The Foundation
Configure it once. Use it everywhere after.
Layer 0 of the operating system: an AI teammate configured once with your knowledge, your standards and your governance, so that every stage of the loop runs on the same reliable engine. Configure it now; use it everywhere after.
Before you run the loop you need something to run it on. Not a chat window you re-explain yourself to every morning, but a configured teammate that already knows your business and holds itself to your standards. This is the one part of the book you do once.
Pick a platform and a project
Use one of the two platforms with a proper project structure: ChatGPT Projects or Claude Projects. A project keeps custom instructions and uploaded knowledge attached to every conversation, which is the whole point. Everything in this book works in either. Choose on feel and, far more importantly, on data terms.
Use a paid tier with clear commitments on training and retention, and confirm your employer permits it before you upload anything. Free consumer tiers may use your inputs to improve the models. This decision outranks every feature. If your company has an approved enterprise instance, use that.
Give it an identity
Configure the teammate as a role with standards, not a search box with a personality. A strong identity answers four questions: who it is, what it does, what it never does, and how it carries itself. The Setup Skill in Appendix A.2 builds this for you through conversation.
Who. A senior sales research analyst and deal coach supporting an enterprise seller.
What. Research, hypotheses, engagement drafts, discovery prep, and honest deal coaching.
Never. Makes commercial decisions, invents facts, or writes final client comms without review.
How. Concise, commercial, direct; separates what it knows from what it is guessing; would rather ask a sharp question than fill a gap.
Think of it as a capable junior analyst, not an oracle. Good junior analysts make mistakes, welcome being challenged, and get sharper the more precisely you brief them. Treat yours the same way, and scepticism stops being a limitation of the tool and becomes a feature of how you use it. The rule to configure it against is the same rule that runs through this whole book: the AI prepares, you interrogate, verify and decide.
Feed it your knowledge
Upload the material that turns a generic model into your teammate: public website and product content, public case studies, your ICP, buyer personas, value proposition and messaging, your discovery methodology, and competitive positioning. Curate, do not dump. A tidy one page summary beats a forty slide deck. And upload only what is public or unambiguously yours to use.
Confidential proposals, pricing, NDA material, internal strategy, personal data and CRM exports never go in. When you want help on a real, sensitive deal, abstract it: strip the identifying detail and describe the shape of the situation. The teammate reasons about patterns perfectly well without the particulars.
Do not fill in a form. Open your AI, run the SIGNAL Setup Skill (Appendix A.2), and let it do the heavy lifting. It works best when you feed it real material, so before you start, have these ready: your website URL, an external-facing pitch or capability deck you can upload, and any public one-pagers or case studies.
The Setup Skill then reads your material, searches for your company where it can, drafts a first version, interviews you only to fill the gaps, and shows you everything it assumed so you can confirm or correct before anything is saved. The result is your Permanent Knowledge document, built from your real assets rather than from memory.
If you have no deck to upload and no web search on your plan, the Setup Skill falls back to interviewing you directly. You still get a complete document. Adding assets and search simply makes it faster and sharper. And you can rerun it any time your positioning changes; each version is a permanent improvement to everything the teammate produces afterwards.
When the AI reads your website or searches the web, it will occasionally infer something wrong or repeat an outdated claim. The Setup Skill forces a validation step where it shows you every assumption before writing your knowledge document. Confirm or correct each one. Never let an unchecked assumption become permanent knowledge.
Teach it to think, and to coach
This is what separates a teammate you trust from a confident guesser. Build these standards into the Project Instructions, then test each one. They are the difference between output you can act on and output you have to check line by line.
- Separate evidence from inference, and label the inferences.
- Never invent. If it does not have the information, it says so and asks.
- Challenge my thinking, including when my logic is thin or my plan has a hole.
- Name what is missing: the stakeholder, the criterion, the competitor.
- Offer two or three hypotheses, not one confident guess.
- State confidence honestly, from well supported to guess.
- Think commercially, always ending with what it means for the deal.
- Coach, do not flatter. Act like a sceptical sales manager, not a copywriter.
Every seller I have coached wanted a tool that made them look good. The ones who improved wanted a tool that told them the truth. The eighth standard is the one that matters most. A teammate that agrees with you is just a faster way to be wrong.
Knowledge and workflow are two different things
Keep this distinction clear, because it explains why the teammate works.
Your company, services, ICP, messaging, competitors and case studies.
The research, discovery, qualification and coaching it runs.
You need both. Knowledge without workflow is a filing cabinet. Workflow without knowledge is a clever stranger. Together they are a teammate.
Getting good output
You do not need to be an AI expert, but a little literacy pays off immediately. The durable principles hold on any platform, this year and next.
- Feed it, do not quiz it. Paste the source material. A grounded teammate beats one working from memory, which is where invention starts.
- Iterate, do not accept. The first answer is a draft. “Sharper,” “shorter,” “now challenge that” gets you the real value.
- Know what good looks like. Good output separates evidence from inference, states its confidence, and ends with what it means commercially. Weak output is confident, generic and unsourced.
- Fix the source, not the symptom. When it is repeatedly wrong, the fix is almost always in the instructions or the knowledge, not the next prompt.
- Assume it can be confidently wrong. These models invent facts that look entirely plausible: names, titles, dates, figures, quotes. Fluency is not accuracy. Check anything you intend to say to a customer against the real source. This is the single most important habit on this page.
The exact clicks in ChatGPT and Claude change every few months, so this book stays principle led. The step-by-step, screenshot-level walkthroughs are best learned from real use, which is exactly what the month ahead is for. Build the habit now; the buttons will follow.
Worksheet 3.1 · Foundation build confirmation
Do not start the loop until every line is ticked.
- Platform chosen and its data terms confirmed.
- Employer AI policy checked and complied with.
- Project created; Companion Skill from Appendix A.3 pasted as custom instructions.
- Cleared knowledge uploaded; nothing confidential included.
- All eight reasoning standards tested, including at least one where the teammate challenged you.
What to do right now
You now have the Foundation concept. The next step is to load both skills into your Project. They live in the Appendix of this workbook as copy-paste ready blocks.
Run the Setup Skill. Go to Appendix A.2. Paste the Setup Skill into a new chat and run it. It will interview you and produce your complete Permanent Knowledge document in one conversation.
Load the Companion Skill. Once the Setup Skill is done, go to Appendix A.3. Paste the Companion Skill into your Project as the custom instructions. This is the skill you use on every account from here on.
Type start. The method begins.
Want the whole method as one document?The complete field guide has the full walkthrough, all six build exercises, and the appendix of ready-to-paste prompts, including the Setup and Companion Skills.
Download the Field Guide