YT YouTube2Text

YouTube2Text → Transcripts → Full Claude Tutorial: Beginner to Advanced in 19 Minutes

Full Claude Tutorial: Beginner to Advanced in 19 Minutes — Transcript

by Futurepedia · 3,921 words · 601 segments · language en · Watch on YouTube

Open in transcriber Summarize with AI Download SRT/VTT

Full transcript

0:00Claude has been my most used tool for
0:02the past year. Now, everyone else is
0:04getting on board. It has projects that
0:06remember everything about your work,
0:07skills that execute your repeatable
0:09processes automatically, connectors that
0:11pull in your actual tools and data, and
0:14much more. I'll cover every feature, and
0:16more importantly, how to connect it all
0:18into workflows that unlock its full
0:20power. This is the pricing. They do have
0:22one of the strongest free plans of all
0:24the LLMs. You can get a lot from it.
0:26But, if you use AI regularly, I
0:28recommend investing in the paid plan of
0:30whichever LLM you prefer most. It makes
0:32a big difference. For me, that's Claude,
0:34although I of course have a paid plan
0:36for all of them. For most people, this
0:37$20 per month plan will be plenty. In
0:39general, you'll only need to move up
0:41when you start using Claude Co-worker
0:42extensively or Claude Code. And those
0:44are separate tools beyond what I'm
0:46covering in this video. Okay, let's get
0:48into the actual product. Once you've set
0:50up an account, you'll be here. A similar
0:52chat interface to all the AI models. We
0:54have our prompt box in the middle, the
0:56model selector on the right, file
0:59uploads on the left. Then, the left
1:01panel opens up and has all your past
1:03conversations, plus some amazing
1:04features which I'll cover later, like
1:06skills, projects, and artifacts. One
1:09other thing I want to mention is under
1:10settings. There's some very common
1:12advice people give that actually makes
1:14your responses worse. So, there are
1:16these fields where you can tell Claude
1:17about yourself, select your work
1:19function, and personal preferences
1:21Claude should consider in responses. You
1:23tell Claude about yourself and give it a
1:24list of how you'd like it to respond.
1:26That sounds great, which is why everyone
1:28recommends it. But, if you actually use
1:30Claude across different topics, it's bad
1:32advice. I use Claude for a wide range of
1:34things, and no custom instruction is
1:36going to apply to everything. It can
1:38lead to some weird responses. I leave
1:40these completely blank and set them up
1:41in projects instead. These are folders
1:44you can give their own instructions and
1:46custom knowledge bases. They're scoped
1:48to each context, so the right
1:49instructions apply at the right time.
1:51It's a much better system, and I'll
1:53cover all the details later. But, since
1:54this advice is so common, I wanted to
1:56cover it right off the bat. Now, back to
1:58the prompt box. You can use this for all
2:00the things we're used to with AI by now.
2:02Research, coding, problem-solving,
2:04learning new things, or writing, which
2:06is what I first started using Claude
2:08for. It has the most natural-sounding
2:10writing by far, and ability to match
2:12your style. It's been the leader there
2:14for a long time, same with coding. But,
2:16over the past few months, it's caught up
2:18and surpassed the other models in just
2:20about everything. Most people expect to
2:22unlock all of that by just typing a
2:24question into this box. With a basic
2:25prompt, you'll get generic results. And
2:27people do that, and then end up thinking
2:28that AI just gives that type of answer.
2:30But, it's all in how you use it. So, I
2:32have a three-part formula that changes
2:34the quality of every output. But, after
2:36that foundation, we can get into the
2:37really powerful stuff.
2:40The three most important parts of a
2:41prompt are what I call ICC:
2:43instructions, context, and constraints.
2:46Instructions define the task and action
2:48you want Claude to take. Context is
2:50setting the stage with your role,
2:52objectives, backgrounds, anything Claude
2:54should know that's relevant to the task.
2:56Err on the side of more context. A full
2:58context dump is fine. Then, constraints.
3:01Specify any rules, style, tone, length,
3:04output format. And what can be
3:06especially helpful, when it applies, is
3:08including an example of the output
3:10you're looking for. Now, the order of
3:11these three isn't as important as long
3:13as you include everything. So, here's a
3:15bad prompt. Recommend five ways to
3:16implement AI in my marketing agency.
3:19These answers aren't bad, but they're
3:21very generic. To make this better, I'll
3:23use ICC. Same instruction, recommend
3:25five ways to implement AI in my
3:26business. And for context, I give some
3:28more information about the business, and
3:30then also added some constraints. Make
3:32sure they're low cost, easy to implement
3:34without technical expertise, things like
3:35that. And also format it as a numbered
3:37list with a one-sentence explanation for
3:39each.
3:40Now, this is much more specific to the
3:41business I listed. There's actually a
3:43couple of these that I'm going to cover
3:44ways you could do right in Claude. ICC
3:46will give dramatically better results,
3:48but even still, don't expect the output
3:50to be perfect on the first try. Iterate.
3:52Go back and forth until it's perfect.
3:54The best results come from treating
3:55Claude like a collaborator. In this
3:57case, I asked for just one sentence
3:59each, so I could skim through it and
4:00then ask to elaborate on the ones that
4:02sound the most promising, rather than
4:04reading through a full paragraph on
4:05every single tip. And that actually
4:07leads me into a pro tip on prompting.
4:09The reason the list still isn't perfect
4:11is, although I added some context, it
4:13can be hard to think of all the context
4:15you need to include. You'll often end up
4:17leaving things out that would have
4:18helped. To solve that, at the end of
4:20your prompt, ask Claude to ask you for
4:22any additional context it needs to best
4:24achieve the task. I call this a context
4:27interview. It will know what it needs
4:28better than you do. Then, it will ask a
4:30series of questions to gather the rest
4:32of that information. After it has
4:34everything, it will give you an answer
4:35that's much more helpful because it's
4:37tailored to your specific situation.
4:39Now, these answers are more customized
4:40around client communication and
4:42reporting because I answered that that
4:43was my biggest time drain. It also asked
4:45what tools I'm already using, so I could
4:47tailor it more to my setup. Once you
4:49combine the three-part formula with that
4:51context interview, your first output is
4:53almost always usable. If you want to go
4:55deeper into using Claude for work, I
4:56have a free resource in the description
4:58called The Complete Guide to Claude AI:
5:01Revolutionize the Way You Work. It has a
5:03lot of what I'm covering, but in written
5:04form, and also a lot more example use
5:07cases. Plus, each section has a prompt
5:09you can just copy-paste to get started
5:11right away. Like in artifacts, you can
5:13create a full quarterly marketing
5:14campaign analysis artifact and a
5:16marketing performance interactive
5:17dashboard. Then, there's a full section
5:19on setting up your AI executive
5:21assistant in Claude. That goes into way
5:22more specifics than I can cover in a
5:24video like this. So, make sure to grab
5:26that. It's a free resource down in the
5:27description.
5:29I mentioned web search, which you can
5:31find under the plus button. It's usually
5:33on by default, and I recommend leaving
5:35it enabled most of the time. But, it
5:36doesn't always use it by default. It
5:38will anytime it's obvious, but sometimes
5:40it isn't, and you'll need to ask it to
5:42search the web in your prompt. One
5:43example is when I'm having it help
5:45critique a script. Sometimes I mention
5:46Nano Banana for generating images, and
5:48I've had it say, "I see you have a
5:50placeholder here," or even just tell me
5:52I made up the name. Since critiquing a
5:53script isn't something it intuitively
5:56feels the need to search the web for, I
5:58have to ask it to search. But, if I just
5:59asked it, "What's the best text-to-image
6:01model right now?" In that case, it would
6:03know. But, this gives up-to-date data
6:05and can also help reduce hallucinations.
6:08Something I do a lot is separate the
6:10task of searching from the task of
6:12answering a prompt. Ground first, ask
6:14second. Like, if I need help with
6:15something in an A 10, I'll first run a
6:17prompt that says, "Research A 10 and
6:19make sure you're up to date on all its
6:20capabilities." Then, I can ask it to
6:22help with my issue. Grounding it in that
6:24context first changes the quality of
6:26every answer that follows. Another
6:28example of ground first, ask second is,
6:31say I was launching a product. I'd ask
6:33it to research Alex Hormozi strategies
6:35and extract all the most important
6:36information and tactics.
6:38Now, it has his full strategies, value
6:40equation, Grand Slam offer, pricing
6:43philosophy, market selection rule, risk
6:44reversal tactics. Now, when I ask it to
6:46help me with my launch plan, it will use
6:48all of that knowledge to inform its
6:50answers.
6:52Using the plus button or just dragging
6:54them in, you can upload files to work
6:56from or analyze. Drop in PDFs, images,
6:59spreadsheets, a 200-page contract, a
7:02competitor's landing page. Claude reads,
7:04summarizes, extracts, and analyzes. And
7:06one note, although it can understand
7:08images, it can't generate them, which is
7:11kind of crazy when every other model,
7:13like ChatGPT or Gemini, can generate
7:15amazing images, but Claude can't. It's
7:17definitely a downside to Claude if you
7:19want to be able to generate images. But,
7:21for uploading, as an example, I'm going
7:23to come here and take a screenshot of my
7:26channel, be the 12 most recent videos.
7:29Drop that in, and I'll say, "Here's a
7:30screenshot of my YouTube channel.
7:32Analyze the title and thumbnail
7:33combinations across these videos,
7:35identify what's working visually and
7:37messaging-wise, what isn't, and any
7:39patterns you notice. Be direct and
7:40specific. I want honest feedback, not
7:42flattery." And then, looking at that, it
7:44gives me some good advice. The clearest
7:46failure, fact that I often just restate
7:48the title rather than extending it.
7:50That's a good tip. List titles
7:52underperform, they feel like homework. I
7:54should use identity or emotional
7:55framing. Yeah, there are definitely some
7:57good notes in here. Something else I can
7:59do is actually export my analytics as a
8:01CSV from the last 90 days. It can work
8:04with all sorts of formats, so I can just
8:05drop in that CSV. Then, I'll say,
8:07"Here's my YouTube analytics data.
8:09Analyze my performance trends and
8:10visualize the key metrics. What's the
8:12data telling me about what's working?"
8:14And it gave me an answer on the left,
8:16and went above and beyond, and decided
8:17to create a full analytics dashboard on
8:20the right. Claude just comes up with the
8:21best way it thinks it can portray
8:23information. Sometimes, that's a
8:24dashboard, sometimes that's just a
8:26little visualization or a chart. And
8:28coming back to this Alex Hormozi prompt,
8:30when it broke down the whole strategy
8:31playbook, it has all this formatting
8:33here. These are called inline
8:34visualizations, and they are actually a
8:36newer feature. I can actually click
8:38between each of these tabs. This had the
8:40whole Grand Slam offer, there's the lead
8:42gen, sales, content, scaling, and
8:45mindset. This is a super helpful way to
8:47get this information across. Otherwise,
8:49it would have been just this massive
8:50wall of text. And this comes in all
8:51sorts of different formats. You can also
8:53have Claude create actual downloadable
8:55files. Now, here's a chat where I had a
8:57folder on my phone where I was just
8:59saving random cool spots I might want to
9:00go in Utah. I just took screenshots of
9:02those, and then asked it to organize
9:04everything. It put together this awesome
9:06document, grouped by region and
9:08difficulty, and I can download that
9:10right to my computer. Or in a previous
9:11video, I had it create a full
9:12acquisition deck as a PowerPoint
9:14presentation. All sorts of formats that
9:16you can view directly inside Claude or
9:18download to your computer. And with
9:19these, we've moved on from just
9:20chatting, now we're creating. But,
9:22there's an even more powerful form of
9:24creation available called artifacts.
9:26That's what this analytics dashboard
9:28was.
9:30Artifacts are standalone interactive
9:32outputs that Claude creates in a
9:34dedicated window alongside your
9:36conversation. Anytime you ask for it, or
9:38it just thinks producing something
9:39substantial will be beneficial, it
9:41generates it in a dedicated side panel.
9:43And you can iterate on it without losing
9:45it in the scroll.
9:47I made a few more examples really quick,
9:49just to show the range. Here's a
9:50flowchart of a customer onboarding
9:52process. I've got an interactive expense
9:54tracking dashboard, a landing page for
9:57an energy drink with our logo if we ever
9:58want to create future fuel. Another
10:00really cool one is it can use p5.js for
10:03all sorts of interactive graphics and
10:04animations. I literally just one shot at
10:07each of these for this demo. It's
10:08incredibly easy to do. Just prompt
10:10whatever you want into existence and it
10:12can do way more than what I just showed.
10:14Or another common one I use is writing.
10:16You can just ask it to open up in
10:18artifacts and you'll have the document
10:20on the right with the chat on the left
10:21so you can ask for changes while viewing
10:23the whole document. I'll show how I
10:25actually use that in a minute it
10:26incorporates skills which I haven't
10:28covered yet but are amazing. And if
10:30Claude doesn't automatically create an
10:31artifact when you expect one, you can
10:33explicitly ask create this as an
10:35artifact.
10:37One more I haven't covered yet in this
10:39realm is research. This completely
10:41changes how Claude finds and analyzes
10:43information. I'll say research how to
10:45use Claude co-work for business and
10:47surface what most people are missing.
10:48Something I could just send in a chat
10:50but I'll click this plus button and come
10:52down to research. And now watch what
10:54happens. Instead of a single search,
10:56Claude operates agentically. I can open
10:58this up in the side panel to see exactly
11:00what's happening. It comes up with a
11:01plan first then conducts multiple
11:03searches that build on each other just
11:05adapting its path and exploring
11:07different angles of your question all on
11:08its own. So Claude spent 5 minutes, went
11:11to 169 sources, synthesized all that
11:14information down into this document
11:16that's fully cited. It can take longer
11:17for more complex investigations but it
11:20does work that would typically require
11:21hours of manual research in minutes and
11:23provides all the citations to make
11:25verification easy. Now here's where
11:27things get different. Everything up to
11:29this point has been about what Claude
11:31can do in a single conversation,
11:32chatting and creating. This next part is
11:35about building a system which needs to
11:37remember and connect things.
11:40Using projects correctly is one of the
11:42most impactful changes you can make to
11:44be more effective with Claude. The
11:46projects are self-contained workspaces
11:48with their own memory, chat histories,
11:50knowledge bases, and custom
11:52instructions. Dedicated environments for
11:54specific workstreams. Within each
11:56project you can input different custom
11:58instructions that only apply to that
12:00project. It's far more effective than
12:02using them at the account level like I
12:04mentioned earlier. I use projects for
12:06everything. If you're new to them, just
12:07start with two or three to learn how
12:09they work. The best candidates for a
12:10project are areas where you have
12:12reference materials you'll use
12:14repeatedly, consistent requirements for
12:15how Claude should respond, or just to
12:17stay organized and find previous chats
12:19easier. Click new project, give it a
12:21name, then add a short description if
12:23you want. This doesn't affect how it
12:25functions. Once it's created, come over
12:27to instructions and that's where you can
12:29add all those custom instructions. Now
12:31good project instructions typically
12:33include context about what you're
12:35working on, process instructions, tone
12:37and style preferences, any other
12:39specific requirements. You also have
12:41project files where you can upload any
12:43sort of document you'd like Claude to
12:45have persistent knowledge of in this
12:47project. Reference documents, you know,
12:49brand guidelines, style guides,
12:50templates, examples of work you want
12:52Claude to emulate like newsletters and
12:55blog posts. So upload your brand docs,
12:57tone guide, or SOPs, write a custom
12:59instruction set, and every conversation
13:01inside that project inherits that
13:03context automatically. Now you don't
13:05have to re-explain things across chats,
13:07rewrite context, or upload the same
13:09files repeatedly. It's all there every
13:11time. That's a huge improvement in the
13:13experience. But here's the limitation of
13:15projects alone. They remember things but
13:18they don't know how to do things. That's
13:20what skills are for.
13:23Skills are preset workflows you build
13:25once and Claude pulls them in
13:27automatically whenever they're relevant.
13:29They're little expertise packages. They
13:31teach Claude how to complete specific
13:32tasks in a repeatable way. And there are
13:34two categories of skills. The Anthropic
13:36skills are created and maintained by
13:38Anthropic. They are available to all
13:40paid users and Claude invokes them
13:42automatically when relevant. Custom
13:44skills are ones you or your organization
13:46create for specialized workflows and
13:48domain specific tasks. For example, you
13:50might create a skill that applies your
13:52company's brand guidelines to
13:53presentations or maybe it structures
13:55meeting notes in a specific format or
13:57executes your organization's data
13:58analysis workflows. It's really easy to
14:01create these. Just tell Claude what you
14:02want to create and answer Claude's
14:04questions. It will interview you about
14:06your workflow. Also upload a reference
14:08materials if you have them or an example
14:11of the output you're looking for. All
14:12these things will help Claude understand
14:14exactly what you're looking for. It will
14:15analyze everything, come up with the
14:17whole workflow, then there will just be
14:18a button to copy it to your skills. You
14:20can also download the dot skill file to
14:22your computer if you want. You can see
14:24all your skills under customize. Then
14:26from that point forward, Claude will
14:28automatically invoke it whenever you
14:30work on relevant tasks. You don't have
14:31to trigger it manually. Another way I'll
14:33create these skills is if you've gone
14:35back and forth with Claude iterating to
14:37get to a final output and that's
14:39something you think you may need to do
14:41again in the future, then I'll ask it to
14:42turn that into a skill. Claude will
14:44synthesize it all into a more refined
14:46workflow. I'll show you a few skills in
14:48action. I have four different skills I
14:50use regularly for YouTube. One is my
14:52script critique skill. The way I write
14:54scripts is I write out the full thing
14:56myself, I'll refine it a bit and get to
14:58a good spot. I do this in my scripting
15:00project that has custom instructions.
15:02I'll paste in my full script, then I'll
15:04just say critique this script and that
15:06will invoke the skill. Now my
15:07instructions specifically ask it not to
15:10rewrite anything. And the output format
15:12is as an artifact. It starts with a
15:13paragraph with a high-level critique,
15:15then it has the full script intact but
15:18after each section it lists issues for
15:20me to review. I have a big list of
15:22things to look for and different
15:23techniques to use like identify areas
15:25where things are unclear that could
15:26benefit from a better definition or
15:28analogy, areas where viewers may be left
15:31with unanswered questions, spots that
15:32are just too boring, or areas where I
15:34use hedging language too much and some
15:37other things. After that, I can go
15:38section by section refining it in the
15:40artifact. I don't really have Claude
15:42rewrite things. It just gives me
15:44suggestions on how it could be improved.
15:46The only section I sometimes end up with
15:48a big rewrite on is the intro. I have a
15:51separate skill for that. So it uses my
15:53existing intro plus the whole script as
15:55input. Then it will craft three
15:57different options for intros based on a
15:59big document of guidelines and examples.
16:01Then I can take inspiration from those.
16:03I also have a title skill that will
16:04write titles based on the script and
16:06also allows me to just input a title
16:08idea and it will explode that out into
16:10new ideas and variations on the
16:12original. I also have a skill for
16:13writing the summary that I put in
16:15YouTube descriptions. That's the one
16:16part I just copy and paste as is.
16:19The four YouTube skills I just described
16:21save me hours every single week. And
16:23every workflow you have regularly is a
16:25candidate for the same thing. They are
16:27the most underutilized feature in Claude
16:29and they work really well when combined
16:31with projects like this. Projects are
16:33persistent, reusable knowledge. Skills
16:36are persistent, reusable processes.
16:39Connectors allow Claude to access
16:41different apps to read information and
16:43perform actions on your behalf.
16:45Depending on the connector and
16:47permissions you grant, Claude can search
16:49your files, retrieve documents, analyze
16:51data, create new content, update
16:53records, and execute tasks across your
16:55connected applications all from within
16:58your conversation. Google Drive, Google
17:00Calendars, Slack, and Asana are some
17:03common ones but there's way more in
17:05here. So here's one I use. I have
17:07Granola connected. After any meeting,
17:09like after a YouTube strategy call, I
17:11can ask Claude to pull the notes in and
17:13begin expanding on those ideas. I also
17:15use Gamma to create better
17:16presentations. There's tons and tons of
17:18others. Whatever tools you use, you can
17:20connect them and bring that information
17:22into your workflows in Claude. Then
17:24Claude becomes the command center for
17:26your entire workflow.
17:29Now there are different models you can
17:30use. These change all the time so I
17:32saved them for the end. Opus is the most
17:34powerful model for complex tasks. That
17:37does take longer and uses more tokens
17:39which are basically Claude's currency
17:41for how much processing a conversation
17:43uses. And you get a limited amount
17:45depending on which plan you're on.
17:46There's a 5-hour rolling window on
17:48tokens and also a weekly cap. Sonnet is
17:52much faster and is still very powerful
17:54especially with extended thinking turned
17:56on. That lets Claude work through
17:58complex problems more carefully before
18:00responding. Haiku I never use. It's very
18:02fast but not as good. The Sonnet with
18:05extended thinking is my default then
18:07whenever I'm building anything I switch
18:09to Opus. There is so much in here beyond
18:11just getting better at prompting. Go
18:13from chatting to creating to connecting
18:15it all together to make it more
18:16powerful. If you do one thing after this
18:18video, set up a project. Then build one
18:21skill. That's the fastest path to a
18:23system that actually works for you. I
18:25also want to mention there is a desktop
18:26app for Claude as well. When you
18:28download that, it has three tabs. One
18:30for chat which is the same interface and
18:32all the features we just covered. Once
18:34you've mastered that, there's another
18:35tab for co-work which is an agentic AI
18:38assistant. It operates directly on your
18:40computer to handle multi-step tasks.
18:42Things like managing files, organizing
18:44documents, and summarizing data. It
18:46works as a co-worker rather than a
18:48chatbot. It autonomously executes tasks
18:51to deliver finished results. It is
18:53amazing. I have an entire video on that.
18:55I'll put it right here. And there's also
18:56Claude Code which is an agentic coding
18:58tool that handles entire development
19:00workflows through natural language. It's
19:02much easier to use than people expect
19:04and requires no coding knowledge or
19:06technical ability. I have a full guide
19:08on Claude Code coming soon so make sure
19:10you're subscribed for that.

About this transcript

This page contains the full transcript of Full Claude Tutorial: Beginner to Advanced in 19 Minutes by Futurepedia, generated from the public captions YouTube serves with the video. The transcript has 3,921 words across 601 segments, with the original timestamps preserved so you can click any line to jump to that moment in the embedded player.

What you can do with it

Use the transcript to take notes, quote the speaker, build a study guide, generate a summary with ChatGPT or Claude via the YouTube Summary tool, or export it as a timed subtitle file with YouTube to SRT. You can also re-open it in the transcriber to translate the transcript into 100+ languages.

Free YouTube transcript tool

YouTube2Text is a free YouTube transcript generator — no signup, no daily limit. Paste any YouTube link and get the full transcript instantly, with timestamps, click-to-jump, translation to 100+ languages, AI prompts for ChatGPT, Claude, and Gemini, and exports to TXT, SRT, VTT, or Markdown.