Find your next campaign message.

Start with the decision at the end of the work: which customer problem should lead the landing page? Describe the product in ordinary language, name the market, and state what the product can actually deliver. A delivery area matters more here than an invented profile such as a particular age, income, or household type.

For North Table, compare three possible needs: making weeknight preparation easier, finding more meals to choose from, and understanding what the box costs. Ask the agent to explore these possibilities without assuming any one is the winner. Name a marketing lead as the person who will decide which message proceeds to testing.

In Integrations, open Provided by Taploop and find Public web research. Confirm that Web search and Read public pages are available. If the page says Setup required, your workspace needs that setup before the agent can use the tools. An available connection should be selected before you continue.

Open AI setup and choose Set up access to my integrations. In Agent access, include research.execute in Allowed Taploop scopes and select the research capabilities you need. Follow that panel's server address and secure token instructions. Ask the agent to confirm those research capabilities and source readiness first. Your workspace owner can use Budget controls to set a research limit; paid lookups should stay inside the agreed scope.

Ask for a focused first pass across relevant provider pages, product FAQs, and other accessible sources. Each finding should include the page, the observation, and the question it helps answer. Several pages from one company still represent one company's perspective. A supplier's own positioning is useful evidence of its offer, but does not establish what customers prefer.

For example, HelloFresh describes pre-measured ingredients, recipe selection, and the ability to skip deliveries on its US How it works page. Those are documented product claims. Treat the table below as a demonstration of how to record them, not as proof that North Table's future customers share particular needs. [1]

A hypothesis is a statement your team intends to test. Here, a reasonable first hypothesis is that reducing the work of planning dinner may be more useful to communicate than simply offering another subscription box. That is an interpretation of the example evidence, not a measured preference or a finding about the whole market.

Ask for one message per need, plus the product facts needed to support it. Example options are 'Less planning. More cooking.', 'A new dinner idea for your week.', and 'See what is in your weekly box.' The value option should explain the actual offer; claims about savings, lower prices, or nutrition need supporting evidence.

Bring the audience brief and message options to a short team review. For the fictional company, propose testing the planning message against the current headline while keeping the offer and main action consistent. Define the action that would count as useful progress, such as a visitor requesting delivery availability, rather than treating every page view as purchase intent.

The agent's deliverable is an experiment plan: the hypothesis, copy, audience context, measurement requirement, owner, and decision rule. Your team still needs to approve the claim, implement the page, and verify tracking. Ask the agent to identify what would make the experiment inconclusive, including too few relevant visits or inconsistent tracking.

Ask the agent for a short final brief containing the research question, strongest observations, competing explanations, recommended message, and links. Keep the original evidence table beside the recommendation so another teammate can inspect the reasoning. Taploop's Runs and evidence gives you a place to review the activity recorded for the work.

Finish by recording a decision: test the proposed message, gather more evidence, or stop because the product cannot support the claim. Public research can narrow a question and prepare useful work. The next evidence should come from the company's actual customers or a properly measured experiment.