A prompt library becomes useful when the output changes your next decision. A positioning statement should guide the message. A segment definition should select an appropriate audience. A campaign review should explain what to keep, change, or investigate.
The AI Marketing Working Playbook contains 40 original prompts organized around four jobs: Understand, Shape, Reach, and Learn. They help a small team move from evidence to approved work without treating generated content as validated strategy.
HubSpot's public Loop Marketing resource was a reference for the broad idea of a staged prompt library. Our prompts, names, worksheets, and examples are independently authored. We do not reproduce its gated collection, claim its results, or imply affiliation. [1]
The four jobs and what each produces
| Job | Question | Useful output |
|---|---|---|
| Understand | What do we know about the customer and alternatives? | Evidence brief, customer questions, positioning hypotheses |
| Shape | What should this audience receive? | Segment rule, message, offer explanation, journey |
| Reach | Where and how should people discover it? | Distribution plan, search brief, partner proposal |
| Learn | What happened, and what should change? | Experiment review, measurement checks, next decision |
This is an editorial workflow, not a proprietary predictive model. A team can use it in existing documents, a CRM, and its current AI assistant.
Begin with a context brief
Before choosing a prompt, write the business goal, offer, target situation, approved facts, available evidence, constraints, and decision you need to make. Add your brand voice and examples of language you want to avoid.
Provide only appropriate, minimized data. Remove personal identifiers from customer notes where they are unnecessary, and keep private customer information out of public examples. Record the source and date of evidence so a convincing answer can be traced back to what actually happened.
If the assistant cannot access a file or current source, it should ask for it or mark it missing. Do not let it describe an imagined competitor review or analytics scan as completed research.
A worked sprint: improve welcome-sequence inquiries
Imagine a small consultancy receives interest in email marketing but few suitable requests for onboarding help. The goal is to learn whether a clearer explanation attracts better-fit inquiries. This is a hypothetical campaign, not a MarketingCoach case study.
First, understand. Use prompts MC-U01 and MC-U04 to organize actual customer language and compare only the competitor material you supply. Ask for gaps between what customers need and what the offer page explains. A missing claim on a competitor page does not prove the competitor lacks that capability.
Then, shape. Use MC-S01 to define a segment such as subscribed founders with an active product and a stated onboarding problem. Use MC-S02 to check the logic. Use MC-S04 to draft an explanation that emphasizes the first customer result instead of the number of emails delivered.
Next, reach. Use MC-R01 to choose a manageable distribution opportunity. Create a useful demonstration, then adapt it for the selected channel with MC-R05. If the audience already opted into email, provide a relevant next step there too. A plan is not a scheduled send; confirm the real destination and authorization before any external action.
Finally, learn. Use MC-L01 to define the experiment before launch. Use MC-L02 to inspect the actual results and MC-L10 to record the decision. If visits increase but suitable inquiries do not, investigate fit and the page before declaring that the channel failed.
Use fewer prompts per session
Do not run all 40 prompts to produce a mountain of deliverables. Choose the next missing decision. If you lack customer evidence, a polished brand guide is premature. If the offer is already clear but onboarding is broken, begin with the journey or activation prompts.
The playbook gives every prompt an input checklist, a requested output, and a review check. Save the reviewed artifact and reuse it as context for the next task. Do not repeatedly regenerate foundational decisions without a reason.
Build a usable brand guide
Separate factual positioning from creative taste. Positioning identifies whom you help, the problem, the approach, and credible proof. Voice guidance explains how that story sounds. A useful guide contains approved examples, prohibited claims, vocabulary choices, and boundaries.
An audience preference brief can help generate creative options, but it does not predict with certainty what a buyer will do. Treat predicted responses as hypotheses and test actual reactions before increasing spend.
Make personalization specific and explainable
Use the segmentation guide to separate membership, treatment, automation, and measurement. Choose a difference that changes the assistance you offer: a beginner needs an example, while an experienced operator may need a comparison or implementation detail.
Avoid personalizing from sensitive inferred traits or unsupported intent scores. Give unknown records a useful default. A small set of explainable segments is easier to maintain than dozens of unclear labels.
Research and distribution still require real work
AI can help draft an interview guide, organize review themes, compare supplied alternatives, inspect a journey, or plan a search-topic cluster. Those outputs are research preparation or synthesis. Interviews, observations, analytics collection, and source verification still need to occur.
For answer-engine visibility, assess useful coverage, clarity, access, and source credibility. Google says established SEO foundations apply to its AI features; a prompt cannot guarantee an AI citation. [2] Record the query, engine, date, and answer when making visibility observations.
For partnerships, evaluate audience fit and a useful joint contribution. Do not treat a generated creator list as verified reach, pricing, or permission to contact someone.
Measure before claiming improvement
Choose a primary outcome and a guardrail. For a lead campaign, the primary outcome might be qualified inquiries; a guardrail could be unsubscribe rate or unsuitable-lead workload. Define denominators, time windows, and source systems.
Distinguish attribution from incrementality. An attribution model assigns credit using a rule; it does not automatically prove the campaign caused the sale. Compare like with like, consider conversion delay, and keep small-sample uncertainty visible.
Do not claim tenfold output, lower acquisition costs, or real-time optimization unless your actual workflow and evidence support those claims. More assets can increase review work. Faster production is useful when quality and customer outcomes hold up.
A first working session
Spend the first ten minutes writing the decision and gathering inputs. Use one prompt to produce the draft. Review evidence, omissions, and fit. Revise the output yourself, then record the action, owner, success measure, and checkpoint.
Download the AI Marketing Working Playbook, start with one prompt, and leave the session with one decision you can act on.
Sources
[1] HubSpot's public prompt-library page, reviewed as a format and framework reference, not as a source of copied prompts. [2] Google Search Central: AI features and your website. Checked September 27, 2026.