Six marketing jobs, six prompts, and one rule under all of them: the AI only gets specific when you hand it your customers, your rivals and your numbers. AI Central's The Marketer's AI Week maps a working week onto that rule, from reading reviews at the start of the week to a Friday report built from your own campaign export.
Why a plain strategy request fails
The guide opens with the failure most marketers have already seen. Ask for a marketing strategy for your product and the answer is a textbook: define your audience, set SMART goals, choose the right channels, create engaging content, measure your results. Every line is correct. None of it is about your brand, because the chat had nothing about your brand to work from.
Advice for any brand moves no metric for yours
Start with customers and rivals, not with ideas
Step one reads your customers. Paste 30 to 50 reviews or support tickets with the names removed, then ask for pains, goals and the exact phrases people use, with a count for each. The guide's prompt asks for the top 5 pains and the top 3 outcomes, each with a count and 2 exact quotes, then 10 phrases customers use that your website does not. Its last line, use only what is in the reviews, stops the model from padding the list with generic pains.
Step two maps the competition. Paste the homepage and pricing text of 3 rivals and ask for a table of main promise, target buyer, price for 10 users and the proof each one shows. Then ask for the 2 claims all three make and the 1 need from your customer research that none of them answers. That need is the useful output, and it only exists because step one came first.
The gap is where you stand out
One positioning line the team repeats
Step three turns the research into a sentence, in three fields: For, who and in which situation; Instead of, what they use today; Proof, a number or a name that backs the claim. The prompt asks for 3 statements under 35 words each, then asks which claim a skeptical CFO would doubt first. That second question switches the AI from writer to critic, so the weakest claim shows up before a buyer finds it.
Plans and ad tests with real limits
Step four gives the campaign the limits a real plan has: a goal as one number and a date, a budget as the money and channels you actually have, and the stages of awareness, consideration and conversion. The worked example asks for 300 demo bookings by 15 March on $20,000, LinkedIn ads, an email list of 8,000 and one webinar, and gets back a 6-week plan by funnel stage with a channel, message, weekly tasks, owner and one KPI per stage. It also asks for the riskiest assumption and a way to test it in week 1.
Step five designs the ad tests as a grid instead of ten random ads: 3 angles, which are reasons to buy taken from your research, against 2 audience segments. Each cell gets a hypothesis, one headline, one line of body copy and the metric that decides it. You paste the channel's character limits so the copy fits on the first try.
Friday: numbers in, decisions out
Step six writes the weekly report from the data, not from memory. Upload this week's and last week's campaign export as a CSV and ask for a 150-word report for your manager: the 3 biggest changes with numbers, the likely cause of each, and 3 actions for next week. The prompt makes the AI mark guesses as guesses and check every figure against the file, which is what makes the report safe to forward.
The brief that feeds every prompt
The pro tip is one file that holds your audience, positioning, offers, words to avoid and last quarter's results. Add it to a project in ChatGPT or Claude, or paste it at the top of each chat, so every answer starts from your facts. The guide's sample brief even lists the words to keep out of the copy, revolutionary and seamless.
Brief it once, use it all week
What to watch out for
- Paste real material every time: reviews, rival pages, the campaign export. The output is only as specific as the input.
- Remove customer names from reviews and tickets before you paste them.
- Keep the instructions that limit the AI to your data, such as use only what is in the reviews and check every figure against the file.
- Run the steps in order. The ad angles in step five come from the research in step one.
Where to start
The closing card is three actions: open ChatGPT or Claude, start with the customer research, and save the prompts that worked. Step one is the natural first run, because later steps draw on its research: the unanswered need in step two and the ad angles in step five.