How to build your AI board of directors: the complete field guide

The original version of this post came out in September 2025. Outlined the concept in eight steps, kept it accessible, included a pitch at the bottom for a done-for-you version. More people reached out about that post than anything else I'd written all year. Not to buy. To ask for the rest of the instructions. Specifically: how do you actually do Step 6? Where do you find the material? How do you pull YouTube transcripts? What do you do when two AI tools give you conflicting research on the same person?

This is the full version. The prompts, the specific tools, the exact workflow. If you have 20 hours and patience for research, you can build a board that knows your business and reasons from a real intellectual framework. If you'd rather skip the build and go straight to the output, the link at the bottom is still there.

Read the whole thing before you start. Steps 1 through 5 move fast. Steps 6 and 7 are where the actual work is.

The problem this solves

Generic AI gives generic advice. That's the whole problem. Ask an assistant that knows nothing about your business for pricing strategy help and you get a framework: research your market, consider your value proposition, test different price points. That's not advice. That's a restatement of conventional wisdom dressed up as guidance.

What makes an actual advisor valuable is that they know your specific situation and reason from a real intellectual framework. Not the conventional answer. The answer for your business, filtered through their model of what actually works.

An AI board of directors fills that gap. Each member has a knowledge file built from their actual published thinking. Each one knows your company from the profile you built in Step 1. Those two things together produce something fundamentally different from asking a general-purpose assistant a question.

Traditional advisory options have the framework piece but not the accessibility:

• Executive coaches: $2,000–5,000 per month for one perspective
• Business consultants: $10,000+ for a project that may not fit your reality
• Peer masterminds: great networking, limited specialized expertise
• Industry associations: broad guidance, not specific to your situation

An AI board doesn't replace human advisors for major pivots. It gives you a team available at 10pm, one that knows your business in detail and doesn't charge by the hour for a 15-minute question.

Here's the honest caveat: this takes 15–30 hours to do well. Most people start enthusiastically and stall out in the deep research phase. That's fine. Either do the work or have us do it. But don't build it halfway. A board with shallow research gives shallow advice, and then you've got a tool you trust that's going to mislead you.

Step 1: Build your company knowledge file

Every board member will read this file before every conversation. It's the single most important input in the entire system. Get it right and your advisors will give you advice that actually applies to your situation. Get it wrong and you'll get generic output.

The knowledge file should cover:

• Company history: when you started, how you've evolved, what inflection points shaped the business
• What you actually do: not the elevator pitch, but the operational reality: who you serve, how you find them, how you deliver, how you bill
• Revenue and growth: current revenue, growth rate, revenue model (project, retainer, product, recurring)
• Team structure: who does what, headcount, contractors vs. employees, gaps
• Current challenges: the 3–5 things that are actually keeping you up at night right now
• Goals for the next 12 months: specific and measurable where possible
• Competitive landscape: who you lose deals to and why, who you beat and why
• Customer profile: who your best clients are, what they have in common, what they value most about working with you
• What you've already tried: strategies that didn't work, advisors you've had, tools you've implemented

This doesn't need to be a polished document. A raw, honest brain dump is more useful than a sanitized company overview.

Budget 45–60 minutes for this interview. If you rush it, your board will give generic advice. The more specific you are, the more useful every future conversation becomes.

Step 2: Deep market and competitive research

Your board members need to understand the market you operate in. This step asks AI to do the research for you, but you need to give it a focused brief.

Run this prompt in both Claude and ChatGPT separately, then compare the outputs. They'll often surface different angles. Claude tends to be more analytical; ChatGPT more exhaustive on breadth. Take the best from each.

Step 3: Strategic gap analysis

Before you can pick board members, you need to know where your strategic gaps are. This step surfaces the 5–7 areas where outside expertise would most change your trajectory.

Common gaps for businesses at the $750K–$3M stage: pricing strategy, team scaling, lead generation systematization, customer retention, financial modeling, operational efficiency, and brand positioning. Your list will be specific to your business. That's the point.

Step 4: The AI avatar hunt

Now you're looking for real people whose expertise maps directly to your gaps. You're not hiring them. You're modeling AI personas after their knowledge, frameworks, and communication style.

Good candidates are people who have:

• Published books, articles, or substantial public content in their domain
• Given long-form interviews (podcasts, YouTube, conference keynotes) where they explain their thinking in detail
• Actually operated businesses in your industry or at your scale (not just advised on it)
• A distinctive point of view, not just conventional wisdom

Step 5: Board member selection

From the candidates identified in Step 4, you're selecting your final board. Aim for 5–7 members. Fewer than five leaves functional gaps. More than seven creates conflicting advice and makes the system unwieldy.

A well-balanced board for a $1–3M service business typically covers:

• Marketing and demand generation: how to find and attract ideal customers
• Operations and delivery: how to scale without breaking what works
• Finance and pricing: margin, cash flow, pricing strategy, financial modeling
• People and culture: hiring, team structure, retention, leadership development
• Strategic planning: long-range vision, positioning, competitive strategy

You can add domain specialists (a sales expert if your close rate is the bottleneck, a technology advisor if you're building product, a PR strategist if brand authority is the gap), but fill the five core seats first.

Step 6: Deep research on each board member

This is where most people stall out. And it's the most important step.

A board member built from three Google searches gives you generic output. A board member built from 200 pages of their actual thinking, in their own words, gives you something that feels genuinely like talking to that person.

For each board member you're going to build a research file. Think of it as their intellectual biography: not their resume, but their frameworks, their mental models, their opinions, their patterns of reasoning.

Where to find source material

Books: Google Books and Internet Archive. Google Books shows substantial previews for most business books, sometimes the full text. Search for your target's name on books.google.com. For older books (published pre-1928 in the US, or later works where the author donated rights), check archive.org and gutenberg.org. HathiTrust Digital Library has academic and historical works that aren't on either platform.

Even partial book previews are valuable. Five chapters of Peter Drucker's actual prose is worth more than fifty summaries of it.

Podcast transcripts: Most major podcast platforms don't publish transcripts, but many show notes include them. Search "[person's name] podcast transcript" on Google. Otter.ai and Descript have indexed a large corpus of podcast transcripts. Some podcasters post full transcripts on Substack. When you find a transcript, paste it directly into your knowledge file.

Long-form interviews: Charlie Rose interviews, Lex Fridman episodes, Tim Ferriss show transcripts, and similar long-form formats are goldmines. These are places where people explain their thinking over 1–3 hours rather than giving 60-second sound bites. Ferriss publishes full transcripts on his blog. Search for them.

Conference keynotes and speeches: TED Talks have machine transcripts. YouTube auto-generates subtitles for most content. Industry conference recordings often end up on YouTube. These give you both the content and the communication style: how they phrase things, what metaphors they use, what they emphasize.

How to scrape a YouTube channel

If your target has a YouTube channel with substantial content (even 20–30 videos), that's potentially hundreds of pages of searchable transcript. The transcripts are free. You just need to pull them.

There are two ways to do this depending on your comfort level with tools.

# macOS (with Homebrew)
brew install yt-dlp

# Windows (with pip)
pip install yt-dlp

# Or download the binary from github.com/yt-dlp/yt-dlp/releases

yt-dlp --write-auto-sub --sub-lang en --skip-download \
  --output "%(title)s.%(ext)s" \
  "https://www.youtube.com/@channelname"

Whichever method you use, the auto-generated transcripts aren't perfect. They miss punctuation and occasionally garble proper nouns. Run them through a quick cleanup prompt before uploading:

How to use public domain works

For board members modeled after historical business thinkers (Peter Drucker, Jim Collins' early work, Dale Carnegie, Napoleon Hill, W. Edwards Deming), much of their foundational writing is either in the public domain or available in full via Google Books previews.

Key public domain resources:

• Project Gutenberg (gutenberg.org): mostly pre-1928 works, but includes early business and management classics
• Internet Archive (archive.org): books donated by authors, out-of-print works, and scanned texts; search is excellent
• HathiTrust Digital Library (hathitrust.org): academic and historical works; some require a library login but many are publicly accessible
• Google Books: substantial previews even for copyrighted works; full text for many older works

The copyright threshold varies by country and publication date, but as a rule: anything published in the United States before 1928 is in the public domain. Works published 1928–1977 depend on whether copyright was renewed. Works by authors who died more than 70 years ago may be in the public domain in many jurisdictions.

You don't need to be a copyright lawyer here. Use Project Gutenberg and Internet Archive confidently. They've already done the clearance work. For everything else, use Google Books previews, which are explicitly licensed for display.

Combining outputs from multiple research tools

Claude, ChatGPT, and Perplexity each have different strengths when it comes to research synthesis:

• Perplexity is best at pulling from the live web (recent interviews, news, current publications). Use it first to find what exists and where.
• ChatGPT (with browsing or the Deep Research feature) is good at comprehensive synthesis across a large corpus of sources.
• Claude has the largest context window and is best at synthesizing long documents you paste directly: full transcripts, chapter excerpts, multiple articles at once.

The workflow that works best:

1. Use Perplexity to find source URLs and identify the best available material for your target
2. Pull the actual source text (transcripts, articles, book excerpts) into a document
3. Paste everything into Claude with a synthesis prompt to distill their frameworks and mental models
4. Ask ChatGPT to independently synthesize the same material
5. Use a comparison prompt to identify what each synthesis missed, then combine into a final profile

Step 7: System integration and GPT creation

You have your research files. Now you build the actual advisor.

You can do this in a custom GPT (requires ChatGPT Plus, $20/month) or a Claude Project (requires Claude Pro, $20/month). Claude Projects have the advantage of a larger context window, which means more source material. Custom GPTs have a more polished interface for multiple personas. Many people build both.

Writing the system prompt

The system prompt is the instructions you give the AI about who it is and how to behave. This is the most technically demanding part of the process. A weak system prompt produces generic output. A strong one produces advice that genuinely sounds like the person you researched.

A well-structured system prompt for an AI board member has five components:

1. Identity declaration: who they are, their specific expertise, their career arc in 2–3 sentences
2. Core philosophy: their 3–5 fundamental beliefs about their domain, pulled from your research
3. How they advise: their approach to giving advice (direct, Socratic, framework-first, example-first)
4. What they push back on: their strong opinions about what NOT to do, based on your research
5. Interaction instructions: how they should respond to questions, including asking clarifying questions before answering, referencing the company knowledge file, and staying in character

Structuring knowledge files

Upload your research material as knowledge files attached to the GPT or Claude Project. Structure matters. Clean, labeled files are easier for the AI to retrieve from than a single giant dump.

Organize files by type:

• company-profile.txt: your company knowledge file from Step 1
• [name]-philosophy.txt: the synthesized advisor profile from Step 6
• [name]-transcripts.txt: cleaned transcripts; paste multiple shorter ones into a single file
• [name]-writing.txt: book excerpts, articles, published content

For Claude Projects, you can paste the content directly into the Project instructions and Files section. For custom GPTs, use the Knowledge section in the GPT editor. Aim for 50,000–100,000 words of source material per board member if you can get it. More is better as long as it's actually their thinking and not summaries of their thinking.

Testing before you rely on it

Before your first real strategic session, run a calibration test. Ask the board member a question you already know the answer to from your research, something they've covered in their writing. Does the response match their actual perspective? Does it feel like them, or does it feel generic?

If the output is generic, the problem is almost always the system prompt, not the research files. Tighten the philosophy section and the communication style instructions. Be more specific about what makes this person's advice distinctive from a generic business advisor.

Step 8: First session and ongoing cadence

Your board is built. Here's how to get value from it.

The first session

Start with an introduction conversation. Share your company knowledge file and ask each board member to read it and tell you what they see as your most important strategic opportunity and your most significant risk. You'll learn a lot about how well each persona was built from how they respond to this prompt.

Ongoing cadence

The best use of an AI board of directors is a weekly check-in, not a quarterly summit. A 15-minute conversation every Monday morning with your most relevant board member, whatever challenge you're navigating that week, compounds over time faster than a monthly deep dive.

Common session formats that work well:

• Decision review: "I'm deciding between X and Y. Here's the full context. What would you advise and why?"
• Problem decomposition: "I'm stuck on this problem. Help me understand why it's actually hard and what I might be missing."
• Pre-mortem: "I'm about to do X. Walk me through how this fails. What are the three most likely ways this goes wrong?"
• Scenario planning: "If [X market condition] happens, what should I do differently than I'm currently planning?"
• Quarterly review: share your progress on the goals from your knowledge file; ask the board to push back on your interpretation of what's working

Update your company knowledge file quarterly. As your business evolves, the profile the board is working from should evolve with it. An advisor working from a year-old company profile will give you advice that's misaligned with your current reality.

Tools and platforms

Everything you need to build this system:

• ChatGPT Plus ($20/month): custom GPT builder with knowledge file uploads
• Claude Pro ($20/month): Projects with large context window for extensive knowledge files
• Perplexity (free tier works): live web research to identify source material
• yt-dlp (free, open source): transcript extraction from YouTube channels
• Google Books (free): book previews and full text for public domain works
• Internet Archive (free): full public domain books, donated works
• Project Gutenberg (free): pre-1928 classics, fully downloadable
• Otter.ai (free tier): podcast transcript search

Total monthly cost for the platform: $20–40 depending on which AI tools you're already paying for. The real investment is time. 15–30 hours upfront, 1–2 hours per week ongoing.

What to do if you get stuck

The most common sticking points and how to get past them:

Can't find enough source material for a board member: Reconsider your candidate. A strong AI persona requires at least 20,000 words of source material in the person's actual voice. If you can't find that, pick someone else who has written more or appears on more podcasts. Don't try to fake depth with summaries. It produces generic output.

Board member sounds generic despite good research: The problem is the system prompt, not the files. Go back and be more specific about what makes their perspective distinctive. What would they say that a generic business advisor wouldn't? What would they push back on that most consultants let slide? The specificity of the philosophy section is what makes or breaks the persona.

Board members give conflicting advice: That's actually good. It mirrors a real board. When your marketing advisor and your operations advisor give you opposite recommendations, you're seeing a real tension in your business. Run a synthesis prompt: "Here's what my marketing advisor says and here's what my operations advisor says. What's the actual tradeoff I'm navigating, and how would I decide between these positions?"

Don't know which board member to consult: Start with the question: what domain is this problem actually in? Pricing is finance. Hiring is people. Go to that board member first. Then bring the same question to your strategic planning advisor with the domain expert's input already in hand. They'll synthesize across functions better than any single advisor will.