AI becomes more useful for marketing when you give it a complete job: understand the offer, organise the evidence, prepare the campaign and help review what happened.

Asking for ten social posts can produce ten pieces of content. Starting with the customer, the offer and the action you want gives you a way to judge whether those posts serve the business.

This guide follows one campaign from brief to results. Northbank Decorating and every example below are fictional. The campaign targets are planning examples, not reported results.

1. Write a campaign brief before asking for content

Use a short document that answers six questions:

  1. Who is this for? Identify the customer and the situation that makes the offer relevant.
  2. What problem are we addressing? Use language you hear from customers.
  3. What are we offering? State the service and its important conditions.
  4. Why should someone believe us? List evidence you can actually show.
  5. What should they do next? Choose one main action.
  6. How will we judge the campaign? Define a useful business result and a review date.

For Northbank, the brief could be:

Brief element Fictional campaign decision
Audience Office managers arranging a meeting-room repaint
Customer problem Finding a suitable time around office use
Offer Discuss an evening repaint, subject to scope and availability
Evidence Approved service details and photographs the business has permission to use
Main action Request a site visit and quotation
Pilot target Six suitable quote requests over three weeks

“Suitable” needs a definition. Here it means the property is in the service area, the requested work fits the service and the customer is considering a real project. The owner must check the target against capacity before launching.

If you are new to using assistants, start with the practical guide to AI for business before building the campaign.

2. Research the questions that affect the message

Begin with information you already have: approved service details, anonymised enquiry notes, common questions and genuine customer feedback you are allowed to use.

Ask the assistant to group that material into customer problems, objections and unanswered questions. Require a reference to the source behind each finding. Separate what customers actually said from the assistant’s suggestions.

For Northbank, useful research questions include: Do enquirers ask about evening access? What information is needed before a quote? Which services and locations can the business confidently advertise?

External research can add context. If you use an assistant’s web-search capability, ask for sources and dates, then open the relevant pages. Check competitors’ own service pages for their stated offers. A competitor’s claim is evidence of its messaging; it does not prove that customers prefer it.

ChatGPT’s official prompting guidance describes supplying relevant documents, requesting web search for current information and asking for sources to check. Confirm which capabilities are available in your own tool and account before relying on them.

Keep a short evidence list. Mark unsupported ideas as hypotheses you could test. Do not turn a suggested customer concern into a quotation or a made-up market statistic.

3. Develop one clear message

Use the approved brief and research to ask for a few genuinely different message options. For each, request the customer problem, the main benefit and the evidence needed to support it.

Northbank might compare a message about working around office use with a message about refreshing a tired meeting room. The owner can choose the first because it matches the offer and the enquiries being addressed.

Keep the wording within the evidence. “Evening work can be discussed” does not support “zero disruption,” “overnight transformation” or guaranteed next-day availability.

A clear main message for this fictional campaign is: “Planning a meeting-room repaint? Ask about an evening appointment.” It gives the reader a recognisable situation and a practical next action.

4. Turn the message into connected drafts

Choose the channels you can manage. For this example, Northbank uses a service-page section and two LinkedIn posts linking to its enquiry page. Each asset should carry the same offer and conditions.

Here is an illustrative first post:

Does your meeting room need repainting, but finding a suitable time is difficult?

Northbank Decorating can discuss evening appointments for office repainting. We confirm the scope, access arrangements and availability before agreeing the work.

Tell us where the office is, what needs painting and your preferred timing. We will explain the next steps for a site visit and quotation.

The owner would replace the fictional business details with verified information and add the correct enquiry link. Any photograph also needs permission and an accurate description.

Use this prompt to prepare your own campaign drafts:

Prepare a marketing campaign draft from the approved material below.

Campaign brief: [audience, problem, offer, evidence, action,
business target and review date]
Sources: [approved service details and relevant research notes]
Voice: [how we speak, with a short approved example]
Channels: [where this campaign will appear]

First identify missing facts or conflicting instructions.
Then produce:
1. One main message and the supplied evidence supporting it.
2. A short service-page section and two distinct social posts.
3. A list of factual claims for me to check against the sources.
4. A simple measurement plan tied to the campaign target.

Keep offer details consistent across the drafts. Do not invent
testimonials, statistics, prices, results or availability. Label
assumptions and leave placeholders for facts we have not supplied.
Prepare drafts only; do not publish or send anything.

If the drafts sound generic, supply a real example of your preferred voice and explain what to change. For choosing an assistant, compare Claude and ChatGPT for marketing on this same task.

5. Review the whole customer journey

Read each asset as the customer would. Does the opening describe their situation? Is the offer understandable? Does the next step lead to a working page with the same promise?

Then check facts against the sources: services, locations, prices if used, dates, evidence and conditions. Remove claims you cannot support. Make sure the visuals show what the text says they show.

For Northbank, the review should catch any wording that turns a possible evening appointment into confirmed availability. It should also catch a post asking for a quotation when the destination only offers a general newsletter signup.

Approve the final content yourself or assign a named reviewer. A chatbot prompt produces drafts; publishing, scheduling or sending requires a separate, supported setup with the right access. Verify it before use and keep human approval for customer-facing actions.

6. Publish, record results and make one decision

Before launch, check how you will recognise a campaign enquiry. You might use an existing campaign-tagged link and a field in your enquiry tracker. If that setup is unavailable, record the customer’s stated source and keep the limitation visible.

For the fictional pilot, record the publication dates, enquiries, suitable quote requests and subsequent bookings. Track preparation and review time as well. Keep customer-identifying information out of any analysis that does not need it.

After three weeks, compare the actual count of suitable quote requests with the target of six. If reliable visitor counts are available, you can also calculate the proportion that became enquiries: enquiries divided by visitors, multiplied by 100. Use the same period and definition for both counts.

Ask AI to organise the figures, identify missing information and suggest explanations. Recheck calculations and treat explanations as hypotheses unless you have supporting evidence. A few enquiries cannot establish that one headline caused the result.

Choose one next step: continue the campaign, clarify the offer, improve the enquiry page or try a different message. For a repeatable review document, follow the AI marketing report guide.

When the process becomes regular, request a free assessment for help setting up one recurring desktop AI workflow. We agree the task, tools and scope before payment, then build it with you and repeat the workflow together.

Frequently asked questions

Can AI create a marketing strategy from just my website?

It can prepare an initial analysis, but your website may omit capacity, margins, customer feedback and commercial priorities. Supply the information that changes the decision and review the recommendations against the actual business.

How much content should I create for my first campaign?

Create enough to explain one offer and direct people to one clear action. Choose a volume you can check, publish and learn from. Add channels when you have a reason and the capacity to maintain them.

What if the campaign gets attention but no useful enquiries?

Check the path from message to offer to enquiry. Review who responded, what the destination promised and whether the next step worked. Use those findings to choose the next change; attention alone does not tell you which part needs fixing.