How to Write a Useful “How I Make This” Statement on Substack

Does a Substack AI scan prove who wrote a post?

No. Substack describes its scan as an estimate of how much text appears human-written or AI-assisted. An estimate can prompt a useful question about process, but it is not an authorship verdict and does not establish who originated the idea, checked the evidence, made the editorial decisions, or accepts responsibility for the result.

Substack says the feature applies to posts and Notes published on or after July 21, 2026. When readers run a scan, creators may add a “How I make this” statement that appears with that scan. Creators can also disable scanning for an individual post or Note; readers then see that the scan is unavailable.

The current availability is narrower than every Substack surface. Substack says scans are unavailable for video or audio posts, standalone Substack and custom-domain web views, and emails. The August 9 Help page lists scanning on iOS and says Android support is coming soon.

If a result looks wrong, Substack provides a Report detection error path. Its July 24 discussion also acknowledges false flags and the limits of what a label can capture. Reporting a result, explaining the process, and disabling a scan are different choices; a process note should describe the actual work, not be tuned to force a desired detector label.

The practical response is not to write a defense against a percentage. Write a concise process statement that gives readers information the estimate cannot: where the work came from and how it became publishable.

What should a “How I make this” statement explain?

A useful process statement is a compact provenance note: it shows where the idea began, which evidence shaped it, where AI assisted, what the creator decided, and who owns the final claims.

That definition is deliberately more specific than “written with AI” or “100% human.” Tool labels compress the work into a category. Readers usually need the consequential details: whether the insight came from the creator, whether sources were read, whether AI supplied or merely organized language, which suggestions were rejected, and who will correct an error.

Keep the statement proportional. A reported investigation may need source and verification detail. A short personal Note may need only the origin, the assistance used, and final ownership. The goal is not to inventory every keystroke; it is to make the decisions that affect meaning legible.

How do you write a credible creator process statement?

Use the Creator Process Note in order: Originate → Source → Assist → Decide → Own. The sequence keeps the creator’s contribution visible without pretending that one tool-use label can describe the whole article.

Originate names the starting material. Source identifies evidence and its role. Assist limits the AI contribution to work the creator can describe honestly. Decide records the judgment that changed the draft. Own closes with the person responsible for the published claims and later corrections.

This process is separate from the broader AI disclosure workflow for creators, which helps classify generated or altered material and place an initial disclosure. It is also distinct from AI disclosure carryover, which governs which representation and provenance facts survive when an approved asset becomes a derivative. A process note explains how this piece was made.

• Originate: “This began with my field notes and a recurring reader question.”
• Source: “I checked the recommendations against the linked primary guidance.”
• Assist: “I used AI to group notes and propose an outline, not to supply the reported facts.”
• Decide: “I rewrote the recommendations, removed an unsupported generalization, and chose the final examples.”
• Own: “I reviewed the final text and am responsible for its claims and corrections.”

Which details add useful transparency—and which overclaim?

Specific process details help because they connect a visible signal to the decision it can support. They become misleading when the statement turns one detail into a sweeping proof of authorship, accuracy, or quality.

What does a realistic “How I make this” workflow look like?

Consider a hypothetical independent gardening creator preparing a newsletter about keeping balcony herbs alive during a heat wave. The article begins with the creator’s planting notes, photographs, and a repeated reader question. The creator reads current horticultural guidance, then separates the source recommendations from observations that apply only to the creator’s own balcony.

AI groups the notes into possible sections and flags repeated passages. It also proposes a broad claim about one watering routine working for every balcony. The creator rejects that claim, rewrites the advice around sunlight, container, and local-condition differences, checks the linked guidance again, and chooses the final photographs and wording.

A concise statement could read: “This guide began with my balcony notes, photographs, and reader questions. I checked the practical recommendations against the linked horticultural guidance. I used AI to organize the notes and identify repetition; I rejected unsupported generalizations, rewrote the advice, selected the examples, and reviewed the final text. I’m responsible for the published claims and corrections.”

The statement does not announce a detector score or claim perfect accuracy. It gives the reader an inspectable account of origin, evidence, assistance, judgment, and accountability. A reusable AI voice packet can preserve the creator’s recurring language and boundaries, while the process note stays specific to this article.

How is AI detection different from blocking AI training?

Substack’s scanning, process-statement, and model-training controls answer different questions. Scanning estimates the composition of published text. “How I make this” gives the creator a place to explain the work when a reader scans. The separate Block AI training setting sends a request about whether a publication may be used for model training.

Substack says that the Block AI training request is respected only by tools that choose to honor it, and enabling it may affect discoverability. That preference does not describe how an article was authored. Likewise, a process statement does not set a crawler preference.

Substack also says neither Substack nor its detection provider Pangram uses publisher content to train generative AI models. Keep that platform statement scoped to those companies rather than treating it as a promise about every outside tool.

When should creators add or revise the statement?

Add the statement when process detail would materially help a reader understand the work. Revise it when the evidence changes, AI performs a different job, an editor makes a consequential intervention, or a correction changes what the creator can responsibly claim about the final piece.

Before publishing, run the Creator Claim Check on consequential sourced claims. Then compare the process statement with the actual draft history. Remove vague assurances, name the assistance that mattered, and make final responsibility unmistakable.

Substack’s estimate can start the conversation. The stronger trust signal is a process readers can understand and a creator who remains accountable after publication.

• Name a real creator-owned origin rather than a generic topic.
• Identify the evidence or source type that shaped the article.
• Describe AI work as a bounded task, not a blanket label.
• Include at least one human decision that materially changed the draft.
• State who reviewed the final piece and owns corrections.