Guide

Why AI-Written Ad Copy Comes Out Generic (It Is Not the Model)

By Diego Cipion · Updated September 24, 2026

AI-written ad copy comes out generic because the prompt usually contains nothing specific about the buyer, the objection that stops them, or the proof the brand can actually offer, so the model does the only thing it can with a thin input: it writes the most typical ad for that category.

Switching to a newer model rarely fixes this. A stronger model writes the same average ad more fluently. The fix is on the input side: give the model the knowledge that makes one campaign different from every other campaign in its category, and the copy stops sounding like everyone else's.

This guide is from Populica, a tool that researches buyer personas for a company from its name, website and sector and turns them into a campaign brief and ad creative. The diagnosis and the fixes below work with any AI writing tool.

Why a language model defaults to the average ad

A language model writes by predicting what text is most likely to come next given everything in its context. When the context says "write a Facebook ad for an accounting firm, make it compelling", the most likely continuation is the ad that thousands of accounting firms have already run: save time, reduce stress, trusted experts, book a call today. Nothing in the prompt pushes the model away from that centre, so it lands there.

This is not a flaw in the model. It is the model working correctly on an input that describes a category instead of a campaign. Words like "compelling", "punchy" or "high converting" do not help, because they are instructions about tone, not information about the buyer. The model can only use what is in its context, and a request for a better adjective is not new information.

The four things a generic prompt is missing

Compare a prompt that produced flat copy with one that produced something usable and the difference is almost always in the same four places.

A specific buyer. Not "small business owners" but one person in one situation: the owner of a two location dental practice who just lost her office manager and is doing payroll at night. A specific buyer gives the model a specific problem to write about.
The objection that stops them. Every buyer has a reason not to act, such as "switching will take weeks" or "we tried an agency and got burned". Copy that answers the real objection reads as written for someone. Copy without one defaults to listing benefits.
Proof the brand can actually claim. A named feature, a real guarantee, a price, a turnaround time, a certification. Without real proof, the model fills the gap with vague superlatives, or worse, invents a claim you cannot legally make.
Constraints. What the brand never says, words the category overuses, regulatory limits, the platform's character count. Constraints push the output away from the common phrasing that makes ads feel machine written.

All four are knowledge about this campaign. None of them can be supplied by the model, because none of them exist anywhere the model has seen. If you are starting a new account with no customer data to draw them from, our guide to building buyer personas when you have zero customer data covers how to draft the first two honestly.

Better model, better template, or better brief?

Teams that are unhappy with AI copy usually try one of three fixes. They are not interchangeable, and each one wins somewhere.

	Upgrade the model	Better prompt template or skill	A written brief with buyer knowledge
What it adds	Fluency, instruction following, handling of long inputs	Method: structure, formats, hook patterns, review steps	Knowledge: this buyer, this objection, this proof, these limits
Same for your competitors?	Yes, anyone can use the same model	Usually, templates are shared and copied	No, it is specific to one brand and one audience
Fixes generic copy?	Rarely on its own	Partly: tidier, more consistent, still generic	Yes, it is the only one that changes what the copy is about
Where it wins	Complex briefs, long documents, strict formatting rules	Consistency across a team and speed on repeat formats	Relevance to a real buyer, which is what makes an ad work
Survives a tool change?	No, it is the tool	Partly, it often needs rewriting per tool	Yes, any model or tool can read it

The honest reading of the table: a good template is worth keeping, and a stronger model does help once the input is rich. But both improve how the copy is written, not what it says. Only the brief changes what it says, because only the brief carries information the model did not already have.

Want to see what specific buyer input looks like for your own company? Enter your company name and sector and get three researched personas, each with the objection that stops them buying. No account needed.

Draft my audience free →

What the difference looks like in practice

An illustrative example, using an invented bookkeeping firm rather than a real client. The first prompt is the common one.

Prompt A: "Write a compelling Facebook ad for a bookkeeping firm for small businesses." The output will almost certainly promise saved time, less stress, accurate books and peace of mind, closing with "book a free consultation". It could run for any bookkeeping firm in any city.
Prompt B: "Buyer: owner of a restaurant with 12 staff who does the books herself on Sunday nights and is three months behind. Objection: she thinks handing over the books means losing track of her cash. Proof we can claim: weekly cash summary by email every Monday, fixed monthly fee, restaurant clients only. Never say: peace of mind, hassle free. Write three Facebook ad variants under 125 characters of primary text."

Prompt B does not ask for better writing. It asks for the same ad with four pieces of knowledge added, and that is what moves the output from the category average to something one restaurant owner might recognise as her own week. None of that knowledge came from the model.

How to stop getting generic copy, in five steps

Write the buyer before you write the prompt. One sentence per audience: who they are, what situation they are in right now, and what they are trying to get done.
Name one objection per buyer. The single most likely reason they will scroll past. If you do not know it, ask the people who sell or answer the phone, they hear it every week.
List only proof you can back up. Real features, real prices, real guarantees. Tell the model explicitly to use nothing else, which also stops it inventing claims.
Add the constraints. Banned phrases, compliance limits, character counts, the tone the brand never uses.
Save it as a brief and reuse it. Paste the same brief into every copy request for that audience, and update it when a campaign teaches you something. Our explainer on what a campaign brief is and why most small teams skip it covers the full structure.

This is also why the brief is worth more than any single tool. Models get replaced every few months and copy tools come and go, but a written brief carries across all of them, makes the output reviewable against something concrete, and gets better every time it is revised. If you are weighing copy tools right now, our comparison of Jasper alternatives for marketers who need research, not just copy looks at which ones help with the input side and which only speed up the writing.

Populica's Free plan covers one company with one audience research run and 10 AI creative drafts a month, no credit card required. Research your buyers, write the brief, then draft copy that starts from it.

Start free →

Compare plans

Frequently asked questions

Why does ChatGPT write generic ad copy?

Because the prompt usually describes a category, not a campaign. Without a specific buyer, the objection that stops them, real proof and constraints, the model writes the most typical ad for that category. Adding that knowledge to the prompt changes the output far more than rewording the request.

Will a newer or more expensive AI model fix generic copy?

Rarely on its own. A stronger model follows instructions better and writes more fluently, which helps once the input is rich. With a thin prompt it produces a more polished version of the same average ad, because the missing information is still missing.

Is a prompt template or saved skill enough to get specific copy?

A template adds method: structure, formats and review steps. That improves consistency, but the template is the same for every brand that uses it. Specific copy needs knowledge about one audience, which has to come from a brief or research, not from the template.

What should I put in a prompt to make ad copy less generic?

Four things: a specific buyer in a specific situation, the one objection most likely to stop them, the proof your brand can actually claim, and constraints such as banned phrases, compliance limits and character counts. Tell the model to use only the proof you listed.

How do I find my buyer's real objection if I have no customer data?

Ask the people who talk to buyers, such as sales staff or whoever answers the phone, and read public reviews of your business and competitors. Treat the first draft as a hypothesis and correct it once a campaign reports back.

Does AI copy sound generic because readers can tell a machine wrote it?

Mostly it sounds generic because it says what every competitor says, with the same benefit words and calls to action. Copy about one buyer's specific situation and objection reads as written for someone, whoever or whatever drafted it.