If you lost your job, Claude could help you test a small product idea, but it cannot tell you whether strangers will buy it. In this video, Duncan Rogoff moves from researching complaints to selling a simple download in five steps. The useful lesson is to make a narrow answer to a problem you already understand, then let buyer conversations and a small launch decide whether it deserves more work.
Research a specific audience, verify one recurring problem, build one small deliverable, show how it works, and sell it on a clear page. Measure buyer conversations and purchases, not just views. Duncan says his own $6 templates brought in $8,923 in three months; that is his reported gross revenue, not a typical result, audited profit or a promise that this replaces a salary.
Watch Duncan Explain the Five Steps
Source note, 4 November 2026: Duncan's video and supplied transcript support the workflow, examples and personal earnings claims below. We checked Anthropic's access guidance, Gumroad's fee page, Reddit's API terms and the Arctic Shift repository. We did not recreate his Reddit search, audit his Gumroad account, or test a product launch. The example pilot and prompt are our editorial additions.
1. Find a Repeated Problem, Then Check the Evidence
Duncan starts with Reddit posts from people in an occupation he knows. In his marketing-coordinator example, the recurring frustration is a weekly client report that takes work to produce but barely gets read. He asks Claude to rank three problems, report the number of posts read and provide verbatim evidence. He then searches for a quote himself. That last step matters: a confident-looking quote or frequency count from an agent is not evidence until you can locate the underlying post.
The video uses Arctic Shift, a third-party archive of Reddit content. An archive can be incomplete or stale and may retain material a person later removed. It is not a blanket license to collect posts for a commercial product. If using Reddit's API, check Reddit's current terms, which require a separate agreement for commercial API use. For a first pass, read permitted public discussions manually, record URLs and dates, avoid storing personal data, and interview potential buyers directly. A complaint is a clue, not an order.
2. Choose One Buyer and One Outcome
Duncan's filters are sensible: choose a problem you know how to solve, can explain simply and want to keep working on. Add a buyer check. Who has this pain often enough to pay, and what have they already tried? A marketing consultant who sends ten client updates a month may value a better report template more than a student writing one assignment. Ask five people in that role for a recent example, how they handle it now and what a useful solution would save them. Do not open with your product pitch.
The goal is not to prove the idea by accumulating likes. It is to find a narrow job with evidence of existing effort, cost or frustration. If every interview says the report is annoying but not consequential, choose another problem before building.
3. Make the Smallest Useful Product
For the unread-report problem, Duncan asks Claude for a one-page template: the one number that matters, what changed, what we will do next. He requests an example with invented numbers clearly labeled as an example, then reviews the writing and exports a shareable file. That is an appropriate first product: a buyer can inspect it and use it immediately without learning a new application.
Bring your own method and examples. Claude can help structure the template and draft sample copy, but it should not invent client results, testimonials or expertise. Anthropic distinguishes Claude chat from Claude Code: ordinary chat is enough to outline and draft a checklist or PDF; Code helps when you need it to create files or automate a workflow and is included with paid plans or API access. Do not buy a larger tool stack before the product needs it.
4. Show the Work in Public
Duncan studies posts that catch his attention, identifies their hook and structure, and adapts the format to his own offer. The stronger version is to learn the job each part of a post performs, then write original material from your own evidence. For the report template, show a before-and-after page, explain why one metric belongs at the top, or walk through a fictitious example. Do not copy another creator's wording, imply a fake client result is real, or use private work without permission.
Pick the channel where your buyer already pays attention. Duncan favors LinkedIn for business buyers and sees video as a trust builder, but that is his experience, not a platform rule. Post enough to learn which explanation earns qualified questions. The aim is a clear route from useful content to the product page, not posting daily forever for its own sake.
5. Sell It, Then Learn From Buyers
Duncan suggests a plain Gumroad page that says who the download is for, what is inside and what it will help them do. His $6 price was an experiment, not a pricing formula. At that price, fees matter: Gumroad currently lists 10% plus $0.50 for a direct sale, with card processing of 2.9% plus $0.30 separately; marketplace discovery sales carry a 30% fee that includes processing. A $6 direct card sale would leave roughly $4.43 before tax, refunds and your time under those listed rates. Check the current checkout and your own payment method before committing to a price.
After a sale, ask what the buyer used, what they still had to change and what they were trying to accomplish. Duncan says an early automation template brought him service inquiries; that is an example of a product revealing a larger job, not a guaranteed path to consulting revenue. He also cites an unverified third-party claim of $175,000 monthly sales. Neither claim should set your forecast.
A Two-Week Product Test
This is the part I would add to the video: a stop rule. A product can be beautifully made and still lack a paying audience. Keep the first test small enough that a weak signal does not consume the time and cash you need after a layoff.
| Checkpoint | Evidence to collect | Decision |
|---|---|---|
| Problem | Five conversations or traceable public examples from the same buyer group. | Continue only if the pain is recurring and consequential. |
| Product | One complete sample and feedback on whether a buyer could use it. | Revise the deliverable before adding features. |
| Distribution | Qualified replies, page visits and how people found the offer. | Change the explanation or channel if attention is unrelated. |
| Economics | Purchases, refunds, fees and hours spent, not gross sales alone. | Continue, revise or stop against a budget set in advance. |
These are editorial checkpoints, not validated conversion benchmarks. A handful of views, a purchase from a friend, or enthusiastic comments do not prove demand. Conversely, two weeks is too short to declare every idea impossible; it is a useful limit on the first experiment.
Use the following original prompt to plan that experiment. It is a drafting aid, not a market-research result or Duncan's exact prompt.
Test one small product
Evidence, deliverable, distribution and a stop rule.
Video Chapters
Find and choose
Build and distribute
The longer game
| Time | Topic |
|---|---|
| 10:01 | Why create content today |
Sources and Limits
Checked 4 November 2026. This is a reading of Duncan's video, not independent proof of his income or a prediction of yours. Product and platform details can change.
| Source | Used for | Limit |
|---|---|---|
| Duncan's video | The five steps, report-template example and creator anecdotes. | First-person account; revenue and follower figures were not audited here. |
| Anthropic's Claude Code FAQ | Chat versus Code and paid-plan access. | Check the current plan and usage limits. |
| Gumroad fee guide | Direct, processing and marketplace fees. | Illustrative $6 math excludes tax, refunds and labor. |
| Arctic Shift repository and Reddit API terms | The archive Duncan mentions and a policy boundary for API use. | Neither establishes that archived content is complete or cleared for commercial use. |