Task & context design
Define the objective, audience, approved facts, source boundaries, tone, output format, and missing-information behaviour.
Prompt engineering & AI workflows
Better instructions. More useful AI work.
Turn repeated marketing tasks into clear, testable workflows your team can use with confidence.
In plain English
Prompt engineering is the design of instructions, context, examples, and checks for an AI task. Useful workflows also define approved sources, expected output, evaluation criteria, and when a human should review or decide.
Who this is for
Marketing teams, founders, agencies, and service businesses repeating content, research, outreach, or reporting tasks with AI assistants.
What we can work on
Reusable prompt library
Build practical templates for briefs, content reviews, outreach drafts, customer answers, and reporting plans.
Evaluation & improvement
Create representative test cases and a rubric for factuality, brand fit, source fidelity, format, and task usefulness.
Workflow handoff
Document where human approval belongs, how versions are maintained, and what the workflow should do when it cannot answer reliably.
How the work is scoped
- Discover: review the business context, available evidence, and the problem to solve.
- Prioritise: agree on deliverables, dependencies, budget, and a review period.
- Deliver: implement the agreed work and document the decisions.
- Review: assess the outputs and performance, then adjust the next steps.
Agree the boundaries first.
Includes task-specific prompt templates, representative evaluation cases, a review rubric, and documented handoff. Custom software, account integrations, private-data connectors, autonomous publishing, and model training are excluded unless separately agreed. The focus is practical SEO, content, outreach, and reporting tasks with explicit human review points.
What your brief should include
- A repeated marketing task and examples of current work.
- Approved source material, audience context, and writing guidance.
- Chosen assistant or model and permitted information boundaries.
- Acceptance criteria, representative difficult cases, and review owners.
How this can look in practice
These scenarios explain the method. They are hypothetical examples, not client case studies or performance claims.
Illustrative scenario
repeatable content briefs
- The situation
- A hypothetical marketing team repeatedly asks an assistant for briefs but receives unsupported claims, inconsistent structure, and unclear source requirements.
- Possible work
- Define the brief format, required inputs, approved facts, audience, and missing-information behaviour. Add examples and a rubric that checks source fidelity, usefulness, and completeness.
- What to review
- Compare drafts on the same representative tasks and record corrections. This illustrates evaluation design, not a claim of measured time savings.
Illustrative scenario
more relevant outreach drafts
- The situation
- A hypothetical outreach specialist receives generic draft emails that invent familiarity or fail to explain why the recipient would care.
- Possible work
- Require verified recipient context, a specific contribution, concise wording, and an explicit prohibition on invented relationships or evidence. Separate draft preparation from approval and sending.
- What to review
- Test missing-context and poor-fit cases as well as suitable opportunities. An acceptable workflow flags gaps; this example reports no campaign outcomes.
How delivery is checked
- Templates distinguish supplied facts, assumptions, unknowns, and information the assistant must not invent.
- Evaluation includes difficult or incomplete inputs and checks failures as well as usable outputs.
- Handoff records the tested model context, prompt version, reviewer role, and circumstances requiring retesting.
Adapting the work to your market
For teams serving the USA, UK, Australia, and New Zealand, make audience country, spelling, currency, terminology, and service availability explicit inputs. Test each relevant market rather than assuming one English-language example covers all four. Regional knowledge and important factual claims still require a qualified human reviewer with suitable source material.
See how international projects are scoped and coordinated.
What progress looks like
Compare outputs on a consistent test set. Review acceptance rate, factual errors, editorial effort, and task completion against your baseline. Results depend on the model, context, data, and task.
Common questions
Is the Prompt Studio a live AI chatbot?
No. It assembles a structured brief locally in your browser. You review and copy that brief into your preferred AI assistant; no model API is connected to the builder.
Can you work with ChatGPT or Claude?
Yes, task design and evaluation can be tailored to how your team uses those assistants. Account integrations, private data connections, or custom software require a separately agreed scope.
Will prompts remove the need for review?
No. Important claims, external communications, customer advice, and publication decisions still need appropriate human review.
Connected services
A clear place to start
Let’s find your next opportunity.
Share your website, audience, and priorities. We’ll put the useful next steps in focus.