Go back to research when the team cannot clearly explain who the buyer is, what situation triggers action, what outcome matters, what objections or constraints shape the decision, or why the current message should be credible. Test when that customer model is coherent and the remaining uncertainty is which supported message, structure, proof, or offer expression performs better. Research resolves uncertainty about the customer. Testing resolves uncertainty about execution in market.
The choice becomes easier when you name the uncertainty.
If the team is still guessing about the buyer, the trigger, the problem, the desired outcome, the objection, or the language customers use, the message needs more research.
If those foundations are reasonably clear and the remaining question is whether one supported headline, angle, structure, proof sequence, or offer presentation works better than another, the message is ready for a test.
Research helps you decide what deserves to be said.
Testing helps you learn how well a particular expression of that understanding performs in the market.
The mistake is using one to compensate for the absence of the other.
Research when the customer model is still fuzzy
A team often says it has a "copy problem" when it actually has an understanding problem.
The symptoms are familiar:
- the target customer is described mainly through demographics or job title;
- several people on the team give different answers about the buyer's main problem;
- the message lists features because nobody can explain the decision context;
- objections are based on internal guesses;
- customer language in the copy comes from competitors rather than customers;
- every new campaign starts from a new theory of what the market wants;
- the team cannot explain why a person would act now rather than later.
Michael Haynes's Genius Talk research approach is helpful here. In his B2B work, he emphasizes understanding the industry, geographic context, and individual decision-maker through detailed conversations with customers. Their value comes from the level of context they can uncover, rather than from the interview length itself.
A buyer's title does not tell you what happened this week that made a problem urgent.
A broad category such as "wants growth" does not tell you whether the person needs more leads, a stronger conversion path, better retention, or confidence that a proposed investment will survive internal scrutiny.
When the team cannot answer those questions with evidence, another copy variation is unlikely to solve the deeper problem.
Look for missing trigger, outcome, objection, and proof
Before commissioning more research, locate the missing part.
A practical message model needs at least four things.
Trigger: What situation makes the buyer pay attention now?
Outcome: What change are they trying to create?
Obstacle or objection: What makes the decision difficult, risky, expensive, or easy to postpone?
Proof: What would make the promise credible in this context?
If one of those is based mainly on internal opinion, research has a job.
David Fradin's Genius Talk product perspective adds an important discipline: observe what customers do as well as what they say. In his work, customer use sometimes differed from the company's assumptions about how a product should be used.
That matters for messaging because a customer may describe an ideal preference while behaving according to a different priority.
Research can therefore include customer conversations, observed product use, sales-call patterns, support questions, reviews, search behavior already available to the business, or other evidence appropriate to the decision.
Collect only the material needed to resolve the specific uncertainty preventing a coherent message.
Test when the customer model is coherent
Testing becomes useful when the team can explain the customer situation with enough confidence to create competing expressions of the same grounded idea.
For example, you may know:
- who the offer is for;
- what event creates the need;
- what outcome matters;
- the main objection;
- which claims you can support;
- what action you want the person to take.
The uncertainty may now be narrower.
Should the page lead with the outcome or the problem? Does a customer example make the idea easier to trust? Is a shorter explanation clearer? Does one offer presentation create more qualified responses? Does the same positioning land better when expressed in the customer's language?
Those are testing questions.
Sarah Sal's Genius Talk approach creates a useful bridge between understanding and response. She describes using direct customer research to change messaging, while also emphasizing a sequence in which an offer first demonstrates conversion, customer acquisition then needs to make economic sense, and scale comes later.
Her sequence should be treated as her marketing framework, not as a universal law. The relevant principle here is that market behavior supplies information an interview cannot.
A customer can tell you a message sounds good. A live test can show whether the message contributes to the action you care about.
Do not test a message you cannot explain
A/B testing can look rigorous while avoiding the real question.
Suppose a team tests five headlines, each built on a different assumption about the buyer.
One emphasizes price. One emphasizes speed. One emphasizes status. One emphasizes risk. One emphasizes ease.
If the team does not know which of those concerns is grounded in customer evidence, the test mixes message discovery with execution testing.
A result may tell you which version performed better in that setting. It may not tell you why, whether the difference will repeat, or whether the winning message attracts the right customer.
Bobby Gillespie's Genius Talk branding perspective is useful as a guardrail. He combines consistency with continuing evaluation. A brand can keep its core stable while still letting evidence challenge a weak expression of that core.
Before running a message test, be able to state:
We believe this because of this customer evidence, and this test is designed to resolve this remaining uncertainty.
If you cannot finish that sentence, research may still be missing.
Analytics can show where to investigate, not automatically why
Philippa Gamse starts website analytics with the business objective.
Her perspective matters because a message should be judged against the action it is supposed to support. A higher click rate is not automatically better if the business needs qualified inquiries. More page engagement is not automatically useful if the reader still cannot understand the offer or take the next step.
Analytics can reveal patterns such as:
- people leave before reaching the proof;
- one traffic source behaves differently from another;
- an important action is rarely completed;
- a page section is not being reached;
- a particular offer creates response but weak downstream quality.
Those signals can justify a test or send the team back to research.
They do not explain motive by themselves.
A drop-off may reflect message mismatch, technical friction, weak proof, irrelevant traffic, price, timing, or something else. The data tells you where the problem appears. Customer evidence helps explain what may be happening there.
That is why research and testing work best as a loop rather than rival methods.
Be careful with low traffic
A message can be worth testing even when traffic is limited, but the kind of claim you can make from the result changes.
The approved Genius Talk synthesis on conversion makes this distinction carefully. Tightly isolated tests may take a long time to produce a useful signal on a low-traffic site. Larger page or offer changes can create more visible directional evidence, but they also make it harder to know which element caused the change.
So do not force false precision onto a small sample.
With limited traffic, consider questions such as:
- Can we test the message in sales conversations first?
- Can we use customer interviews to check comprehension?
- Can we expose a larger, meaningful change rather than tiny copy differences?
- Can we collect qualitative evidence alongside behavior?
- Is the cost of waiting for a cleaner test greater than the value of the decision?
The goal is decision quality, not the appearance of scientific certainty.
Use a simple decision tree
When the team is unsure what to do next, use this sequence.
1. Can we name the buyer and the buying situation?
If not, research.
2. Can we explain the trigger and desired outcome in customer terms?
If not, research.
3. Do we understand the main objection or constraint?
If not, research.
4. Do we have support for the promise we want to make?
If not, research or narrow the claim.
5. Are those foundations coherent, but we have two plausible ways to express them?
Test.
6. Is there enough traffic or response volume to learn from the proposed test?
If yes, design the test around the business outcome. If not, use a lower-cost directional method and keep the conclusion modest.
7. Did the result create a new question about why customers behaved that way?
Return to research.
This loop prevents the team from treating research as a one-time phase at the beginning of marketing.
Market response can expose a new uncertainty. That uncertainty can become the next research question.
Know what each method can tell you
Research is strongest when you need context.
It can uncover the words customers use, the sequence of the buying situation, the alternatives they consider, the objections they raise, and the outcome they are trying to reach.
Testing is strongest when you have a concrete execution and need behavioral evidence about performance in a defined environment.
Neither method answers everything.
An interview cannot prove that a message will convert at scale.
A click cannot explain the full reason a person clicked.
The useful discipline is to stop asking, "Should we research or test?" as though one method is more sophisticated.
Ask instead:
What do we still not know?
If the unknown lives inside the customer's world, research it.
If the unknown lives inside a grounded execution in market, test it.
Then let the answer determine the next question.