Side Project Marketing Emails That Teach

Side project marketing emails that teach earn the next open by leaving the reader able to do something they could not do before the send — not by stacking five CTAs on a feature list. Teaching is the product sample. The pitch is one sentence at the end, after the idea is complete. That order is the whole strategy.

This post is for builders who need waitlist or launch mail without sounding like a drip campaign written by a model that has never used the repo. You will get a structure, a comparison of tools including an accurate Cadensend placement, and a workflow that keeps sources honest. Cadensend is an open-source, self-hosted email curriculum engine (MIT, GitHub, no hosted signup). It is not ConvertKit, not Ghost, not a bulk ESP, and the MVP sends to your verified address only. Use it to practice and store a sourced series. Use a real ESP when you have consent and a list.

Key Takeaways:

• One email, one skill; the CTA is optional and last.
• Ground every technical claim in a doc you would show a skeptical engineer.
• Cadensend compiles private curricula; ConvertKit-class tools send to opt-in lists. Do not swap the names.
• AI may tighten structure; it may not invent user counts, benchmarks, or “why we are different” facts.
• If you cannot teach the idea without the product, you do not have educational marketing — you have a brochure.

Table of Contents:

• What “Emails That Teach” Means
• A Repeatable Issue Structure
• Tooling: Curriculum Engine vs ESP vs CMS
• AI Without Lying to Your List
• A 5-Issue Launch Series (Practice vs Production)
• Metrics and Ethics
• Implementation Notes
• Worked Example: Teaching Idempotency Without a Brochure
• Frequently Asked Questions
• Conclusion

What “Emails That Teach” Means

Short answer: The reader could forward the body to a colleague who will never buy, and it would still be worth their time.

Examples that qualify:

• How to key an idempotent webhook handler, with a failure story. Then: “We productized this queue pattern in X.”
• How RAG citations should be rendered from IDs, not from model prose. Then: a link to your open-source engine.
• How to redact a Gmail thread before a copilot sees it. Then: your privacy feature.

Examples that do not:

• “5 reasons teams choose us” with no procedure.
• A model-written “ultimate guide” that restates public docs with your logo.
• Fake scarcity on a waitlist.

The pedagogical bar matches HinterBuild’s engineering blog: first sentence answers, tables for choices, FAQ for leftover queries. See AI for writing technical blog posts and write better emails with AI.

If the lesson is for you first, you are in learning-system territory: Notion AI vs dedicated learning tools. Marketing starts when someone else opted in.

A Repeatable Issue Structure

Short answer: Promise, teach, prove, offer.

Subject lines: teachable. “Idempotency keys go with the response” beats “Big news from our startup.”

Body: no carousels of features. If you need a comparison, make it a real one — including products you are not. That is how this post treats Cadensend versus ConvertKit.

Edit pass: Grammarly vs Claude for editing. Claude will turn your modest metric into a category-defining claim. Reject that.

Tooling: Curriculum Engine vs ESP vs CMS

Short answer: Pick the tool for the audience contract, not for “email + AI” branding.

Cadensend positioning, quoted from the product itself: not a marketing suite, not a CRM, not an autonomous agent, not a bulk sender. Plan a series from a learning goal, ground issues in your sources, structured citations, deliver approved issues on a timezone-aware schedule with idempotent sends. Products page. Contact if you want help self-hosting or wiring delivery — that is backend API engineering, including the same ideas as idempotency in distributed systems.

Practice workflow: Draft the five issues as a Cadensend series to your own address. Read them like a stranger. Fix citations. Then copy the approved text into ConvertKit (or whatever ESP you use) for the real list.

Illegal workflow: Export a random CSV and fire Cadensend. The product will not (and should not) pretend to be that. Consent, unsubscribe, and suppression are later, gated on those controls being real.

Ghost remains a fine place to host the public canonical of the same lesson. Still not Cadensend.

AI Without Lying to Your List

Short answer: The list opted into you. A hallucinated benchmark is a trust incident.

Rules:

1. Claims table before draft — same as technical blogging.
2. No invented customer logos.
3. No “used by 10,000 teams” unless finance agrees.
4. Sources opened this week: Perplexity research without copying.
5. Reading pile processed as extracts, not as fake mastery: batch process a reading list with AI.
6. If the model cannot find a source, the sentence dies. LLM hallucination causes and fixes.

Cadensend’s writer is a bounded agent: retrieve, write, critique, gate, revise with a hard limit. Retrieved text is data, not instruction (prompt injection). Copy that attitude into whatever ESP editor you use. Do not paste a “write viral launch email” prompt and hit send.

Privacy: waitlists contain emails and sometimes employers. Do not dump the CSV into a consumer chatbot to “segment personas.” PII.

A 5-Issue Launch Series (Practice vs Production)

Issue 1 — The painful default. Teach the failure (double charge, silent data leak, unread 80-page Notion). No product yet.

Issue 2 — The mechanism. One design idea (idempotency key stored with response; citation IDs rendered deterministically). Still no pricing table.

Issue 3 — The constraint. What you will not build (Cadensend will not be a CRM; your side project will not be Slack). Teaching via scope.

Issue 4 — The walkthrough. Reader can reproduce a minimal version without you. If they cannot, you are not teaching.

Issue 5 — The offer. Now the CTA. Same voice. Link to docs or GitHub. For Cadensend, the honest CTA is clone the repo, not “sign up for hosted.”

Run 1–5 to yourself with Cadensend. Calendar the reads: calendar AI scheduling for deep work. Interview-style founders can also drill the spoken version of issue 2: AI for interview prep — investors ask the same questions.

When you have a public essay that should live forever, promote issue 2 to the blog. When you have a personal list that is not a launch, personal newsletter with AI assistance.

Metrics and Ethics

Count:

• Replies that ask a technical question (signal you taught).
• Unsubscribes after issue 1 (your targeting was wrong, or you pitched too soon).
• Click on the doc, not only the pricing page.

Do not count:

• Open rate as proof of teaching.
• Model-suggested “urgency.”

CAN-SPAM / GDPR / CASL: HinterBuild is not your lawyer. You still need a lawful basis. Educational content does not exempt you from consent. About our team if you want engineering help, not legal advice.

Implementation Notes

• Idempotent sends on the ESP side too. Retries double-mail launches. Design as in idempotency in distributed systems.
• Timezone. A 07:00 lesson in the founder’s zone is 23:00 for the reader. Cadensend stores UTC plus IANA zone for its schedules; your ESP has its own rules. Do not mix them blindly.
• Versioning. If you fix a wrong claim, that is a new issue version, not a silent rewrite of history in someone’s inbox.
• Gmail. Do not compose the blast in Gemini with the full list in the prompt. AI in Gmail without leaking data.

HinterBuild will help you self-host curriculum plumbing or ESP-adjacent APIs. We will not help you scrape inboxes. Contact.

Worked Example: Teaching Idempotency Without a Brochure

Short answer: Five issues, one mechanism, CTA only when the reader could implement a toy version without you.

Issue 1 body (practice, to yourself via Cadensend). Teach: networks retry; SQS is at-least-once; a timeout is not a negative. Proof: your 12-row audit table, anonymized. No product name in the subject. Sources ingested: AWS docs, your redacted RFC. Citations rendered from chunk IDs — the reason you did not prompt Claude to “add footnotes.” LLM hallucination causes and fixes.

Issue 2. Teach: store the key with the response. Table: Stripe-style header vs your DB unique key. Proof: the crash-window test. This is the issue you might later promote to AI for writing technical blog posts. Link the long canonical idempotency in distributed systems only if you already published it; do not cite a URL you have not shipped.

Issue 3. Scope. “We will not build a CRM. We will not browse the web for you. We will not send this to a list until consent exists.” If you are HinterBuild, this is also Cadensend’s positioning: not ConvertKit, not Ghost. Readers learn how to evaluate tools. Products.

Issue 4. Walkthrough they can run: a 40-line handler, a test that retries. If they need your SaaS to complete the tutorial, you failed the teach test. Cadensend’s honest walkthrough is clone the repo, not “sign up.”

Issue 5. Offer. One link. For Cadensend: GitHub. For a hosted side project: the waitlist ESP. You copied the approved text from the private series into ConvertKit. You did not blast from Cadensend. Cadensend.

AI in the middle. Outline critique on issue 2: allowed. “Add a statistic that startups lose millions”: banned. Grammarly on the ESP paste: allowed, with the vendor data question. Grammarly vs Claude for editing. Subject lines: write better emails with AI. Research: Perplexity research without copying, batch process a reading list with AI. Notes: Notion AI vs dedicated learning tools. Holds: calendar AI scheduling for deep work. Inbox compose: AI in Gmail without leaking data. Job-search overlap: AI for interview prep. Ongoing list that is not a launch: personal newsletter with AI assistance.

Delivery engineering. ESP retries must not double the launch. Idempotency. If you need that designed, backend API engineering. Contact. About.

Frequently Asked Questions

Can Cadensend send my side-project launch to 5,000 people?

No. The current build sends to a single verified recipient. Opt-in audiences are a later update, gated on consent and suppression. Use an ESP for lists. Cadensend is not ConvertKit.

Is Cadensend a Ghost alternative for newsletters?

No. Ghost is a publishing CMS (hosted or self-hosted). Cadensend is an email curriculum engine that plans and grounds lessons. Different job.

How do I use AI to write teaching emails without sounding like everyone else?

Write the proof block from a real constraint you hit. Ban the model from adding metrics. Keep one idiosyncratic example per issue.

Should the CTA be in every email?

It can be a quiet text link. It should not interrupt the procedure. Issue 5 is allowed to be louder because you already paid in teaching.

What if my product is not educational?

Then do not fake a curriculum. Send a short, honest pitch. Teaching-shaped marketing for a non-teaching product is a category error — the same error as calling Cadensend a CRM.

Can I practice the series in Notion?

You can draft. You will not get coverage validation, citation structure, or exact-once delivery. Practice in Cadensend to yourself, or accept that Notion is a doc: Notion AI vs dedicated learning tools.

How long should a teaching issue be?

Long enough to complete one procedure, short enough to read in a single hold — often 400–900 words of body, not a 3,000-word canonical. If you need the canonical, link it. HinterBuild’s engineering posts are a different genre; the email is the sample, the blog is the archive. A model asked to “make it comprehensive” will destroy the teach-one-idea rule.

Do I need a visual in every send?

No. A table that compares two designs beats a generated hero image. If you use a diagram, you must be able to explain every box. Cadensend’s own path is diagrams-as-data (Mermaid/D2) rendered server-side — still not an excuse to decorate a brochure.

Conclusion

• Teaching emails complete an idea before they sell.
• Use Cadensend to compile and receive a sourced series yourself — MIT, self-hosted, not an ESP.
• Use ConvertKit-class tools for opt-in lists; use Ghost-class tools for web publishing.
• AI edits structure; humans own facts, consent, and the CTA.
• Measure technical replies, not theatrical open rates.

Keep going with write better emails with AI, personal newsletter with AI assistance, and LLM hallucination causes and fixes. Products, contact, Abdul Sami on LinkedIn.