AI Foundations and Working with Wisdom
This domain covers the AI layer inside the platform and, more heavily, the judgement around it. Expect questions on where AI assists (drafting in Perch, summarising in Lumen, reply drafting in Nest), how to prompt it, and what a human must always check before anything ships.
Prompt anatomy
A usable prompt names the role and context, the task, the audience, the hard constraints (length, tone, banned claims), the output format, and one or two examples of approved copy. Adjectives like "engaging" carry no information; constraints do.
- Iterate on the draft with specific edits rather than regenerating from scratch - regeneration loses what already worked.
- Feed brand voice as examples, not as a description. The model imitates samples better than it interprets adjectives.
- Keep prompts reusable: the same constraint block across a campaign keeps output consistent.
Rules of judgement
- Human-in-the-loop before publishing. Every statistic, date, price and product claim is fact-checked against source.
- Never paste personal data, customer identifiers or unpublished financials into a third-party model.
- Disclose AI-generated imagery where the platform or regulation requires it.
- AI classification (sentiment, themes) is a first pass to be validated on a sample, not a verdict.
Worked example
Three LinkedIn captions for a sustainability consultancy launch. Prompt: role = B2B copywriter for a sustainability consultancy; task = 3 caption variants; audience = procurement leads; constraints = under 200 characters, no unverified carbon claims, one CTA, plain tone, no emoji; plus two approved past posts as voice samples. Then the edit pass: verify the carbon figure against the published report, cut a hype adjective, insert the campaign UTM. Reasoning: the claim constraint prevents the most expensive failure - a fabricated or overstated statistic - and the voice samples do more work than any tone adjective.
Traps
- Publishing the first output unedited.
- Asking for "viral" copy with no defined audience or constraint.
- Trusting an AI sentiment summary over your own verified sample.
- Accepting generated statistics or citations as sourced.
- Assuming the model has learned your brand guidelines from memory.