Brand and customer research: the work before the ads

Why research comes first

Ad platforms have made targeting mostly automatic. On Meta today, the ad itself tells the algorithm who it’s for. That means the words and images in the ad do the targeting, and the words have to come from somewhere. If they come from the marketing team’s guesses, they sound like marketing. If they come from customers, they sound like the person scrolling.

Good research answers four questions: who buys, why they buy, what almost stops them, and what they compared you to. Everything in this guide exists to answer those four.

Step 1: Start with what the brand already knows

The founder and the customer service team know more about the customer than any tool. Most of it has never been written down. Before any outside research, we ask the brand to fill in a short onboarding section. The questions that pay off most:

• Describe your best customer as one real person. What’s going on in their life when they buy?
• What are the top three reasons people say they bought?
• What almost stops people from buying? Price, sizing, trust, shipping, “do I need this”?
• What are the five questions customer service gets before someone orders?
• What are the most common complaints and return reasons?
• Which customer group surprised you? The best new avatar often hides here.
• Which words do customers use for your product that you would never use yourself?
• Your three best ads ever, and why you think they worked. The offers you’ve tried, and what happened.

Then ask for the raw material, because raw exports beat summaries: a review export, customer service tickets from the last few months, post-purchase survey answers, the best emails and organic posts, and any recorded customer calls.

Step 2: Collect the customer’s own words

Now go where customers talk when the brand isn’t listening. The goal is 50 or more exact quotes, each with a link, from at least four different sources. One source gives you one kind of customer; four give you the full picture.

A few search strings to copy. Replace the brackets with the product type, the problem or a competitor:

• site:reddit.com [product type] recommend
• site:reddit.com "[competitor]" worth it
• "[product type]" "I wish" OR "the only problem"
• best [product type] for [situation], for example best rain jacket for running at night
• why does my [product type] [problem], for example why does my running jacket get wet inside

Read reviews in this order: 3- and 4-star first, because they’re the most specific; then 1- and 2-star for objections; then 5-star for desires. On competitor products, the negative reviews are the most valuable. Every complaint about them is a reason someone might choose you.

“Wetter from sweat inside than from the rain” is a better hook than anything a copywriter would invent from scratch. In this step you’re collecting headlines as well as insights.

Step 3: Let AI sort it, never invent it

AI makes this step much faster. It can read 500 reviews in a minute and tell you which complaint comes up most. But an AI tool is not a source. Asked about customers without real material, it produces plausible-sounding quotes that nobody ever said, and ads built on them sound like every other ad.

The rule: paste real material in, and check that every quote it returns exists in what you pasted. A prompt we use:

Below are [N] customer reviews of [product] from [source]. Group them into desires, pains, objections, alternatives they tried, and phrases that come up again and again. For each group, list the themes from most to least frequent, with a count and 2-3 verbatim quotes. Only use quotes that appear word for word in the text below. If a theme has fewer than 3 mentions, label it "weak signal".

[paste reviews]

AI is also good at the step before: ask it for 30 searches real customers would type at different stages, from “why does my jacket get wet inside” to “[brand] vs [brand]”. If you use an AI tool with web search to find threads and videos, open every link yourself before you use it.

Step 4: Build the language bank

The language bank is one table with every useful quote, sorted by type. It’s where headlines, hooks, landing page sections and email subject lines come from later.

Next to each quote, note the source, the link and how often the theme came up. A theme that appears once is an anecdote. A theme that appears 40 times is a campaign.

Step 5: Connect features to desires

Brands describe their products in features, because that’s what they built. Customers buy desires. The ladder connects the two:

Do this for every important feature, then check each desire against the language bank. A desire with real quotes behind it is an angle you can test. A desire without any is your assumption, so mark it as one.

Two more lists belong here. Problems and solutions: each problem in the customer’s words, how the product solves it, and the proof. What they tried before: every alternative, including doing nothing, and why it disappointed them. That second list is where “better than X” angles come from.

Step 6: Map desires across the year

What people want from the same product changes with the season. A desire calendar makes that visible, so the ads and offers change before the market does.

Step 7: Research the competitors

Pick three to five direct competitors and note the same things for each:

• The hero product and price range, and the offers they run: bundles, gifts, free-shipping thresholds, discounts.
• Their longest-running ads in the Meta Ad Library. An ad that has run for months is probably profitable. Note the angle, the avatar it speaks to and the format.
• Where their ads send people: a product page, an advertorial, a listicle, a quiz.
• What their customers love and what they complain about, in quotes.
• The gap: the desire or the customer nobody in the market speaks to yet.

Then list the claims the brand could make that competitors can’t, and next to each, how you’d prove it. Drop any claim you can’t prove.

Step 8: Audit what already worked

If the brand has run ads before, the account holds research nobody has read. Pull the top creatives by spend for the last 30 days and the last six months, with their CPA or ROAS. Then look deeper:

• Hook rate (3-second views divided by impressions): which opening image and words stopped the scroll?
• Hold rate and outbound CTR: which message kept people watching and made them click?
• Landing pages: where did the best ads send people, and how did each page convert?

Write the learnings about the customer, not the ad. “The video was short” teaches nothing. “Runners who train in the dark responded to the visibility angle” is something the next ten ads can use.

Step 9: Turn research into avatars and angles

With the language bank, the ladder and the competitor gaps in hand, you can build avatars: a specific person in a specific situation, defined by desire first and demographics last. Then write angles from the gap between what they want and what they do now. We walk through that part step by step in our avatar and angle tutorial, and turn angles into openings in the ad hooks library.

Research is ready to hand over when:

• There are 50+ exact quotes, each with a working link.
• At least four sources were used besides the brand’s own reviews.
• Every avatar has at least five real quotes behind it.
• Every claim has proof, or is marked as needing it.
• The brand has checked the messaging against its own rules and signed off on the top angles.

How long it takes

For one brand with one hero product: about half a day of onboarding for the brand, seven to eight hours of source research, two to three hours of sorting, and half a day to turn it into avatars and angles. Two to three days in total. It’s the cheapest part of any campaign, and the one most often skipped.

Common mistakes

• Starting with demographics. “Women 25–45” tells you nothing about why anyone buys.
• Rewriting quotes into marketing language. The rough version is the one that works.
• Using only the brand’s own 5-star reviews. They describe the people you already convinced.
• Asking AI about the customer instead of giving it the customer’s words.
• Doing the research once. Markets move; refresh it every quarter and before every big season.

FAQ

More guides