How an experiment gets designed and run

You have a question about your market. Which benefit makes buyers act? Which of three product ideas is worth building? Does your positioning still hold in a new country? A Heatseeker experiment puts that question to your audience and brings back an answer you can act on. Here is how you get from question to answer, what you decide along the way, and how to read what comes back. If you want to know why a market test can be trusted in the first place, start with How a market test works.

What you bring

An experiment is built from things you already have in your workspace.

• Your product. The thing your question is about. Its name, description and selling proposition shape every question Heatseeker asks and every ad it writes. If you have several products, you say which one this is about.
• Your brand. The ads run under your brand's page and in its tone of voice. If you also have a stealth brand, you can ask the market incognito, so competitors never see who is asking.
• Your audience segments. The people whose answer you want. Your saved segments are offered to you once the question is set. If none fits, you create one on the spot.
• Your ad accounts, for live runs. Live experiments show real ads to real people on Meta, LinkedIn or TikTok, so those accounts need to be connected first. See Connect Meta to Heatseeker, Connect LinkedIn to Heatseeker and Connect TikTok to Heatseeker. Simulations need no ad account.

Step 1: Turn your question into a goal

Open a new experiment and say what you want to know, in your own words. "We think the safety angle beats the price angle with parents." "We have six benefits and no idea which one our customers actually care about." "We are entering a new market and are not sure our positioning travels."

Heatseeker AI asks one question at a time, and only to pin down three things:

1. What you want to compare. The one thing that will differ between the ads: headlines, benefits, pain points, images, products.
2. Why now. The problem, opportunity or disagreement that makes this the question worth answering.
3. What you will do with the answer. In your words: brief the next campaign, pick the launch message, decide which product to build.

If you have a hunch, say it. It is recorded as your hypothesis, and the result proves or disproves it. Heatseeker never invents one for you. If you already know exactly what you want, say "skip the questions" and it states its assumptions and asks you to confirm them in one turn.

You do not have to choose an experiment type. The experiment library shows all thirteen, but Heatseeker recommends two or three that fit your question and explains what each one compares. Your job is to recognize your question in one of them.

Once your goal is confirmed, two more choices close the step.

Choose who answers

Your saved audience segments appear as cards, the best fit for your question first. Pick one, or create a new segment. You can also choose more than one. Each segment then gets its own answer, which is how you tell a strong message apart from a well-matched audience. Experiments with more than one segment run live.

Choose how they answer

Your segment can answer in two ways. Live, where real people in that segment see real ads and their clicks are the answer. Or as a simulation, where the segment's AI customers react to the same ads and answer in minutes, at a fraction of the cost.

Heatseeker recommends one for you and says why. It looks at what rides on your answer: if the result picks the message for a major campaign, it recommends live, because a wrong answer there is expensive. If the result only narrows twelve ideas to three, a simulation is enough. It also looks at how well your AI customers are trained. They learn from every live experiment you run for this product. If they have not been trained yet, Heatseeker recommends live and treats any simulation as a directional read.

You can override either way by saying so. TikTok experiments and experiments across more than one segment always run live.

From concept to campaign

Heatseeker asks the market the same way at every stage of a product's life: an idea before it is built, a proposition before it is positioned, a campaign before it is funded. The questions get more specific as you move along, and the biggest return is at the start, where a wrong call costs the most and shows up last. Thirteen experiment types cover the journey, and the method underneath is the same one throughout: real people in your segment, split at random, counted on what they do.

Which idea? Before you build

The Concept Test, Product Demand Test, Feature Test, Jobs-to-be-Done Test and Pain-Point Test. Which concept earns attention, which product your segment reaches for, which feature matters, which job or pain is worth solving. These are the questions most teams still answer with a survey, and they are where a market test changes a roadmap rather than a headline.

Which promise? Before you position

The Value Proposition Test, Benefit Test, Buying Driver Test and Competitor Benchmark. What to lead with, which benefit moves buyers, what actually drives the purchase, and how your claim fares against a competitor's. This is where most teams start, because everything downstream inherits the answer: the campaign brief, the pack copy, the sales deck.

Which execution? Before you spend

The Messaging Test, Image Creative Test, Video Creative Test and Full Creative Test. The words, the visual, or the whole ad, once the promise is settled. This is where market testing meets the media plan, and where the A/B tests you already run pick up.

Under the hood: four shapes

What makes every answer clean is what the ads hold constant. The thirteen types reduce to four shapes:

• One image, the words differ. The message shape, used by the Value Proposition, Messaging, Benefit, Feature, Pain-Point, Buying Driver, Jobs-to-be-Done and Concept Tests and the Competitor Benchmark.
• Your images or videos, one copy line. The Image Creative and Video Creative Tests, so people react to the visual and not the words.
• Whole ads, head to head. The Full Creative Test. The winner is a direction, not a single element, so Heatseeker offers a follow-up that isolates why it won.
• Your products, one script. The Product Demand Test. Each ad shows a different product on one copy template where only the product name changes.

Step 2: Approve the ads

Heatseeker drafts the ads for you. Each one makes a single, distinct case. If two ads cannot be told apart in one sentence, the test produces noise, and Heatseeker says so before you launch.

Where the ideas come from. If you know which angles you want to test, those are the ads, in your words. If you are not sure, Heatseeker draws on what has already worked for you: your past experiments, the research in your Data Hub and your current market signal. It tells you where each angle came from, and says plainly when it has nothing to go on.

What gets written. Every ad gets a short hook on the image, an optional second line beneath it, a headline where the platform shows one, and the body copy. It is written for your segment, about your product, in your brand's tone of voice, and within each platform's length limits. Any words you supply run exactly as you wrote them.

Images and video. For message tests, Heatseeker generates one shared image, or uses one of yours. For visual, concept and product tests, you upload one creative per ad. You can attach an image in the chat and ask for it to be used or edited. Video is always yours: Heatseeker does not generate it. A Video Creative Test takes one video per ad and runs live on Meta. On TikTok every experiment runs on your uploaded videos, one shared video for a message test or one per product for a product test, and a video experiment cannot be simulated.

How many ads. Two is the minimum, ten the maximum, and three to five is where most teams land. More ads means more breadth, not a weaker answer. Each ad shares the same budget, so a wider set needs more spend or a longer run to give every ad enough exposure to separate.

Making changes. Edit any ad directly, or tell Heatseeker what to change: "make the second one punchier", "swap the background", "make it brighter", "split that headline over two lines". Ask for a rewrite and it first tells you where it thinks the lift is, then offers directions.

For live runs you also say where the ad sends people: your landing page, or a lead form with your privacy policy on Meta and LinkedIn. TikTok ads send people to a website only.

Nothing goes to market without your sign-off on the ads.

Step 3: Set the budget and launch

The review step is short. The thinking is done.

The launch checklist. Heatseeker will not launch an experiment that is not ready. Anything still missing is listed on the draft, and Launch stays disabled until every item is in place: your product, your brand, the goal, the method, the segment, live or simulation, at least two approved ads and at most ten, an image on every ad, where the ads send people, a budget for a live run, and legible terms text if you added any to the creative.

Budget, for live runs. Heatseeker proposes a maximum budget sized for your segment. Change it with the slider or by naming a figure. The budget is a cap, not a target: the experiment stops spending as soon as the answer is decisive. A lower budget may give you a readable lean without a decisive answer. On Meta and TikTok the review also shows which ad account is charged.

Run length. Heatseeker recommends a duration for the platform and the size of your segment, typically about a week on Meta and seven to nine days on LinkedIn, and you can tweak it the same way you tweak the budget. The experiment can finish early once the evidence is decisive, and a completed test can still come back Inconclusive.

Launch. The summary lists your product, the method, the goal, the ads and the budget. You click Launch. A simulation has no budget step: its summary appears as soon as you arrive, and the answer follows within minutes.

Step 4: Get your answer

Results open on the winning ad. Every ad is scored on what people did with it: a click counts more than a like, a form submit more than a click, and everything is normalized by how many people saw it, so the winner is the message and not the budget.

The answer card shows Signal, with four levels describing how conclusive the finding is. The accompanying advice explains what to do next:

• Decisive. The evidence clears the decision threshold for the tested choice. Make the supported choice within the tested goal and population; it does not mean certainty.
• Strong. The leader is clear but has not cleared the decisive bar. Directional, and often enough when you are narrowing a field.
• Indicative. A useful provisional finding, with substantial uncertainty. Use it to guide a confirming test.
• Inconclusive. The evidence does not support a dependable conclusion. A completed test can be inconclusive.

A simulation never shows Decisive, however clearly its votes separate. Strong is its ceiling; a panel with limited live evidence may be capped lower.

If you asked more than one segment, each gets its own winner. Within a segment, results can be split by demographics on Meta and by firmographics on LinkedIn, so you can see which message landed with whom. Every figure is drillable to the underlying responses.

The full read of the results page, from the winner card and the Signal to the by-segment heatmap, is in How to read your results. Ask Heatseeker AI about your result on the results page, or from your own assistant through Connect to Heatseeker with MCP. Heatseeker also proposes the next move: a follow-up that isolates why a Full Creative winner won, a live run to confirm a simulation, or the next question on your roadmap. And every live experiment trains your segment's AI customers, so your next simulation is a little sharper.

Changing your mind

You can go back at any point before launch. Change the question, the segment or the ads, and Heatseeker takes you to the right step. If ads already exist when you change the question or the segment, it warns you once that they may need reworking, then makes the change. Switching to a different experiment type is a change to the question, so it happens in the goal step rather than by rebuilding the ads in place.

What next

Once the experiment has run, How to read your results walks through the results page: the winner card, what % of Winner and the Signal mean, and how to read the heatmap of who responded. If you want the reasoning behind the method itself, start with How a market test works.