This is the one prompt we use to produce a 20-page founder business audit: a visual, cited report on a founder and their company that shows their best proof, the gap between what they sell and what a stranger can verify, and the easy wins their team or agents can do this week. Copy it, fill in the brackets, and paste it into whichever AI you use.
Who this is for
Use it when you want to give a founder real value before asking for anything: a friend you just met at a conference, a prospect before a strategy call, a partner’s client, or yourself. The report earns the next conversation because it shows you understand their goal, found their strongest proof, and did the work to make it obvious.
It is the front door to our maintained method, How We Audit. The prompt tells the AI what to do; the method and the 20-page blueprint behind it say what “pass” means. If the two ever disagree, the method wins.
The master prompt
Fill in the brackets. Leave a bracket as “unknown” if you don’t know it. The prompt tells the AI to label guesses instead of stalling.
You are preparing a 20-page founder business audit for [FOUNDER NAME] of [COMPANY] ([WEBSITE]). Prepared by: [YOUR NAME] and [PARTNER NAME, if any]. Deliver to: [EMAIL THREAD, PROJECT, OR CHAT]. GOAL Give [FOUNDER] real value before asking for anything. When they finish page 2, they should feel understood, see proof they forgot they had, and know the first three things to do this week. DONE WHEN A 20-page PDF passes every check in step 5 and is delivered as described in step 6. 1. CONTEXT — use what is here; do not ask me for anything already answered - Who they are and how I know them: [where we met, what we talked about, why I want to help] - Their goal for the next 90 days, in their words if possible: [GOAL or "unknown"] - Who they need to trust them and buy: [BUYER or "unknown"] - Proof they mentioned: [1–3 results, clients, talks, interviews, videos] - Handle with care: [anything sensitive, or "none"] If the founder is with me, let them speak: who they are, what the firm does, where they are credible, and what success looks like. Treat their words as the brief. If goal or buyer is unknown, infer it from public evidence, label it INFERRED, and keep going. 2. RESEARCH — inventory everything public Find and verify: websites, LinkedIn, YouTube, podcasts they hosted or appeared on, speaking, press, books, awards, reviews, case studies, clients, team members, partners, and every positive mention. Confirm identity with two independent facts and rule out namesakes. Check what Google, ChatGPT, Gemini and Perplexity say about them and their company, and whether a Knowledge Panel exists. Put every item in a sheet: link, date, what it proves, and a score on the 30-point authority scale (10 Who, 10 Where, 10 What). Label each claim OBSERVED, INFERRED or UNKNOWN. A bio they wrote is a claim, not independent proof. 3. ANALYSIS — find what is not obvious - Greatest hits: the 10 strongest pieces of proof, and where each one is buried. - Network map: people and organizations they are publicly connected to. A photo together is not a friendship or an endorsement. - Proof vs. promise: what they sell versus what a stranger can verify. That gap is the plan. - Market: SWOT and five-forces view of how they compete. - The magic berry: the one asset that, amplified, does the most. - Search and AI: name search, entity consistency, Knowledge Panel state, and whether AI recommends them for their buyer's real questions. - Dollar a Day: proven organic winners worth a $1/day test. Frame findings as opportunities. Be truthful, not flattering. Nobody should look bad or get in trouble for reading this report. 4. THE 20-PAGE PDF — built like a magazine, not a wall of AI text - Page 1: cover with a real photo of the founder, "Prepared by" with headshots, four headline facts. - Page 2: executive summary — 5 to 8 non-obvious insights, each with its evidence and the action it leads to. No biography recap. - Pages 3–19: goal and buyer; identity and search; scores; greatest hits; network map; highest-value assets; offer and funnel; Dollar a Day; targeting; content plan (Produce → Process → Post → Promote); topic wheel; entity plan; Knowledge Panel and AI visibility; agent plan; 30/60/90-day plan; risks and open questions. - Page 20: method, sources, and how to check our work. Every page: one clear takeaway, one real visual (photo, screenshot, chart or diagram), and a citation beside every fact. Real photos only — no stock art, no AI-generated faces. Write so a fifth grader could follow it. Action plan: red/yellow/green by priority. Each item gets an owner, a due date, and how we will know it is done. Flag easy wins that take 10 minutes or less. Mark which an agent can do and which need a person. Include a QR code with the printed link https://localservicespotlight.com/install/ so their agents can install the skills and start on the plan. 5. QA — loop at least twice before saying done Check: exactly 20 pages; every fact has a citation and every link opens; page 1 stands alone; no placeholder text; every page has a real visual; the QR scans to the install page; no private phone numbers, emails or addresses appear. Fix every failure and re-check. List any check you could not run as UNTESTED — never as passed. 6. DELIVER Reply on the existing thread with the PDF attached and a three-sentence note: the strongest finding, the first easy win, and the one decision we need from them. Then write a short meta article on what you did and what to do better next time, and set a follow-up for four business days later.
Read the audit methodInstall the audit skills
Why each part is there
Every line in the prompt came from a correction Dennis made while dictating audits to agents. A request only made it in if the earlier reports were worse without it.
| Part of the prompt | The problem it fixes | First asked for |
|---|---|---|
| Goal and “done when” at the top | Agents started researching before they knew what finished looked like. | Oct 2026, after a review of 8,106 dictations found “done when” said only 4 times |
| Let the founder speak | Second-hand context missed what the founder actually wants. At live events the founder now describes the firm and success in their own words. | Oct 2026, Hypergrowth Singapore |
| 30-point authority scale | Without a score, every mention looked equally important. | June 2026 |
| Proof vs. promise | Reports listed problems instead of the one gap that drives the plan: what they promise versus what anyone can verify. | August 2026 |
| Opportunities, not problems | A report that embarrasses the marketing team gets buried, not used. | August 2026 |
| Non-obvious executive summary | Page 1 kept opening with a flattering bio. Readers want the surprise that changes their next decision. | August 2026, repeated in nearly every request since |
| Real photos, diagrams, colour | Text-only reports “looked like AI wrote it” and went unread. | June 2026 |
| Red/yellow/green plan with owners | Recommendations without owners and dates never happened. | August 2026 |
| QR code to install agents | Founders agreed with the plan but had no way to start it. | September 2026 |
| Citations beside every fact | Founders often pay agencies and consultants who need to check our numbers. | October 2026 |
| Truthful, not flattering | Praise felt good and changed nothing. | October 2026 |
| Easy wins under 10 minutes | A 20-page plan needs a first step small enough to do today. | October 2026 |
| QA loop and meta article | Agents said “done” before checking. Each run should make the next one better. | August 2026 |
How to fill it in fast
- Speak it, don’t type it. Dictate the context section right after you meet someone. The details are freshest then. Our dictation study shows why voice briefs carry more useful context than typed ones.
- Give the goal even if you’re guessing. “Unknown” is fine. A guess the founder can correct is better than a blank.
- Name one to three pieces of proof. Search engines miss a founder’s best talk or client story more often than you’d think.
- Say where it goes. The existing email thread, project, or chat. One place, so nobody has to forward anything.
What a finished audit looks like
These published audits were produced with earlier versions of this prompt:
- Hoopin Nate: a creator with a billion views and almost no search presence.
- Ryan White: 3 million followers and zero organic search keywords, with a claims-versus-evidence ledger.
- Garrett McClure: an agency roll-up president whose homepage still had placeholder text.
- George Leith: a well-connected founder’s network mapped into a plan.
Common questions
Does it have to be exactly 20 pages?
Yes. A fixed length forces choices: one takeaway per page, nothing padded. Raw research goes in a separate sheet, not extra pages.
Which AI should I use?
Any of them. The prompt is model-independent. Use the strongest model you have for analysis and QA, and a cheaper one for bulk research if your setup can split the work.
What if the founder doesn’t have much proof online?
That is the finding. The audit shows where their real work isn’t visible yet, and the plan becomes recording and publishing it. Missing proof is not the same as a bad reputation.
Can I run it on myself?
Yes. Put your own name in the brackets and ask for the truth, not compliments. It is often the most useful version.
How is this different from a Quick Audit?
The Quick Audit is a short local-search exam for service businesses. The founder audit covers the person and the company together: proof, network, search, AI visibility and an agent plan.
How we auditInstall the skills
Prompt compiled October 19, 2026 by an AI agent from 61 of Dennis Yu’s dictated audit requests (June–October 2026). It is a starting point for the maintained method at How We Audit, not a separate standard.
Originally published .

