A home services client signed in January. 6 vans, one city, 11 Google reviews that all say some version of "great job, on time". No email list worth exporting, no support inbox, launch in 2 weeks.

The room did what rooms do. Energy savings, financing, before-and-after photos: 8 angles went live inside a month. By February the Meta Ads Manager could tell us which one had taken the most spend, and nothing about why.

Short answer: Audience research for a new product means borrowing someone else's buyers. Pick the competitor closest to you in price rather than the largest one, read their reviews for objections and their forum threads for pre-purchase language, and treat the first round of ads as the test that tells you which source was right.

The takeaways

Borrow from the competitor whose price you match, not the one whose logo you know. Their reviewers bought partly on recognition, and recognition is an input you cannot reproduce in week one.
Reviews are post-purchase language. Nobody writes a review before buying, so a review pile hands you objections while your cold ad has to speak to someone who has decided nothing.
Budget the first round as research. Run 3 angles from 3 different sources at one spend level, and whichever source produced the winner is the one worth another 2 days.

Why does audience research break when the product is new?

Every voice-of-customer method assumes you own the sources. Survey your customers, mine your reviews, read your support tickets, pull your search terms. A new product has none of those, so the advice collapses into two substitutes: survey people who might buy one day, or fill a persona template from demographics and instinct.

The first produces stated preference. Ask someone what would make them switch heating providers and you get a considered, slightly flattering answer nobody has ever given while scrolling. The second produces a slide called Marketing Mary. I wrote about the evidence standard a persona has to clear separately.

There is a harder break underneath both. With no customers you cannot check the research either. Mine your own reviews and you can test a hook against people who already bought. Borrow language and that check disappears. The cold start is 2 problems stacked: finding words, and knowing whose they are.

Whose buyers should you borrow from?

Pick the competitor whose price and buying trigger match yours, even when they are far smaller than the market leader. The leader's reviewers weighed a decision your ads cannot cause: part of what they bought was a name they had already heard.

Work through three tiers before you open a single review page.

Tier one is the direct competitor at your price point. Their buyers weighed the same trade-off yours will.

Tier two is the substitute: what people do instead of buying the category at all. For home services that is the neighbour with a van, a YouTube tutorial, or another winter of doing nothing. Substitute language carries the objection your ad beats before any competitor comparison matters.

Tier three is the adjacent category people graduate from. Weakest of the three, and the only one showing what a buyer believed before they knew your category existed.

Why are reviews the wrong corpus for a cold ad?

Because nobody writes a review before they buy. Every review was written by someone who had already chosen, paid, and lived with the outcome. That pile is excellent for objections and the exact wording of a broken expectation. As a model of the undecided person your cold ad interrupts, it is poor.

A one-star review is the least contaminated one you will find. A happy customer has usually absorbed the vendor's own words and hands them back ("the onboarding was smooth"). An angry one describes the situation in their own vocabulary and says what they expected.

Even the angriest review starts from a purchase, though. It tells you what to promise and what to defuse, and says nothing about what that person typed into Google the week before. Cold traffic is made of those people. Amazon, the Apple App Store and Google Play are objection libraries. Stop asking them to do the other job.

Where does the pre-purchase language live?

In threads where somebody is asking rather than reporting. The "which of these should I get" post, the "is it worth it" post, the "has anyone dealt with this" post. Reddit and niche category forums carry most of it, and the valuable part is the question itself, before any vendor gets named.

Three things to lift from a pre-purchase thread. The criteria the person lists unprompted, in their order, because that ordering is the buying logic you are trying to enter. The alternatives they name, which gives you the real consideration set, rarely the one in the client's deck. And the fear, one line, near the end.

Those quotes sit at the earliest of Eugene Schwartz's 5 awareness stages, where cold traffic sits too. Post-purchase quotes cluster at the most-aware end and read beautifully in retargeting. 20 to 30 usable quotes across both piles is enough to see what repeats. The mechanics of pulling and tagging are the same either way, and I covered that method in voice of customer research for ads.

How do you design the first round so it teaches you something?

Run one angle from each source rather than the three best overall: a post-purchase objection, a pre-purchase question, a substitute-category frustration, at the same spend and offer. Whichever wins tells you which pile describes your buyer, and that is worth more than the winning ad.

Fund it properly or skip it. Three angles at €20 a day for 4 days produce numbers you cannot separate, so you choose on instinct after paying to avoid exactly that. Two sources with enough events behind each beat three ambiguous ones.

Then the honest part. Borrowed language buys a hypothesis and nothing more. Their buyers are not your buyers, the overlap is unknown, and the only instrument that measures it is spend. The first round is the research. Everything before it exists to make the first round less stupid.

Doing this by hand on a new account eats 2 days, which is why I built it into Adscalr. It searches 5 named sources (Reddit, Amazon reviews, the Apple App Store, Google Play and custom forums) seeded with your competitors' URLs as well as your own, tags each quote with 2 to 4 phrase markers plus emotion and register, and maps it to one of Schwartz's 5 awareness stages, so the two piles separate themselves. Seeding it with only competitor URLs is the cold start: more on the audience intelligence page.