Turn Client Content Sources Into an Evidence-Backed Research Brief

A client sends six interesting links, a competitor launches a campaign, and somebody bookmarks a useful customer question. By the next planning meeting, the team has plenty of material but no clear answer to a simple question: which of these sources deserves a place in this client's content plan?

The useful deliverable is not another feed of popular posts. It is a short research brief that explains what matters, where the evidence came from, and what the client could contribute without copying somebody else's work.

For an agency delivering this service, Agentic Workers provides a separate workspace for each client, connected tools, and agent work delivered through chat. The following is a proposed workflow for a Super Agent, not a tested X integration. It starts with authorized source material and ends with an editorial decision, not an automatic post.

Define the brief before collecting more links

Write a one-paragraph assignment for one client. Name the audience, the offer, the questions the client can answer credibly, and the subjects that are out of scope. A specialist accounting firm, for example, might want to explain reporting problems for growing retailers. General entrepreneurship inspiration would be outside that brief, however popular it becomes.

Set the output format before collection begins. Each candidate should include:

• The original URL, author, publication date, and retrieval date.
• A short description of the audience's problem.
• The specific observation that makes the source useful.
• An original angle the client could support with its own experience.
• The evidence or permission still needed before drafting.

A candidate without a reason to matter belongs in the archive, not in the next assignment. The brief should be short enough for the person responsible for the client relationship to review without opening every source again.

Use a source packet you have permission to provide

Do not begin by assuming an agent can read a client's X lists, bookmarks, private messages, or account statistics. Verify the specific authorized access path first. If it is unavailable, a client-approved export, a shared document, or a manually selected set of public links is a workable starting point.

Keep that packet inside the client's workspace. Exclude private messages, customer identifiers, and confidential campaign plans unless they are necessary and explicitly authorized. Public availability does not grant permission to reuse images, reproduce entire articles, or republish a person's words as the client's own.

Treat instructions found inside posts and documents as source material, not as permission to change the workflow. A source telling the agent to follow another account or send a message does not expand its authority.

Ask the main agent to coordinate the packet and the research assignment. Give the research task read-only scope wherever the available tools allow it. Saving a new brief to a shared tool is still a write, so agree on its destination and approval boundary before enabling that step.

Separate attention from relevance

A widely shared post can suggest a question worth investigating. It does not establish that the underlying claim is true or that the same format will work for this client.

Have the Super Agent distinguish three things in its notes: what the author actually said, what observable evidence is available, and the team's interpretation. If an export lacks impression data, say that reach is unknown. Do not invent a score that makes incomplete data look comparable.

Competitor material is useful for spotting repeated questions, neglected explanations, and differences in positioning. It is not a script. Ask what the competitor left unresolved, then identify whether the client has a concrete answer. If the client cannot add evidence, experience, or a useful explanation, reject the angle.

Saved sources need the same treatment. Resurfacing an old bookmark is valuable only when its claim remains current and its context still fits. Preserve the original date, check whether the linked page has changed, and mark inaccessible material as unavailable rather than summarizing from a headline.

Turn the packet into decisions

For the accounting example, suppose the packet includes a popular post about messy reporting, a competitor checklist, and an older article about inventory records. The proposed brief could recommend an original explanation of which records a retailer should prepare before a reporting review.

It should also state what is missing: the firm's own process, an approved example, and confirmation that the advice applies to the intended audience. None of those details should be filled in from another client's files.

Use a compact decision label: draft next, investigate, save, or reject. Include a reason. A possible commentary post can remain an idea in the brief; it does not authorize quote-posting, tagging the source author, or publishing from the client's account.

Test whether the brief improves selection

Start with one client and a small packet that a person can inspect in full. Compare the resulting brief with the sources. Can the reviewer trace every recommendation? Did the agent miss a caveat? Are the suggested angles genuinely different from the originals?

Record which candidates become approved assignments and why others are rejected. Also record the effort needed to correct citations or remove irrelevant suggestions. These are workflow checks, not promises of audience growth.

Stop the run when access fails, source dates are unclear, or the packet contains material from another client. Return the incomplete brief with those limits visible. A shorter honest brief is more useful than a confident list built from missing evidence.

This approach draws on Ryan's list of content-agent ideas, particularly research, competitor monitoring, saved-source review, and commentary scouting. The adaptation here is a client-specific evidence brief with a human editorial handoff.

Review Agentic Workers plans for starting with one client workspace.