Source packs are the new briefs

Definition

What is a source pack?

A source pack is a compact, approved bundle of facts, links, claims, examples, constraints, and measurement notes that a human or agent can use to complete a repeatable marketing workflow. It is not a folder of random references. It is a governed working object.

In AEO/GEO and ContentOS work, the source pack answers the questions a brief usually hides: which facts are canonical, which claims are allowed, which claims are not ready, which pages should be cited, which external surfaces matter, which competitor sources explain the gap, and what proof will count as progress after publication.

That is why source packs are more important than prompts. A better prompt can make an agent sound smarter. A better source pack makes the work safer, easier to review, and more likely to improve the source graph that AI systems use.

Old brief

The classic brief was built for a human writer

Traditional marketing briefs assume a human writer can fill in the gaps. They list the topic, audience, keywords, tone, rough outline, and maybe a few links. That can work when the writer already knows the company, category, and constraints.

It breaks down when the work becomes agentic. An agent will happily connect weak sources, invent missing transitions, overstate claims, reuse outdated facts, or optimize for fluency instead of truth. The issue is not that the model is bad. The issue is that the work object is underspecified.

A human can read a vague brief and ask the founder, sales lead, or strategist for context. An agent needs that context packaged before the workflow starts, and a reviewer needs to see which context was used after the workflow ends.

New primitive

A source pack turns context into an operating asset

The source pack is not a prettier brief. It changes the job. Instead of asking "what should we write?", it asks "which source-backed claim should become easier for humans and AI systems to cite?"

That shift matters because AEO/GEO is not only content production. It is source-surface production. The page, case, note, service description, profile, or external adaptation should make a business claim more discoverable, more defensible, and easier to reuse in an answer.

In practice, a source pack is the object a team can approve before agent work starts. It prevents the agent from treating every public page, internal note, old deck, Slack comment, and competitor claim as equally valid.

Anatomy

What should be inside a source pack?

The exact shape depends on the workflow, but the useful pattern is stable: facts, claims, evidence, constraints, canonical rules, and proof. If a source pack lacks one of those, the reviewer will have to reconstruct it later from the agent transcript.

AEO/GEO

AEO/GEO needs source packs because AI systems cite sources, not intentions

AEO/GEO work often starts with the wrong question: "What article should we publish?" The stronger question is: "Which missing source would make the right answer easier to assemble?"

If an AI system does not mention a company, describes it incorrectly, or cites weaker competitor material, the fix is rarely one isolated article. The fix is usually a source gap: unclear owned pages, missing cases, weak profile consistency, absent third-party validation, thin service pages, or a lack of answer-first pages that connect the claim to evidence.

The source pack makes that gap explicit before the team creates content. It should say which answer is currently weak, which source surfaces explain the weakness, and which canonical asset should be produced first.

ContentOS

ContentOS should run on source packs, not topic lists

A topic list is a queue. A source pack is a control plane. The difference matters when a team starts using agents for research, writing, QA, distribution, and measurement.

In a ContentOS workflow, the source pack should move through the production corridor: strategy approves it, the content agent drafts from it, the editor checks claims against it, the website agent verifies crawl and schema implications, the publisher prepares adaptations, and the proof loop compares the result against the baseline.

Without that object, each step becomes a separate interpretation of the same vague idea. With it, the team has one inspectable truth boundary.

Marketing agents

A marketing agent should receive a source pack and return a result packet

This is the clean interface for ordinary teams. The human gives the agent an approved source pack. The agent returns a result packet: what it produced, which sources it used, which claims it avoided, what is uncertain, which checks passed, and what decision the human needs to make.

The source pack prevents bad autonomy at the beginning. The result packet prevents invisible work at the end. Together they create a workflow that can be audited, repeated, and improved.

This is also how a workspace accumulates memory without becoming a messy chat archive. Approved source packs and reviewed result packets become reusable assets. Prompt transcripts stay secondary.

Human gate

The source pack is where humans should spend judgment

Teams often waste human review at the wrong point. They wait until a draft exists, then argue about facts, claims, sources, tone, examples, and positioning all at once. That is expensive because the agent has already produced a polished object from incomplete instructions.

A better review sequence is to approve the source pack first. Is this the right business claim? Are these sources allowed? Are we missing evidence? Is the canonical destination right? Do we know what we will measure after publishing?

Once those questions are answered, agent work becomes less magical and less risky. The reviewer can focus on whether the output followed the source pack, not whether the agent guessed the business context correctly.

Example

A source pack for "promotion in neural networks"

For the Humanswith.ai funnel, a source pack for "promotion in neural networks" would not start with a headline. It would start with the source graph we want to make clearer.

FAQ

Questions this page should answer

Source trail

Internal sources and next surfaces

Read next

Continue the thread