Marketing claims review¶
Domain: Marketing and advertising compliance
For: Marketing, legal, and compliance teams
Checks advertising claims against substantiation, disclosure, and release controls.
Source converted¶
This example combines official method sources with a synthetic company policy and a synthetic case. Raw official documents are not bundled.
| Type | Source | Publisher / status |
|---|---|---|
| Official source | Advertising and Marketing on the Internet: Rules of the Road | U.S. Federal Trade Commission |
| Official source | .com Disclosures: How to Make Effective Disclosures in Digital Advertising | U.S. Federal Trade Commission |
| Official source | Guides Concerning the Use of Endorsements and Testimonials in Advertising (16 CFR Part 255) | Electronic Code of Federal Regulations |
| Synthetic policy and case | Published in the demo directory under the repository MIT licence |
— |
Skill generated¶
Instead of compressing the source into one summary, the skill separates reusable knowledge into six files:
SKILL.md: when to use the skill and the workflow;public-method.md: an independent method summary from official sources;company-policy.md: synthetic company rules with stable identifiers;evidence-map.md: which claims may come from which source;output-schema.md: the expected answer structure;scenario-guide.md: missing information, conflicts, and abstention behavior.
The locked evaluation expects decision class approve-with-edits, option
evidence-bounded-campaign, and 9 rule identifiers.
Final human route: Legal · Compliance.
LLM only vs LLM + skill¶
The same case and prompt were run twice. The only difference was that the skill was installed for the second run. A deterministic script using the locked answer key, not another model, scored both answers.
| Check | LLM only | LLM + skill |
|---|---|---|
| Policy rules cited | 0 / 9 | 9 / 9 |
| Exact decision class | no | yes |
| Named option | yes | yes |
| Human route | yes | yes |
| Trace score | 40 / 100 | 100 / 100 |
Control answer · Skill answer · Scorecard
This is a one-scenario, one-host comparison; it is not proof of production accuracy or regulatory compliance.
Copilot packages¶
- Cowork
.skill - GitHub Copilot for VS Code
- Microsoft Scout
- Copilot Studio GitHub harness
- Copilot Studio classic setup
The classic package is not a direct solution import. It contains setup material for a human to apply in the target Copilot Studio environment.
Reuse this example¶
The source manifest, synthetic inputs, scenarios, raw answers, and scorecard are
public under demos/marketing-claims-review/. Copy the structure for your own source, but publish
only material you have the right to share. Review Safety & reuse.