AI marketing prompts: how to get usable campaign output
Short answer Usable campaign output comes from the process, not one clever prompt. Write a short campaign brief once and paste it into every request. Ask in steps: angles first, then each channel's pieces in its exact format, then a review against the brief. Give real examples and past results, then check every fact and limit yourself.
Want help applying this to your business? Talk it through on a free 30-minute call.
Book a callAI marketing prompts: what makes campaign output usable
A draft is usable when you could run it today: it sells the right offer to the right person, fits the channel's limits, says only true things and sounds like your business. Most AI drafts miss on at least one of those because the prompt left it out.
Five habits fix most of it:
| Habit | What it fixes |
|---|---|
| One campaign brief, pasted into every prompt | Ads, emails and the landing page that drift apart |
| Ask in steps, not one giant request | Long, average output nobody can edit |
| Name the exact format and limits | Headlines too long to paste into the ad tool |
| Give examples and past results | Copy that sounds like every other business |
| Make it review its own draft, then check facts yourself | Made-up details and broken promises |
Anthropic's prompting guide gives a useful mental model:
"Think of Claude as a brilliant but new employee who lacks context on your norms and workflows." (Claude Platform Docs: Prompting best practices)
The same holds for ChatGPT or Gemini. A new hire with no brief writes a generic ad. Give the same person a one-page brief and they write yours.
Start with a campaign brief the AI can't misread
A brief is a short summary of the campaign that every piece must follow. Write it once, in plain sentences, and paste it at the top of every prompt (or save it in a ChatGPT project or custom instructions).
| Field | Example (a lawn care company) |
|---|---|
| Offer | First mow free with a 6-mow spring plan, booked by April 15 |
| Who it's for | Homeowners in our 3 zip codes with no lawn service yet |
| The one action | Book online or call |
| Proof you can show | 412 Google reviews, 4.8 stars; same crew every visit |
| Tone | Friendly, plain, no hype |
| Never say | "Guaranteed", "best in town", any price not on our price list |
| Channels | Google search ads, one email, one Facebook post, landing page |
The "never say" line matters most. A model fills gaps with likely words, and likely marketing words include claims you can't back up.
OpenAI's own advice for ChatGPT starts here too:
"Ensure your prompts are clear, specific, and provide enough context for the model to understand what you are asking." (OpenAI Help Center: Prompt engineering best practices for ChatGPT)
Prompt in steps, not one giant request
Asking for "a full spring campaign" in one message gets you a long, middling answer that's hard to fix. Split it, and look at each result before the next step. Anthropic calls this chaining and names the most common pattern:
"The most common chaining pattern is self-correction: generate a draft → have Claude review it against criteria → have Claude refine based on the review." (Claude Platform Docs: Prompting best practices)
A four-step run for a small campaign:
- Angles. "Using the brief, give me 5 different angles for this offer, one line each, and who each one appeals to." You pick one.
- Pieces. "Using angle 3, write the Google ad text, the email and the Facebook post. Formats below." One channel per message if the output gets long.
- Review. "Check every piece against the brief. List anything that breaks the 'never say' list, any claim not in the brief, and any line over its limit."
- Fix. "Rewrite only the lines you flagged."
Picking the angle in step 1, before any copy exists, saves you from editing three drafts that argue for different things.
Ask for the exact format each channel takes
AI doesn't know which tool you'll paste into. Name it and give the limits. Google's limits for responsive search ads (the standard Google search ad):
"The headline fields for responsive search ads support up to 30 characters. The description fields support up to 90 characters each, and the path fields support up to 15 each." (Google Ads Help: About responsive search ads)
Google also says you can enter "up to 15 headlines and 4 descriptions" per ad, and it mixes them into combinations. So each headline must make sense next to any other one.
A format line that works:
Give me 15 headlines (30 characters max, each makes sense on its own, at least 3 include "lawn care") and 4 descriptions (90 characters max). Return a table: text, character count.
Check the counts yourself before pasting. The ad tool won't accept a line over its limit, and a model's own count isn't proof.
For email, say: subject line, preview text, body under a word count, one link. For a landing page, say: headline, three benefit lines, the form's call to action. For more channel prompts, see ChatGPT prompts for ad copy and ChatGPT prompts for email copy.
Feed it examples and past results
Examples show the AI what you mean better than adjectives like "punchy" or "friendly". OpenAI's prompting guide describes the technique:
"Few-shot learning lets you steer a large language model toward a new task by including a handful of input/output examples." (OpenAI API docs: Prompt engineering)
In marketing, your best examples are your own: the email that got the most bookings, the ad with the best click rate, a review in a customer's words. Paste two or three with one line on how each did ("this subject line got twice the opens of the others"). Ask for new pieces that keep what worked and change the angle.
Keep customer names, emails and phone numbers out. On ChatGPT Free, Plus and Pro, OpenAI says data sharing to improve its models "is enabled by default"; you can turn off "Improve the model for everyone" in Settings, Data controls (OpenAI Help Center: Data controls FAQ). Summarize results as numbers, not customer lists. To keep a consistent voice across campaigns, see ChatGPT prompts that keep your brand voice.
What the AI can't check for you
A self-review catches limits and off-brief lines. It can't confirm facts it never had. OpenAI is direct about this:
"Always verify quotes, data, technical information or references to external documents." (OpenAI Help Center: Does ChatGPT tell the truth?)
Before anything runs, a person checks:
- Offer terms: price, dates, what's included, the fine print. These must match what staff will honor.
- Numbers and claims: review counts, years in business, "fastest", "licensed". Each must be true today.
- Every promise of results. Delete any the brief didn't contain.
The law treats an AI-written ad like any other. The FTC's guide for small business says "Advertising must be truthful and non-deceptive" and "Advertisers must have evidence to back up their claims" (FTC: Advertising FAQ's). In Canada, the Competition Act bars any representation to the public "that is false or misleading in a material respect" (Competition Act, section 74.01).
Example
A lawn care company in Ohio (US dollars)
The owner writes the brief above, asks for 5 angles and picks "your Saturdays back". One more prompt returns 15 headlines, 4 descriptions, an email and a post. The review step flags "Ohio's best lawn care" (breaks the never-say list) and two headlines at 32 characters. After fixes and a fact check, the ads go to Google's review with a $600 a month search budget.
A bookkeeper in Ontario (Canadian dollars)
She pastes her three best past emails with their open rates and asks for a January campaign for her $250 tax-prep package. The first draft promises "maximum refund". She deletes it: she can't prove it, and the brief never said it. She asks for two subject lines to test against each other and sends each to half her list.
FAQ
Is there one best prompt for marketing?
No. A saved brief plus the four-step run (angles, pieces, review, fix) works better than any single template, because the brief carries what only you know.
Which AI tool is best for campaign copy?
ChatGPT, Claude and Gemini follow the same prompting basics. Pick the one you'll use daily; keep your brief where any tool can use it.
Why does AI keep making up details like awards or prices?
It predicts likely words and fills any gap in the brief. List the facts it may use and say "use no other facts or numbers."
How many versions should I ask for?
Ask for several angles or headlines, then pick. Ask for one full draft at a time after that, so each review is short.
Different for you?
- Running Google or Meta ads and want the ad side done for you: Paid Ads.
- Worried Google will rank AI-written pages lower: does Google penalize AI content?
- The campaign sends people to a weak page: landing page best practices.
- You need copy for one page, not a campaign: ChatGPT prompts for copywriting.