Core guide / PULSE
A source-to-publication workflow for AI-assisted content
Turn a source packet into a reviewed article and visual without losing evidence, ownership or the approval decision.
Source review: . Prepared with AI assistance; references checked against official documentation. Editorial method and limits.
Download a blank evaluation worksheet (CSV). The example row is fictional and has not been run.
Download a worked claim-review example — fictional brief, flawed draft, evidence decisions and corrected copy.
Create the source packet
Assign a brief ID and record the audience question, original URLs, access date, source dates and uncertainty. Keep source observations separate from your interpretation. A product announcement supports what a vendor announced; it does not establish independent performance.
For a fictional launch brief, use fields for claim, supporting passage, limitation and reviewer. If a claim cannot be supported, remove it or label it as a question to investigate.
Draft within the evidence boundary
Give the drafting tool the packet and ask it to identify missing evidence before writing. Request a short answer to the audience question, followed by reasoning and references. Never let a polished paragraph substitute for opening the source.
Suggested instruction: “Use only this packet for factual product claims. Mark unsupported claims as unresolved. Separate vendor statements from our interpretation.” This is a proposed prompt, not a benchmarked one.
Run two different reviews
First review facts: names, dates, numbers, source support and disclosure. Then review usefulness: does the reader know what to do and when the advice does not apply? Assign a responsible human for the final decision.
Do not approve a visual independently of its caption. A correct article can become misleading when a chart label drops a qualifier or a graphic crops out the date.
Make the handoff explicit
Use a record with brief ID, revision, reviewer, decision and approved artifact URL. If using automation, require approval for the exact revision being sent onward. Changing a draft after approval should create a new review decision.
Start with a draft-only destination. Test declined approval, missing source and duplicate input before allowing a publishing action. The linked Zapier guide describes its approval feature; the proposed record design here is tool-independent.
Publish and maintain
Link to the original references and state how the article was produced. Date substantive reviews accurately. A new timestamp without a content check does not make an old claim fresh.
Create an update queue for source changes, broken references and reader corrections. Prioritize changes that could alter a reader’s decision. Preserve an explanation when a material recommendation changes.
Official references
References support documented product capabilities. Procedures, examples and acceptance criteria above are editorial proposals, not vendor promises or hands-on benchmarks.
Related tool decision guides
- Canva for a small content team: brand consistency before automation
- Zapier for AI content handoffs: approvals before publishing
- Notion AI for team research: sources, scope and permission checks