here. We have a really fascinating guide
• 0:46
  to explore today. It outlines a clever six -step
• 0:49
  framework for using Gemini. The mission here
• 0:52
  is transforming a basic forgetful chat bot. Right.
• 0:56
  We are turning it into a highly personalized
• 0:58
  business assistant. A true partner, really. Exactly.
• 1:02
  We are going to walk through this entire process
• 1:03
  together. First... We transfer your business
• 1:07
  brain into the AI system. Then we build a highly
• 1:11
  customized business knowledge base. We establish
• 1:14
  a dedicated workspace and refine your brand voice.
• 1:17
  That brand voice part is so cool. It really is.
• 1:20
  Then we set up some very strict operational guardrails.
• 1:24
  Finally, we automate your daily recurring business
• 1:27
  workflows. It is a complete overhaul. Okay, let's
• 1:30
  unpack this entire process from the beginning.
• 1:32
  You can't ask a total stranger to run your business.
• 1:34
  No, definitely not. You have to introduce them
• 1:36
  to the company first. Right, and that requires
• 1:38
  a comprehensive business handover document. A
• 1:42
  handover document. Yeah. Think about your past
• 1:45
  chats with Claude or ChatGBT. You have probably
• 1:48
  already shared a lot of business context over
• 1:50
  time. Oh, for sure. You have explained your products
• 1:53
  and your general philosophy. You can actually
• 1:55
  ask them to write this handover document for
• 1:57
  you. Oh, wow, really? Yeah, you don't even have
• 1:59
  to write it yourself. You tell the AI to write
• 2:02
  entirely in the second person. It should speak
• 2:04
  directly to your new Gemini assistant. That is
• 2:06
  brilliant. It's essentially the ultimate new
• 2:09
  hire employee onboarding packet. Exactly like
• 2:12
  that. You include your target customers and your
• 2:14
  core business goals. You also outline your specific
• 2:18
  nuanced decision making style. You even tell
• 2:21
  it what to avoid promising entirely. I still
• 2:24
  wrestle with prompt drift myself. We all do.
• 2:26
  It is incredibly hard to keep AI on track during
• 2:29
  long conversations. It starts strong, but eventually
• 2:33
  forgets its original persona. Yeah, it just wanders
• 2:36
  off. You really need this foundational document
• 2:38
  to permanently anchor it. Right. And there is
• 2:41
  a crucial refining step here. You take that drafted
• 2:44
  document from Claude or ChatGPT, then you feed
• 2:47
  it directly into Gemini itself. OK, so you move
• 2:50
  systems. Yeah. You command Gemini to review it
• 2:52
  as a rough draft. The most important instruction
• 2:55
  is about handling missing information. You tell
• 2:57
  Gemini to mark anything unclear as Unconfirmed.
• 3:01
  Unconfirmed. Why is that specific instruction
• 3:04
  so critical for the AI? What is the specific
• 3:07
  danger of skipping this unconfirmed labeling
• 3:11
  step? Because AI models inherently want to fill
• 3:14
  in the blank spaces. Right. They are predictive
• 3:16
  text engines designed to be incredibly helpful.
• 3:19
  If you don't explicitly forbid it, the AI will
• 3:23
  guess. It just makes things up. Exactly. It wants
• 3:26
  to give you a complete answer right away. It
• 3:28
  might invent a business model or a fake pricing
• 3:31
  structure. You want to completely eliminate those
• 3:34
  dangerous assumptions upfront. forcing it to
• 3:37
  say unconfirmed, you change its primary goal,
• 3:40
  you make accuracy more important than just providing
• 3:42
  a complete sentence. So we force it to admit
• 3:44
  ignorance instead of hallucinating facts. Exactly.
• 3:47
  You give it permission to say, I don't know.
• 3:49
  To sex silence. OK, so Gemini knows exactly who
• 3:52
  you are now. Yes. But knowing who you are isn't
• 3:54
  enough to actually do the work. It needs the
• 3:56
  actual raw data to build its answer. It needs
• 3:59
  the raw materials, yeah. It needs the raw materials
• 4:01
  to start generating value. This brings us to
• 4:03
  step two, building the business knowledge base.
• 4:06
  The knowledge base? You need to organize your
• 4:09
  internal documents into three clear buckets.
• 4:12
  OK, what are the buckets? The first bucket is
• 4:14
  exactly what you sell. Like products? Yeah, this
• 4:16
  includes product features, tier pricing, and
• 4:19
  overall market positioning. The second bucket
• 4:21
  is exactly who you sell to. The audience. Right.
• 4:25
  This goes beyond demographics into actual customer
• 4:28
  pain points. It includes their midnight anxieties
• 4:30
  and their daily frustrations. The deep psychological
• 4:33
  stuff. Exactly. And the third bucket is exactly
• 4:35
  how you deliver your value. This is your internal
• 4:39
  workflows, logistics, and daily processes. I
• 4:42
  really love the analogy the guide uses for this
• 4:45
  part. The Legos. Yes, the Lego blocks. It's like
• 4:48
  giving the AI a very specific bucket of Lego
• 4:51
  blocks. You are telling it to build whatever
• 4:53
  marketing strategy it wants, but it can only
• 4:56
  use the specific blocks in this bucket. Right.
• 4:59
  It can't go outside the bucket. It cannot pull
• 5:01
  random bricks from the rest of the internet.
• 5:03
  If it needs a red brick and you only have blue
• 5:05
  ones, it has to stop and ask you for a red brick.
• 5:08
  It cannot just invent a red brick out of thin
• 5:10
  air. Right. because that ruins the whole structure.
• 5:13
  You instruct Gemini to only use these attached
• 5:16
  documents. It must cite the specific source document
• 5:18
  for every factual claim. But business data is
• 5:22
  inherently messy over time. Oh, totally messy.
• 5:24
  What if two uploaded documents contain directly
• 5:27
  conflicting information? That is a very common
• 5:30
  issue with older business data. Pricing changes,
• 5:34
  but the old marketing PDF still exists. You give
• 5:37
  Gemini explicit working instructions for this
• 5:39
  exact scenario. It must point out outdated or
• 5:42
  unclear information immediately. It flags it.
• 5:46
  It has to highlight the contradiction for human
• 5:48
  review. It must do this before suggesting any
• 5:51
  strategic action. It flags the contradiction
• 5:53
  first before making any recommendation. Precisely.
• 5:55
  It stops the assembly line before a broken product
• 5:58
  gets shipped. So the AI has the context and the
• 6:02
  clean data prepared. But we need a dedicated
• 6:05
  permanent workspace for all this information.
• 6:08
  Yeah, you do. We definitely don't want to upload
• 6:10
  this stuff every single time. No, that'd be terrible.
• 6:13
  That sounds completely exhausting and totally
• 6:15
  inefficient. Yeah. That logical problem brings
• 6:18
  us directly to step three. Forging the gem. Forging
• 6:21
  the gem. Creating a Gemini gem permanently changes
• 6:25
  your entire daily workflow. Let's define that
• 6:28
  term quickly so nobody gets lost. What exactly
• 6:31
  is a Gemini gem? It's a custom version of the
• 6:33
  AI for specific tasks. Got it. So we create this
• 6:37
  dedicated, specialized workspace. Right. But
• 6:41
  the guide says something that initially feels
• 6:43
  a bit counterintuitive. Oh, about the uploads.
• 6:46
  Yeah. It says we must re -upload all our foundational
• 6:49
  documents. Why do we have to upload the handover
• 6:52
  document again? Because gems purposefully don't
• 6:55
  automatically pull from your past chat history.
• 6:57
  Oh, I see. They're designed with isolated, completely
• 6:59
  independent context windows. Like a walled garden.
• 7:02
  Exactly. They start as a blank slate, requiring
• 7:05
  fresh, specific context. This actually prevents
• 7:08
  your past casual chats from contaminating the
• 7:10
  workspace. That makes a lot of sense. You upload
• 7:12
  the handover document and the knowledge base
• 7:14
  files. Then you give the gem a very clear functional
• 7:17
  name. Like what? Something like marketing assistant
• 7:19
  or customer research analyst? Naming conventions
• 7:22
  really help keep things organized mentally. They
• 7:25
  do, yeah. You treat them like specific departments
• 7:28
  within your actual company. Exactly. Then you
• 7:30
  give the gem its core working instructions. The
• 7:33
  guide highlights one instruction that seems incredibly
• 7:36
  important for business owners. The facts versus
• 7:39
  suggestions instruction. Yes. You demand the
• 7:42
  AI clearly separate confirmed information from
• 7:45
  suggestions. Why is separating facts from strategic
• 7:49
  suggestions so vital? Because as a business owner,
• 7:52
  you need to know what is real. Right. A suggestion
• 7:54
  is just a creative guess from the AI model. A
• 7:58
  fact is grounded in your actual verified, uploaded
• 8:01
  documents. Blurring those two things together
• 8:04
  can cause massive business liability problems.
• 8:07
  Imagine confusing a hypothetical pricing strategy
• 8:09
  with your actual current pricing. It would cause
• 8:12
  absolute chaos for your sales team. Right. It
• 8:15
  keeps us from confusing a creative guess with
• 8:17
  hard data. Yeah. It draws a thick, bright line
• 8:20
  between memory and imagination. Too sex silence.
• 8:23
  I love that phrasing. We are going to take a
• 8:25
  very quick break here. When we come back, we
• 8:27
  will get into teaching the AI how to actually
• 8:30
  sound like you. Mid -roll sponsor break. OK,
• 8:34
  we are back. So the AI is loaded with our verified
• 8:37
  facts. It lives happily in a safe, dedicated
• 8:40
  workspace. Right. But if I send a generated email
• 8:43
  to a client right now, they will know. They will
• 8:46
  instantly know a sterile robot actually wrote
• 8:48
  it. Yeah, it sounds completely unnatural. It
• 8:51
  needs to sound human. Specifically, it needs
• 8:53
  to sound like your kind of human. This is where
• 8:55
  you establish the AI's brand voice. This is step
• 8:57
  four of the framework. Step four, okay. You feed
• 9:00
  Gemini your absolute... best past written content.
• 9:03
  Like what kind of content? We are talking successful
• 9:06
  newsletters, popular blog posts, and converting
• 9:08
  email campaigns. The greatest hits. Exactly.
• 9:11
  You ask the AI to act as a brand style analyst.
• 9:14
  It meticulously reverse engineers your unique
• 9:17
  style from those writing samples. So it looks
• 9:19
  at the mechanics of how you write. It extracts
• 9:22
  your specific sentence structure and natural
• 9:24
  language patterns. It notes how often you use
• 9:28
  humor. or ask rhetorical questions. It catches
• 9:30
  all that. It also identifies the writing styles
• 9:33
  you should actively avoid. What's fascinating
• 9:35
  here is how incredibly objective the AI is. What
• 9:39
  do you mean? It analyzes your subjective voice
• 9:41
  much better than you ever could. Oh, interesting.
• 9:44
  We are blind to our own communication habits.
• 9:47
  The AI notices subtle patterns and rhythms you
• 9:51
  use completely unconsciously. Right. It spots
• 9:54
  your exact transition phrases and your typical
• 9:56
  paragraph lengths. That makes total sense. We
• 9:58
  don't hear our own accents. No, we don't. But
• 10:00
  what if a new business doesn't have a large content
• 10:03
  archive? That happens a lot. Should we just let
• 10:06
  the AI guess our tone without samples? No, you
• 10:09
  should absolutely never let it guess. Never guess.
• 10:11
  If it guesses You just get terrible generic corporate
• 10:14
  speak. Synergy and paradigm shifts. Exactly.
• 10:18
  LLMs are trained on the entire vast internet
• 10:21
  if you just ask for a professional email. It
• 10:23
  averages every professional email ever written
• 10:26
  in human history. Oh, wow. The result is pure,
• 10:31
  unreadable, distilled mediocrity. That sounds
• 10:34
  awful. It is awful. You must provide actual examples
• 10:38
  to pull it away from that center. Never let it
• 10:40
  guess. Only use patterns from actual provided
• 10:42
  examples. Exactly. Even if you just write three
• 10:45
  emails from scratch today. Okay. Use those as
• 10:47
  the baseline samples. Okay, so it sounds exactly
• 10:50
  like you now. Yet. And it has all of your private
• 10:53
  verified business data. But before we let this
• 10:56
  thing run free... We need safety. Yes, safety
• 10:59
  is huge. We need strong safety nets to prevent
• 11:01
  an absolute disaster. Step five is putting up
• 11:04
  strict operational guardrails. Guardrails. Guardrails
• 11:07
  are absolutely essential for any business AI
• 11:10
  assistant. You have to set very strict limitations
• 11:13
  in the core instructions. Like, what kind of
• 11:15
  limitations? You tell it no inventing pricing
• 11:17
  under any circumstances whatsoever. Right. No
• 11:20
  making up business metrics or projecting false
• 11:22
  financial results. No promising revenue or growth
• 11:25
  to any potential customer. Because it wants to
• 11:28
  please the user, so it might just say yes to
• 11:30
  everything. Exactly. You have to block that urge.
• 11:32
  You also force the AI to ask clarifying questions
• 11:35
  constantly. Yeah. If any context is missing,
• 11:38
  it must stop and ask you. It has to hit pause.
• 11:41
  It has to highlight potential risks before suggesting
• 11:44
  major structural actions. To sex silence. Whoa.
• 11:50
  Imagine scaling this out across a huge company.
• 11:53
  It is mind blowing. Having an assistant that
• 11:56
  never goes off script or makes a false promise,
• 11:58
  no matter how much you throw at it. Yeah. It
• 12:01
  is like having an infinitely patient compliance
• 12:04
  officer. That is a great way to put it. Well,
• 12:06
  wait, I have to challenge you there on the mechanics.
• 12:08
  OK, let's hear it. If I tell the AI never to
• 12:11
  invent pricing, if I say it can never promise
• 12:14
  growth, aren't I just putting it in a creative
• 12:16
  straight jacket? I see where you're going. At
• 12:19
  that point, aren't I just building an overpriced
• 12:21
  FAQ bot? Do these strict guardrails stifle the
• 12:25
  AI's natural creativity? That is a very common
• 12:28
  fear, but it isn't actually true. Okay, explain
• 12:30
  that. You have to understand how constrained
• 12:33
  creativity works. Guardrails only restrict the
• 12:36
  invention of fake factual data. Right. You are
• 12:39
  constraining the facts, not constraining the
• 12:41
  underlying logic. Ah, I see. The AI can still
• 12:44
  brainstorm incredibly brilliant marketing strategies
• 12:46
  freely. It can still write incredibly persuasive,
• 12:50
  emotionally resonant copy. It just uses real
• 12:53
  ingredients. Yes. It just bases those creative
• 12:56
  strategies on real, verified facts. Ah, guardrails
• 12:59
  protect the facts, but still allow creative strategy.
• 13:02
  Exactly. Constraints actually breed better creativity.
• 13:05
  They do, yeah. It forces the AI to be clever
• 13:08
  with the real data. To sex silence. We have finally
• 13:11
  reached step six of the process. The final step.
• 13:14
  Automating your daily recurring business workflows.
• 13:17
  This is a big one. The ultimate goal here is
• 13:19
  moving past a conversational chatbot interface.
• 13:22
  We want a task executing machine that actually
• 13:24
  drives the business forward. This is where the
• 13:26
  magic really happens. You start using Gemini
• 13:29
  as a dedicated workflow architect. A workflow
• 13:31
  architect. Yeah, you ask it to analyze your current
• 13:34
  daily business operations deeply. Okay. then
• 13:37
  it designs highly repeatable blueprints for your
• 13:40
  mundane tasks. Like what kind of blueprints?
• 13:42
  Well, these blueprints have very specific elements
• 13:45
  that ensure consistent high quality. Every blueprint
• 13:48
  starts with a crystal clear business objective.
• 13:51
  It lists all the required inputs and specific
• 13:53
  output requirements. It defines the final format
• 13:57
  and the exact level of necessary detail. It also
• 14:01
  establishes very clear decision rules for the
• 14:03
  AI. Decision rules are so important. For example,
• 14:06
  if a customer email is angry, use an empathetic
• 14:09
  tone. If the email is a simple question, be concise
• 14:13
  and direct. And there is one absolute necessity
• 14:16
  in these workflow blueprints. What is that? You
• 14:19
  must establish mandatory human review points
• 14:22
  in the process. Right. You cannot just let the
• 14:25
  AI run entirely unsupervised in the background.
• 14:28
  So what does this all mean for the daily workflow?
• 14:30
  Why define human review points instead of just
• 14:33
  checking final outputs? Because automated workflows
• 14:35
  can become incredibly complex very quickly. They
• 14:38
  really can. They often involve multiple chain
• 14:41
  steps of AI reasoning. If the AI hallucinates
• 14:45
  early in that chain, the final output is completely
• 14:48
  ruined. Oh, it cascades. Exactly. A small error
• 14:51
  in step one becomes a massive error in step four.
• 14:54
  Viewpoints let you catch minor errors before
• 14:56
  they completely cascade. It keeps the human in
• 14:59
  the loop before the AI derails. Right. You become
• 15:02
  an editor instead of a ... Blank page writer.
• 15:04
  Two secs silence. We covered an immense amount
• 15:06
  of ground today. You really did. This shifts
• 15:08
  the entire paradigm of how we interact with technology.
• 15:12
  Can you synthesize this entire six step journey
• 15:14
  for us? We started with a generic amnesiac chatbot
• 15:18
  interface. Blank slate. It didn't know you and
• 15:21
  it didn't care about your business. We systematically
• 15:24
  moved to a highly specialized, safe, automated
• 15:26
  business partner. You front -loaded the extremely
• 15:29
  hard work of establishing business context. You
• 15:31
  built the handover document and the verified
• 15:33
  knowledge base. The Lego blocks. Exactly. You
• 15:36
  created a secure gem and defined your exact brand
• 15:39
  style. You put up guardrails and mapped out repeatable
• 15:42
  workflows. By doing all this, you stopped treating
• 15:45
  AI like a magic eight ball. You finally start
• 15:48
  treating it like a customized software engine.
• 15:51
  That shift in perspective changes absolutely
• 15:54
  everything about working with AI. It does. It
• 15:56
  demands a serious upfront investment of your
• 15:59
  time and focus. You cannot just click a button
• 16:02
  and expect instant miracles. No, you have to
• 16:05
  put in the work. But the operational payoff on
• 16:07
  the back end is truly massive. I want to leave
• 16:10
  you with one final thought today. OK. If you
• 16:13
  take the time to build this properly. Yeah. If
• 16:16
  you build a Gemini Gem that perfectly understands
• 16:19
  your complex business model. If it perfectly
• 16:22
  mimics your unique brand voice and executes workflows
• 16:25
  flawlessly, what happens to your actual role
• 16:28
  as the business owner? That is the big question.
• 16:30
  When your AI knows your business this intimately,
• 16:33
  everything shifts. It frees you up to start just
• 16:35
  running the daily operations. You get your time
• 16:37
  back. You can finally step back and start actually
• 16:39
  growing the business. Thank you all so much for
• 16:42
  taking this deep dive with us today. Out You
• 16:43
  Row Music.