AI audience research that names real buyers
Audience research is the step everyone skips because it is slow and produces a document nobody rereads. Done properly it decides every word that comes after it. Here is what a useful segment contains and how to judge the ones a machine hands you.
A demographic is not an audience
Ask most tools who your audience is and you get an age range, a gender split, a city and an adjective. Women, 30 to 45, suburban, values convenience. It feels like research and it is unusable, because you cannot write a single sentence out of it. Nobody has ever bought anything on the grounds of being 38.
A usable segment is a situation, not a category. It is the moment a person is in, the thing that is currently going wrong for them, and the words they would use to complain about it to a friend. That is writable. You can take a hook straight out of it.
What a useful segment contains
The targeting stage returns a fixed shape for each audience, and every field exists because a later stage consumes it rather than because it looks thorough on a page.
| Field | What it holds | What uses it |
|---|---|---|
| Name | A short, specific label. Not a demographic bucket | You, when picking which to run |
| Who they are | A portrait of one real person in the segment | The copy stage, for voice |
| Pain points | What is going wrong right now | Hooks, which are usually a pain stated plainly |
| Desires | What they actually want, not what you sell | The offer framing |
| Objections | Why they would refuse you | The line that pre-empts it |
| Messaging angles | Distinct approaches that would land | One ad per angle, so the batch diverges |
| Keywords | Their own vocabulary for the problem | Wording, so the ad sounds native |
| Tone notes | Register, humour, what to avoid | Stops it writing brochure voice |
| Why it fits | Why this business wins with this segment | You, deciding whether to believe it |
| Size note | Rough reachable size and competitiveness | You, prioritising |
| Fit score | 0 to 100 commercial fit, spread honestly | You, ordering the list |
| Group search terms | How these people look themselves up | The group finder |
The last field is the one that turns research into action. A segment that cannot tell you where those people already gather is a description you will admire and never use.
Why it reads the live web
A model writing about your market from memory produces the average of every marketing article it ever read. It is fluent, plausible and about four years out of date, and it says the same things about a plumbing business in Leeds as about one in Phoenix.
The targeting call runs with web access so it is reading about your actual category, its current complaints and its current competitors, and reasoning from that. It is the difference between a segment that describes your market and one that describes the idea of a market.
It is still prose rather than data. Nothing it returns is a measured figure, and the size notes are characterisations, not numbers to put in a deck.
Fit scores, and reading them honestly
Every segment comes back with a commercial fit score out of a hundred, and the prompt asks it to spread the range rather than award everything an 85. A list where all five score in the nineties is a list that has told you nothing.
| Score | What it usually means | What to do |
|---|---|---|
| 80 and up | Clear fit, clear demand, you can serve it today | Generate ads for it first |
| 60 to 79 | Real but competitive, or slightly off your core | Worth one batch as a test |
| 40 to 59 | Plausible, needs an offer you do not have yet | Park it, revisit next quarter |
| Under 40 | It is being thorough, not recommending | Delete it, and do not pay to write ads for it |
The scores are the model's opinion, not a measurement. The useful part is the ordering rather than the number, and where your own knowledge disagrees with the ordering, you are right and it is wrong.
Correcting a segment it got slightly wrong
The common failure is not a wrong segment. It is a right segment with one wrong detail: an objection that is not really the objection, or vocabulary a half-step off how people in your trade actually talk.
Edit the field rather than fighting it in the ad copy afterwards. The copy stage reads the audience record, so one correction there fixes every ad written from that segment from then on, and the correction is permanent because audiences are saved.
- Fix the objections first. They are the field that most changes what the copy argues.
- Then the keywords, since they decide whether the ad sounds like an insider or a visitor.
- Then the tone notes, if the drafts keep coming out brighter than your groups would tolerate.
- Leave the angles alone unless one is plainly wrong, since they are what keeps a batch varied.
From audience to somewhere to post
Research that ends in a document is research that was a waste of an afternoon. A segment here has two exits, and both are one click.
- Generate ads for it. The copy stage takes one audience at a time and writes a batch on distinct angles, with spintax and first comments already in place. That pipeline is covered in the AI Facebook ads guide.
- Find its groups. The finder takes the segment's own search terms and returns candidate Facebook groups, which you accept or drop, saved as a reusable list. See the group finder guide.
The link works in both directions, which matters more than it sounds. If you already have a group list and no idea how to describe the people in it, point a generation at the list and it reads the groups and writes the audience. Either door, same pairing at the end.
Audiences are saved, not rerun
One audience costs one unit, once, when it is written. After that it belongs to the business and is free to use forever: generate ads against it in January and again in June without paying for the research twice.
That is why the meter has two units rather than one. Ad creatives are consumable, because each one is a fresh render. Audiences are an asset, because the research does not spoil.
How to tell a bad segment fast
Four questions, thirty seconds each, and they will kill about a third of any batch.
- Could I name three real people in it? If not, it is a category rather than a segment.
- Does the pain point describe a moment? Wants better value is not a moment. Just got quoted twice what they expected is.
- Would the objection actually be said out loud? Real objections are embarrassing and specific. Generic ones mean it was guessing.
- Do the search terms return real groups? Run the finder. If nothing comes back, those people do not gather here, whatever the fit score says.
Using segments outside this tool
Nothing about a segment is locked to group posting. The same record is a perfectly good brief for cold email, a landing page, a paid campaign or a script for the person answering your phone, and the fields that matter most for those are the objections and the keywords.
If you take one habit from this page, take that one: whatever you write next, write it to one named segment rather than to everybody. It is the cheapest improvement available to any small business, and it costs nothing but the decision.
Still stuck? Mail us and a person will answer.
Add to Chrome