Most businesses use AI the same way. Open a chat window. Type "write me a Facebook ad for my business." Paste whatever comes back.
What comes back sounds like every other business that typed the same sentence.
Blame the job you gave it. Language models learn from enormous amounts of existing text, so when you ask one to create from scratch it drifts toward the average of what already exists. You get average copy for an average offer, delivered at speed.
There is a far more valuable job for these tools, and almost nobody hands it over.
Stop asking AI to write for you. Make it argue with you.
Give it your offer, your price and your buyer. Tell it to play the sceptic. It will drag out the objections and weak claims your team is too close to see.
Why a critic beats a copywriter
Every business has blind spots about its own marketing. You know the product too well and you have heard your own pitch a thousand times. Your team is too polite, or too invested, to tell you the offer is weak.
Real critics are expensive and hard to find. A focus group costs money. A blunt mentor is rare. As for a customer who explains exactly why they did not buy, most owners never meet one.
A chatbot has no ego to bruise and no job to protect. Ask it the right way and it will pick your best idea apart in seconds, at eleven at night, when you finally have time to think about your marketing.
AI writing your marketing makes you sound like everyone. AI attacking your marketing makes you better than everyone.
Criticism also plays to what these tools do well. Spotting a vague sentence or comparing two blocks of text is pattern work. Writing something distinctive about your business is much harder for a machine, because the distinctive part lives in your head and never made it into the training data. We covered the writing side in how to use AI without sounding like a robot. For most businesses, this half is worth more.
Five ways to make it argue
1. The objection list
Give it your offer, your price and a description of your buyer. Ask it to play a sceptical customer and list every reason they would not buy. Ask for more. Then have it rank them by how likely each one is to kill the sale.
You will recognise most of them, which is the uncomfortable part. Now check whether your website and your sales conversations answer each objection head on, or just hope it never comes up.
A prompt that works: "You are a cautious small business owner considering this offer. List every reason you would hesitate or say no. Be specific. Do not be polite."
2. The offer stress test
Paste in your offer and ask it to find the weakest point. Where is the risk still sitting with the customer? Which line is vague? Is there anything that would make a buyer wait until next month? And how would a competitor undercut it?
Fix one thing and run it again. Three or four rounds can sharpen an offer faster than a month of internal meetings.
Then push on one more angle. Why would a buyer choose to do nothing at all? Doing nothing is the competitor most small businesses forget they have to beat.
If you only run one exercise from this list, run this one, because the offer drives everything downstream. As we argued in why the offer still beats the algorithm, no amount of targeting or creative rescues a weak one.
3. The competitor comparison
Copy the public claims from your website and two competitor websites into the conversation. Strip out the names. Ask which business a careful buyer would choose and why, and which one sounds interchangeable.
If it cannot tell you apart, neither can your customers.
Follow up by asking what specific, checkable claim would make one of the three stand out. Usually the answer is a number or a guarantee that nobody in the category is stating plainly.
4. The hostile headline review
Give it ten headlines and ask it to act as a tired, distracted person scrolling on a phone. Which would they skip, and why? Which make a claim they would not believe?
It cannot pick the winner for you. Only the market does that. What it can do is flag the obviously weak ones before you pay to test them.
5. The pre mortem
This one is borrowed from the psychologist Gary Klein, who described the project premortem in Harvard Business Review in 2007. You assume the project has already failed, then work backwards to why.
Before launching a campaign, describe it in full and give the model this prompt: "It is three months from now and this campaign failed badly. What went wrong?" Expect a list of risks you had not considered, from tracking gaps to a landing page that does not match the ad.
Turn the answers into a checklist. Fix what you can before launch. Write the rest down, so you know what to watch in the first fortnight.
A worked example: three rounds on one offer
This one is made up, to show how the rounds play out.
Say you run a Gold Coast personal training studio selling a twelve week program for $900. The headline on your site says "Transform your body with expert coaching."
Round one, the objection list. The model, playing a sceptical 40 year old office worker, lists: I have tried gyms before and quit, $900 is a lot upfront, I do not know what happens in week one, I am embarrassed about my fitness level, and every studio says "expert coaching."
Round two, the stress test. Asked where the risk sits, it points out the customer carries all of it. They pay upfront, with no guarantee and no way to try a session first.
Round three, the do nothing test. Asked why someone would simply not act, it says: "I will start in the new year when things calm down."
None of that is genius, and all of it is useful. Next you check your sales calls. Do people really say these things? Suppose the embarrassment and the upfront cost come up constantly. So you change two things: you add a payment plan, and you describe the first session step by step, including "no mirrors, no audience, just you and your coach." Then you let real enquiries tell you whether it worked.
How to prompt a useful critic
A lazy prompt gets a polite answer. Anthropic's own research found that AI assistants tend to tell people what they want to hear, so you have to give yours explicit permission to be harsh.
- Give it real context. Your offer, price, audience and the claims on your website, word for word.
- Assign a role. A sceptical buyer, a rival business owner, a demanding procurement manager. A role gets you sharper objections than "give me feedback."
- Ask for weaknesses only. No praise, and no rewrites yet.
- Push back on its answers. Ask "what else?" and "which of these matters most?" The good material tends to show up in the second and third rounds.
- Protect your data. Never paste customer personal details or confidential figures into tools your business has not approved.
One more trick: feed it real customer words. Paste in anonymised reviews or common questions from enquiries and ask which objections they reveal that your marketing never answers. Your customers have usually told you what is wrong already, as we explored in your customers already wrote your best marketing.
Turn objections into marketing
An objection list is only worth something if it changes what the customer sees.
- Pick the top three. Choose the objections you have also heard from real buyers, ranked by how often they kill a sale.
- Answer each one on your website. Give it a clear section or FAQ entry, in plain words.
- Change the offer where words are not enough. A buyer worried about risk needs a guarantee or a smaller first step. One stuck on price needs to see what it replaces, or payment terms.
- Brief your ads with it. Each major objection is an ad angle. "Worried you will quit again?" is a stronger hook than another feature list.
- Arm your sales conversations. Whoever answers the phone should have a short, honest answer to each one.
Keep the judgement human
A critic does not make the decision. The model has never met your customers and has no idea what your margins are. Some of its objections will be wrong or irrelevant, and a few will contradict each other.
Your job is to weigh them. Keep the ones you have heard from real buyers, or that match what your sales calls and reviews already tell you. Then ask whether each fix is worth what it costs.
Be especially careful with any facts it offers about your market or your competitors. Even the best models sometimes generate text that is factually incorrect, and they say it in the same confident voice as everything else. Treat each claim as a lead to verify.
Then check the important ones against reality. Ask customers. Review the deals you lost. Run the test. AI can raise the right questions faster than any meeting, but only your market can answer them.
If you want a human critic in the room as well, someone who has seen hundreds of offers and will tell you plainly which part is weak, that is what our consulting work is for.
Try it tonight
Take the offer you are running right now. Paste it into the AI tool you already use, with your price and a short description of your ideal buyer.
Ask it to play your most sceptical prospect and give you ten reasons not to buy, with no praise and no rewrites.
Pick the objection that stings the most, and fix that one thing on your website this week. It is worth more than a hundred generic ads the same tool could have written for you.
Frequently asked questions
How can I use ChatGPT to improve my marketing?
Use it as a critic, and leave the writing for later. Give it your offer, price and a description of your buyer, then ask it to play a sceptical customer and list every reason they would not buy. Have it find the weakest point in your offer, compare your claims with two competitors, and imagine why a campaign failed. Then fix what matters most.
Why does AI written marketing sound generic?
Language models learn from huge amounts of existing text, so when asked to write from scratch they drift toward the average of what already exists. Without specific details about your offer and your customers, the output sounds like every other business. The distinctive parts of your marketing come from knowledge the model does not have unless you provide it.
What is a pre mortem in marketing?
A pre mortem imagines a campaign has already failed and asks what went wrong, before you launch it. Describe the full campaign to an AI tool or your team and ask for the reasons it flopped. The answers, such as tracking gaps or a landing page that does not match the ad, become a checklist to fix before spending money.
Is it safe to paste business information into AI tools?
Your offer and your pricing are fine to share, since they are already public or close to it. Never paste customer personal details, confidential financials or anything covered by a privacy obligation into tools your business has not approved. Anonymise reviews and enquiries before using them, and check your chosen tool's data settings before you start.