Build the brief before asking for content
Write down the business, the customer, the service, the geographic scope and the decision the page should support. List the information that is confirmed and the information still missing. Separate the desired outcome from the deliverable: visibility is an outcome; a page plan, draft or implementation patch is a deliverable.
Make responsibilities explicit. ChatGPT can propose work and use the tools actually available in the session. A claim that it published a page, inspected a website or checked a live result needs corresponding access and evidence. Ask it to identify that boundary before starting.
For the agency-level overview, read Can ChatGPT do SEO? on the Click2Flow website. The exercise below concentrates on writing a brief that a trainee can execute and review.
A worked public teaching prompt
Fictional exercise: Example Home Services is a made-up business offering residential gutter cleaning in Randburg. The exercise supplies that service and area only. It does not establish prices, availability, reviews, qualifications or a real contact destination.
Act as an SEO planning assistant for this fictional training exercise.
Objective: plan a useful gutter-cleaning service page for homeowners in Randburg. Work towards relevant Google visibility and AI discoverability; do not claim achieved rankings, citations or recommendations.
Confirmed facts: business name = Example Home Services; service = residential gutter cleaning; area = Randburg. This is a fictional business.
Unknown facts: prices, availability, credentials, reviews and contact details. Do not invent them. Put information gaps in a separate list.
Task: produce a page-purpose statement, an intent-based outline, a short sample introduction and a proposed internal-link plan. Apply Authority, Semantics, Architecture and Performance to explain your decisions.
Acceptance checks: every business claim must trace to a supplied fact; each section must answer a customer question; proposed links must be labelled as proposals; no publication or results may be claimed.
Handoff: return the draft, assumptions, open questions, checks performed and next reviewer. Stop for review before any live change.The prompt names a role, defines the input boundary and requests a reviewable output. The target is clear without turning an ambition into an unsupported result. Replace the fictional facts only with information approved for the real client.
Use a sequence of smaller task prompts
After the initial plan is accepted, ask for one bounded artifact at a time. For research, request a question list with supporting sources and a distinction between observation and inference. For content, request the selected section using the approved fact register. For implementation, identify the exact page or file, the permitted edit and the checks required before release.
Choose tools by the task. A research task needs current source access. A coding task needs the relevant source files or a staging environment. A writing task needs the approved brief and editorial criteria. More reasoning time can help a difficult analysis, but it cannot supply missing facts or credentials.
When using several agents or conversations, keep one approved master brief and one change log. Give each worker a bounded responsibility and require a handoff with inputs, sources, output, checks and unresolved issues. A reviewing agent should check the actual artifact instead of relying only on the previous agent’s claim that it is correct.
Review the result against the original objective
Read the draft as a customer first. Can the person understand the service, its scope and the next step? Then inspect the evidence. Highlight statements that imply a price, outcome, qualification or availability that the brief did not establish.
Ask ChatGPT to identify weaknesses, but retain an independent human release decision. Compare the text with the source facts, test the intended user journey and confirm that any code change affects the correct environment. If there is no tool access to perform a check, the handoff should say so.
For measurement, give the query set, market, devices or platforms, date window and baseline. Record Google visibility separately from AI mentions, citations, recommendations and enquiries. The brief can keep those goals in view throughout delivery; it cannot control how a search engine or assistant will respond.
Practise a correction, not just a generation
Deliberately add one unsupported price and one invented testimonial to a copy of the fictional draft. Ask the trainee to find both, explain the risk and repair the prompt so the same errors are less likely to recur. Evaluate the correction against the supplied facts.
This public example introduces the briefing method. Click2Flow’s staff programme extends it into website planning, content, schema, implementation, assessment and independent practical review. Use the competency map to see how prompting fits the wider engineering role.
Two practice checks
Consider each question before opening the explanation. These public checks are practice; graded staff assessments are inside the private portal.
The brief requests Page 1 visibility. May ChatGPT state that the new draft already ranks?
No. A target does not establish a result. The response should identify the artifact produced and leave ranking claims to dated observations.
A coding agent says “implemented” but had no website or repository access. What should the handoff say?
It should describe a proposed implementation or supplied code, identify the access limitation and name the remaining implementation and verification steps.
Sources & further reading
Use primary documentation to verify technical details. Click2Flow’s framework reference describes the agency’s own method.
Build understanding into a repeatable skill.
Explore another public guide, or sign in to the assessed staff programme if your employee account and browser are approved.
The Training Centre is part of Click2Flow, the AI SEO engineering agency.