Latest / AI Fire Daily / #542 Neil: Claude AI Skills That Turn Social Data Into Better Content
Transcript
- 0:00It is two in the morning. You are staring blankly
- 0:03at your analytics dashboard. Yeah, we've all
- 0:05been there. The video you just spent two solid
- 0:08weeks writing, shooting, and editing is sitting
- 0:11right there. And it is completely dead. Right.
- 0:15And it always seems to happen when you're incredibly
- 0:17confident about it. Yeah, exactly. So what do
- 0:19you do? You panic. You change the title. You
- 0:22swap the thumbnail to something much louder.
- 0:25You are entirely relying on your gut feeling.
- 0:28Beat. But what if I told you the exact mathematical
- 0:31reason it failed is already sitting inside your
- 0:34data, and you just aren't looking at it? We live
- 0:37in an era where that information is entirely
- 0:40visible. We just refuse to do the math. Right.
- 0:43I mean, the math simply takes too much time for
- 0:44a human to do. Welcome to the deep dive. Today,
- 0:48we are on a very specific mission for you. We
- 0:50are unpacking an incredibly detailed guide on
- 0:53five distinct Claude AI skills. And just to clarify,
- 0:57these aren't general writing tricks. No, not
- 0:59at all. These are built specifically for high
- 1:01level social media analysis. Let's frame the
- 1:04value here immediately. This isn't about using
- 1:07AI to mindlessly generate generic, you know,
- 1:12soulless posts. It is about using AI to do the
- 1:17heavy, tedious analytical lifting. It frees the
- 1:19human to actually be creative. Right. So here
- 1:23is our roadmap for today. We will start by analyzing
- 1:27your own channel's hidden patterns. Which is
- 1:29crucial. Then, we look at safely expanding your
- 1:32audience without alienating your core fans. After
- 1:35that, we scout your competitors. Oh, that's my
- 1:37favorite part. We track their permanent viral
- 1:39breakouts. Finally, we reverse engineer the perfect
- 1:42hook. It is a complete strategic ecosystem. Okay,
- 1:44let's unpack this. If you want to grow, your
- 1:47instinct is usually to look out into the wider
- 1:49internet. But before we do that, we have to look
- 1:51inward. Right. We have to analyze what is genuinely
- 1:53working on your own channel. Because a strong
- 1:55video leaves an incredible trail of clues behind.
- 1:59But finding those clues across 50 or 100 videos,
- 2:02that is just beyond human working memory. The
- 2:05guide suggests taking your raw channel data.
- 2:08You take your titles, view counts, hooks, and
- 2:11full transcripts, and you upload all of it straight
- 2:13into Claude. Yeah, and the guide specifically
- 2:15recommends using the Claude Sonnet 5 model for
- 2:18this. They strongly advise against using Opus
- 2:204 .8. Why that specific model? Well, Sonnet 5
- 2:23is phenomenally good at handling massive files.
- 2:27It processes large context windows efficiently.
- 2:29Plus, the API cost is significantly lower. Let's
- 2:33pause and define that quickly for everyone listening.
- 2:35Sure. The tiny fee paid every time software talks
- 2:38to the AI. Perfect. So we dump all this data
- 2:41into Sonnet 5. We give it a prompt to act as
- 2:43an elite content analyst. It starts ranking your
- 2:46videos. It specifically compares the top 20 %
- 2:49against the bottom 20%. And that comparison is
- 2:51the actual magic trick. It isolates the variables.
- 2:54Right. Let's say you run a tech review channel.
- 2:57Claude might notice that in your bottom 20%,
- 3:00you always start by listing technical specs.
- 3:02But in your top 20%, you always open with a relatable
- 3:05problem. It's like stacking Lego blocks of data
- 3:08to build a perfectly clear picture of your audience's
- 3:11brain. I love that. And what is fascinating here
- 3:14is the strictness of the prompt. It enforces
- 3:17a hard rule. What kind of rule? It must find
- 3:20these patterns in at least two top -performing
- 3:22videos. Ah, because one hit is just luck, like
- 3:25an algorithm glitch. Exactly. One viral video
- 3:28is an anomaly. Two starts to form a reliable
- 3:31pattern. Once it establishes those multi -video
- 3:34patterns, it hands you data -backed content gaps.
- 3:37Then it suggests five brand new video ideas based
- 3:40on real data. But I have a very practical question
- 3:43here. My analytics dashboard is often a disaster.
- 3:46Oh, sure. Same. What happens if the data we upload
- 3:48is just too messy? Or what if we are completely
- 3:51missing key metrics, like average watch time?
- 3:54Will the AI just guess? The prompt is explicitly
- 3:57built to prevent that. It commands Claude to
- 3:59flag missing data immediately. It completely
- 4:02refuses to hallucinate fake numbers just to please
- 4:05you. Right, so it refuses to guess if the evidence
- 4:07isn't there. Exactly. It forces you to operate
- 4:09in reality. So Claude has perfectly optimized
- 4:12your content for your existing fans. Two -Sex
- 4:15Silence. But there's an obvious ceiling to that.
- 4:18Yeah, you hit a wall eventually. If you only
- 4:21talk to people who already agree with you, you
- 4:23stop growing. You have to step outside your bubble.
- 4:27But how do you do that without accidentally pivoting
- 4:29your entire channel? You don't want to alienate
- 4:32the people who actually buy your product. That
- 4:34is the eternal struggle of the creator. You want
- 4:36explosive growth. But you really don't want empty
- 4:40views from people who don't care about your actual
- 4:42mission. The guide introduces a framework for
- 4:45this exact problem. You upload your ideal customer
- 4:48profile and your core product information. Right.
- 4:50Then. Claude builds what they call an audience
- 4:53bullseye. I really love this concept because
- 4:56it creates boundaries. It breaks your potential
- 4:58audience into five distinct rings. Ring one is
- 5:02your core buyers, the true fans. Yeah. Ring two
- 5:05is the problem -aware audience, people who need
- 5:08you but don't know you. Ring three is the adjacent
- 5:10audience. Right. And ring four is the broader
- 5:13relevant audience. Finally, ring five is broad
- 5:16interest. Think of ring five as the general public.
- 5:19Let's make this concrete. If I run a channel
- 5:22about specialized marathon training, my ring
- 5:25one is competitive marathon runners. Makes sense.
- 5:28What is my ring four or five? Well, ring four
- 5:31might be people who just want to lose weight
- 5:33by jogging. They care about fitness, but they
- 5:35don't care about marathon pacing. Ring five would
- 5:38be anyone who owns sneakers. broad generic interest.
- 5:42The prompt then enforces a strict constraint
- 5:44on Claude. It is called the 3 -2 -2 content mix.
- 5:49This is brilliant pacing for a content calendar.
- 5:52Yeah. It dictates three posts strictly for ring
- 5:55one, two posts for rings two and three, two posts
- 5:57for ring four. And ring five is completely excluded.
- 6:00You don't make content for the general public
- 6:02unless there is a highly specific strategic reason.
- 6:05Absolutely. Wait, isn't the whole point of social
- 6:08media to go as broad as possible to get maximum
- 6:11views? You'd think so, but if we connect this
- 6:14to the bigger picture, raw views don't pay the
- 6:17bills. If you make a viral video for Ring 5 about
- 6:20tying your shoes, you get a million views. But
- 6:23those people will never buy your advanced marathon
- 6:25guide. Because they don't hear about marathons.
- 6:27Exactly. You just diluted your core message.
- 6:29You ruined your algorithm targeting. OK, but
- 6:32if I'm asking Claude... for ideas for ring four.
- 6:36How do we stop the AI from just suggesting generic
- 6:40viral trends? I don't want it telling me to do
- 6:43a running dance challenge. Because the prompt
- 6:45explicitly forbids broad labels. It bans generic
- 6:49words like marketers or fitness fans. It demands
- 6:52highly specific painful problems for every single
- 6:55audience group. It acts like a bouncer. keeping
- 6:58your relevant views out of your strategy. Exactly.
- 7:00You expand your reach, but you expand it safely.
- 7:03Okay, you have your audience rings clearly defined
- 7:05now. You know exactly who you are talking to.
- 7:08The next logical step is looking outward. You
- 7:10want to see what is working for your competitors
- 7:12in those exact same rings. But you have to be
- 7:14extremely careful here because raw views are
- 7:17a complete lie. That is a bold statement. Walk
- 7:21us through why raw views are a lie. Look at it
- 7:23proportionally. A 50 ,000 view video on a massive
- 7:27channel with two million subscribers is actually
- 7:30a miserable flop. Oh, wow. Yeah. But on a tiny
- 7:33channel with a thousand subscribers, that exact
- 7:36same view count is an absolute goldmine. It signals
- 7:39something highly resonant. The scale dictates
- 7:42the success, not the raw number. Precisely. That
- 7:45is why the prompt ignores raw views. It calculates
- 7:48an outlier score instead. Let's define that term
- 7:51for the listener. A video's views divided by
- 7:53that channel's normal average views. So it levels
- 7:56the playing field completely. Small channels
- 7:59suddenly get a fair shot at being analyzed. Yes.
- 8:01The prompt takes those scores and ranks them
- 8:04into three tiers. Okay. Extreme outliers are
- 8:06three times the channel's normal average. Strong
- 8:09outliers are two times the normal. Promising
- 8:11outliers are 1 .5 times normal. It also excludes
- 8:14any channel with under 10 valid videos, right?
- 8:17That ensures the mathematical average actually
- 8:19means something. That's right. You need a baseline.
- 8:21Whoa. Beat. Imagine instantly spotting hidden
- 8:25trends across 50 different channels before they
- 8:27go mainstream. It is a massive tactical advantage.
- 8:30You aren't just looking at what is generally
- 8:32popular. You are looking for what the guy calls
- 8:35emerging signals. Which are what? Exactly. These
- 8:38are hidden patterns that are only just starting
- 8:40to appear in recent outliers. OK, wait. If I'm
- 8:43feeding an AI, the exact video that caused a
- 8:48competitor's channel to explode. Yeah. My instinct
- 8:51is to just say, hey Claude, write me a script
- 8:53exactly like this. But if everyone does that,
- 8:57social media just becomes an echo chamber of
- 8:58clones. Completely. What's to stop us from just
- 9:01taking these outlier ideas and completely ripping
- 9:03them off? The prompt is way ahead of you on that.
- 9:05It explicitly bans copying hooks or structural
- 9:08layouts. Really? Instead, it analyzes why the
- 9:11outlier worked. Maybe it was a unique framing
- 9:13of a common problem. It forces the AI to generate
- 9:16completely unique adaptations tailored to your
- 9:19specific niche. It extracts the underlying recipe.
- 9:22but forces you to cook your own meal. Beautifully
- 9:25said. You leverage the proven mechanics without
- 9:27ever sacrificing your originality. Spotting a
- 9:30single competitor outlier is highly valuable.
- 9:33That gives you a great video idea. Definitely.
- 9:36But spotting the exact moment a competitor's
- 9:39entire channel permanently leveled up, that is
- 9:42the holy grail of strategy. It really is. We
- 9:45are moving from finding a good video to detecting
- 9:48a creator's fundamental breakout moment. The
- 9:51guide? has a specific skill designed just for
- 9:54this. You take a creator's entire video history,
- 9:57you put it in strict chronological order, and
- 10:00then you upload it into Claude. First, Claude
- 10:03looks at their last 10 videos. This establishes
- 10:06their normal baseline. Once it has that average,
- 10:08it starts scanning forward. It hunts for a massive,
- 10:12uncharacteristic jump. Specifically, it looks
- 10:15for a video pulling in at least double their
- 10:17normal viewership. It's like looking at an X
- 10:19-ray that reveals the exact moment A creator's
- 10:22DNA mutated for the better. Yes, but the prompt
- 10:25does something absolutely crucial next. Finding
- 10:28a sudden spike is easy, but Claude is instructed
- 10:31to check the next five videos immediately following
- 10:33that spike. To see if the new baseline holds.
- 10:35Exactly. Because if the views drop right back
- 10:38down to the old average, it was just a one -time
- 10:40viral fluke. And a fluke doesn't teach you anything
- 10:43useful. Not at all. But if the new baseline holds
- 10:46steady, you are looking at a sustained breakout.
- 10:49Once Claude confirms it, it generates a detailed
- 10:52before versus after table. And this is where
- 10:54you analyze what actually changed. You aren't
- 10:57just looking at thumbnail colors. No, it goes
- 10:59way deeper. You look at their storytelling structure.
- 11:01You check their pacing. You examine how they
- 11:04changed their calls to action. Let's say a creator
- 11:06went from getting 5 ,000 views to 50 ,000 views
- 11:10consistently. OK. The AI might reveal that in
- 11:14the before phase, they always introduced themselves
- 11:16for 30 seconds. In the after phase, they dropped
- 11:20the intro entirely and started mid -action. But
- 11:23let me challenge that. We know platforms change
- 11:25their recommendation engines all the time. Sure.
- 11:27Could an algorithm glitch trick the AI into thinking
- 11:31a breakout happened when it was just a temporary
- 11:33platform push? That's exactly why it analyzes
- 11:36the next five videos. It looks for structural
- 11:39narrative changes in the transcripts. It isn't
- 11:41just looking at the view count. It demands proof
- 11:44of a lasting qualitative change in the actual
- 11:47content. It requires sustained evidence, filtering
- 11:50out random spikes of algorithmic luck. The AI
- 11:53is ruthless about demanding evidence. Mid -roll
- 11:55sponsor read. Okay, we're back. Think about what
- 11:57we've built today. We have a strategy based on
- 12:00internal channel data. We mapped our audience
- 12:02rings. We found competitor outliers. We tracked
- 12:05their permanent breakouts. It is a phenomenal
- 12:07foundation. Two sec silence. But absolutely none
- 12:11of it matters if a viewer scrolls past your video
- 12:13in the first three seconds. The hook is the ultimate
- 12:16gatekeeper. If the hook fails, the entire ecosystem
- 12:20we just built fails right along with it. This
- 12:22brings us to the fifth and final skill, building
- 12:25a hook machine. First, you upload your own pass
- 12:28-winning videos specifically to extract the opening
- 12:31hooks. Claude takes those specific hooks and
- 12:33groups them into repeated structural categories.
- 12:36Like what? It finds your underlying frameworks.
- 12:39Things like a personal experiment framework,
- 12:41an unexpected result setup, or a specific problem
- 12:44framing. Then, it builds a 100 -point scoring
- 12:48rubric to judge your future hooks. And this rubric
- 12:51is based on four critical pillars, curiosity,
- 12:55tension, specificity, and payoff. Those four
- 12:58elements are the psychological levers that dictate
- 13:01human attention. Let's break that down, because
- 13:03it sounds a bit abstract. Let's look at tension.
- 13:05A bad hook says, I bought a new camera. Right.
- 13:08There's zero tension there. Exactly. But if Claude
- 13:12applies tension, it might rewrite it as, I spent
- 13:14$5 ,000 on a camera, and it ruined my most important
- 13:18shoot. Right. Claude recognizes that contrasting
- 13:21a high price with a disastrous outcome creates
- 13:24instant psychological tension in the viewer's
- 13:26mind. And specificity works the same way. I lost
- 13:29weight was boring, but I lost 14 pounds eating
- 13:32only potatoes grabs you immediately. Specific
- 13:35details prove that you actually have a real story
- 13:37to tell. I'll be honest, I still wrestle with
- 13:39judging my own hooks by instinct alone. It is
- 13:42painfully subjective. We all do. You spend 50
- 13:44hours on a script. You are entirely too close
- 13:47to the material to know if the first 10 words
- 13:49are objectively interesting. That is why this
- 13:52rubric is an absolute lifesaver. You paste a
- 13:54completely new hook into the chat. Claude scores
- 13:57it out of 100 based on those four pillars. It
- 14:01identifies the absolute weakest link in your
- 14:04text. Then it writes five improved variations
- 14:07for you. But here's the critical part. It bases
- 14:10those variations only on your proven past structures
- 14:13that it identified earlier. new hooks for me,
- 14:18doesn't run the risk of sounding like a robotic
- 14:20clickbait generator, like you won't believe what
- 14:23happens next. No, because the prompt acts as
- 14:25a strict guardrail. It forces the AI to only
- 14:29use the distinct, authentic structures that your
- 14:31actual human audience previously loved and validated.
- 14:34It strictly refuses to use generic clickbait
- 14:37tropes. Your past successes literally become
- 14:40the mentors for your future ideas. That is exactly
- 14:43how the system works. It removes the guesswork
- 14:45while preserving your unique voice. Let's synthesize
- 14:47all of this. The overarching philosophy of this
- 14:50deep dive is incredibly clear. The era of staring
- 14:53at a blinking cursor hoping inspiration strikes
- 14:56is over. And the era of staring at raw view counts
- 15:01and just guessing why something worked is entirely
- 15:03done. These five skills do not replace the creator's
- 15:07unique voice. They replace the tedious manual
- 15:10guesswork of content strategy. It's about leveraging
- 15:13AI as your personal, elite data analyst. It crunches
- 15:18the massive data set so you can spend your valuable
- 15:20time actually creating. Here is where it gets
- 15:23really interesting for you, the listener. We
- 15:25talked a lot today about finding breakout moments
- 15:28in social media data. We did, spotting that permanent
- 15:31structural shift from the baseline. But what
- 15:33if you applied that same before -and -after analysis
- 15:35to your own career trajectory? Oh, wow. Two -second
- 15:38silence. What invisible baseline are you trying
- 15:41to break out of right now? That is a phenomenal
- 15:43question to ponder. Thank you for exploring these
- 15:45concepts with us today. Thanks for joining this
- 15:47deep dive. Stay curious and keep analyzing your
- 15:50world.