Why Your Meta Ad Copy Is Getting Ignored (And the Research System That Fixes It)

Most people approach Meta ad copywriting the same way. They open ChatGPT, type something like "write me a Facebook ad for my product," get back something that sounds vaguely professional, paste it into Ads Manager, and wonder why it isn't converting.

The problem isn't the writing. It's the research, or more accurately, the lack of it. Good Meta ad copy doesn't start with a blank page or an AI prompt. It starts with your customer's actual words, their real frustrations, the phrases they use when nobody's trying to sell them anything. When you build copy from that foundation, the difference in how it performs is dramatic.

This post walks you through the research system that sits behind the best-performing Meta ad copy, and how to turn that research into headlines and primary text that actually get people to stop scrolling.

Why Generic AI Copy Stops Working

When AI tools like ChatGPT first arrived, the output felt genuinely impressive. You could generate an ad in thirty seconds that was better than most small business owners could write themselves. That gap has since closed, and not because the tools got worse.

The issue is that everyone is using the same tools with the same basic prompts. So now the AI is essentially producing the same patterns, the same sentence structures, the same opening hooks, across thousands of different brands. Your ad starts to blend in with everything else on the platform. And blending in on Meta is the same as being invisible.

The fix is not to stop using AI. It's to feed it something that nobody else is feeding it: genuine, unfiltered customer language that you've gone and found yourself.

The Research Document: Where Everything Lives

Before you write a single word of copy, you want to build a research document. Think of it as a training file for your AI. It's the place where all your customer insights live, and every time you go back to write new ads, you draw from it and add to it.

The document has a few key sections:

• A tab for Reddit research, which we'll get to in a moment

• A tab for insights pulled from your own ad account, including comments, questions, and objections that real customers have left

• A tab for competitor ad examples

• A section for the outputs you generate when you run this material through an AI model

Done properly, building this document for a new brand takes a couple of hours. That probably sounds like a lot when you're used to firing off a quick AI prompt. But this document will serve you for months, and the quality difference in your copy will be obvious almost immediately.

Reddit: The Most Underrated Research Tool in E-commerce

Your own product reviews are useful, but they have a blind spot. The people who leave reviews mostly liked what they bought. You're getting the happy customers. You're not hearing from the people who almost bought but didn't, or the people who have been thinking about your product category for six months and can't quite commit, or the people who tried a competitor and were disappointed.

Reddit gives you all of those people.

The way to use it is simple. You search for your product category, your niche, or your competitors, and you start reading threads. You're not looking for product mentions specifically. You're looking for the language people use when they talk about this space without any marketing influence. What words come up again and again? What frustrations keep appearing? What does someone say when they're explaining to a friend why they finally bought something like this?

A real example: working through this process for a tea leaf reading brand, a few things surfaced quickly that you'd never find in a five-star product review. People talked about difficulty interpreting the symbols as a beginner, and how intimidating it felt not knowing the "right" answer. People talked about their grandmothers doing this, and the emotional connection to that memory. People talked about the ritual of it, the quiet and the calm, not just the reading itself.

None of that language would come from a generic AI prompt. But all of it is gold for copywriting, because it's the actual vocabulary your potential customer uses when they think about this product.

What you do with these threads is straightforward:

• Copy and paste the most relevant answers into your research document

You're building a library of real customer language, real objections, and real emotional triggers. Later, when you prompt your AI with this material loaded in, the output will sound nothing like a generic ad. It will sound like someone who actually understands the customer.

Competitor Research: What the Impressions Sort Just Changed

The Meta Ads Library has always been useful, but a recent update made it significantly more powerful. You can now sort competitor ads by impressions, from high to low.

This matters because before this update, the best proxy for a high-performing ad was how long it had been running. The assumption was that if an ad had been active for months, it was probably working. That was a rough guess at best. Impressions gives you something much closer to a direct signal. If an ad is getting a huge number of impressions, it's because money is being put behind it. That means it's working.

When you're researching competitors:

• Sort by impressions high to low

• Spend time with the top results

• Look at the copy structure

• Analyse the primary text

• Look at the headline

What does the primary text actually say? Does it lead with a pain point or a dream outcome? How long is it? Does it use bullet points? What does the headline do? Is it specific, or is it vague and lazy?

You'll often find that even big brands with serious ad budgets are leaving a lot on the table with their copy. Fashion brands in particular tend toward lines like "back in stock" or "latest drop" with nothing underneath it. That's not selling anyone on anything. It's just announcing. If your copy actually speaks to a desire or a frustration, you're already ahead of a lot of your competition.

Paste five to ten of the strongest competitor primary text examples and headlines into your research document. You're not there to copy them. You're there to understand the patterns and then do them better.

The Copy Structure That Actually Works on Meta

Most people either write one sentence in the primary text field or dump a wall of unbroken prose. Both of these approaches leave conversions on the table.

The structure that tends to work best on Meta has three parts:

1. A hook

2. A body that uses a recognised copywriting framework

3. A clear call to action

The Hook

The hook is the most important line you'll write. It's the first thing someone reads before they decide whether to hit "see more" or keep scrolling.

A hook that works does one of two things:

• It names a specific feeling or situation the reader recognises in themselves

• It creates a question they feel compelled to answer

"Some days you're everybody else's problem solver, but nobody solves your problem." That's a hook. It lands immediately for a particular kind of person and does nothing for everyone else, which is exactly what you want.

The Body

The body of your primary text should follow a framework.

The most useful ones for Meta ads are:

• PAS (Problem, Agitation, Solution)

• BAB (Before, After, Bridge)

• AIDA (Attention, Interest, Desire, Action)

The specific framework matters less than the discipline of using one, because it forces the copy to actually move somewhere rather than just sitting there describing the product.

One practical addition worth making to any framework is bullet points. You cannot use actual bullet point formatting in Ads Manager, but you can use dashes or emojis at the start of lines to achieve the same visual effect. This matters because when someone taps "see more" and a wall of text expands, most of them will not read it. But they will scan it. Bullet-style lines with emojis at the start give their eyes somewhere to land, and if any one of those lines resonates, they'll go back and read the whole thing.

The Call to Action

Your call to action should be specific.

"Tap to shop" tells someone nothing about why they should bother.
"Brew a cup tonight and get answers that feel personal" tells them exactly what the experience will be.

Specificity creates a picture. Generic instructions do not.

Two Variations, Every Time

For any given ad angle, you want two different primary text variations. Not because one will definitely outperform the other, but because different people respond to different emotional registers.

Pain-Led Opening

One variation should lean into the pain. It starts in the problem, sits with it for a moment, and then offers the way out.

"By the time the kids are in bed, you're fried. Brain racing, decisions piling up, zero headspace left for yourself."

That's a pain-led opening. It works for people who are currently feeling the weight of whatever problem you solve.

Dream-State Opening

The other variation should lean into the dream state. It starts with the outcome, the way life looks after the problem is gone.

"Imagine ending your day calm, clear, and certain you're on the right path."

Same product, same customer, completely different emotional entry point.

Having both on your breakthrough board means you're not betting everything on one approach. Meta's algorithm will also find different audiences for each variation over time, which is useful data about who your most responsive customers actually are.

Headlines: Shorter Than You Think

The most common headline mistake is length. ChatGPT, if you ask it for headlines without specific constraints, will tend to produce lines that look like news article titles. They're full sentences, often fifteen or twenty words, and they get cut off in most placements before the reader even finishes them.

A headline that works on Meta is usually under 45 characters. That's not a lot of space, which is why every word has to pull its weight.

The best headlines tend to follow one of a few patterns:

• They name the problem directly ("Mum burnout isn't normal")

• They make a specific promise ("10 minutes of total clarity")

• They use a pattern interrupt ("Don't buy this if you love chaos")

• They speak directly to a specific identity ("For busy mums who've forgotten themselves")

When you generate headlines using your research document as context, you'll get much better results by giving the AI an explicit character limit and asking for ten variations at once. Then you look at the preview in Ads Manager to confirm nothing is getting cut off.

Finding the Words Your Customers Actually Use

The single biggest upgrade you can make to your copywriting is getting specific with language. Not the words you think sound good, but the words your customers actually say to each other when you're not in the room.

This is where the Reddit research pays off, but there are other sources too:

• Comments on your own ads

• Questions and objections

• Bad reviews on competitor products

• Recorded and transcribed customer conversations

A useful way to think about it: if you're selling tea leaf reading cups and the two questions your customers most want answered are about their relationships and their careers, then those words should appear early in your copy. Not because you engineered it that way, but because those are the words running through your customer's head when your ad shows up in their feed. The closer your copy matches what they're already thinking, the less work it has to do.

The Process, Start to Finish

If you want to implement this, here is the order of operations.

1. Start by building your research document.

2. Pull Reddit threads related to your product category, your competitors, and the broader problem your product solves.

3. Copy paste the most relevant answers into your document, organised by theme: pain points, desires, buzzwords, objections.

4. Go through your own ad account and pull out any comments that reveal customer language.

5. Add competitor primary text and headline examples.

6. Upload your document to your AI of choice.

7. Run it through a framework prompt that asks for two primary text variations per angle, using a strong hook, a body built on a recognised copywriting framework, bullet-style lines with emojis, and a specific call to action.

8. Run a follow-up prompt asking for ten headline variations under 45 characters each.

From there, you're not taking the output wholesale. You're reading through it, pulling the lines that actually land, and assembling the best version yourself. The AI is doing the structural and generative work. You're doing the editorial judgement.

Copy that comes from this process sounds different from generic ad copy. It sounds like it was written for a specific person, because in a meaningful sense it was.