Product discovery is the process of deeply understanding users’ problems in order to determine the best ways to solve them. It’s also the very first phase of the AI product development lifecycle, which is a cyclical framework for conceptualizing, building, launching, and iterating on software products. But while discovery is one of the most important parts of building a great product, it often flies under the radar.

AI can help product teams make sense of product discovery evidence: what customers ask for, where users struggle, and which questions still need an answer. It can summarize feedback, compare research notes, and suggest explanations to investigate. The value comes from connecting those suggestions to real customer evidence, not treating a generated answer as research.

Here are five ways to use AI across that process without losing the source behind a finding.

Five ways to use AI for product discovery

Start with a specific question and a defined audience. Use an approved source of customer evidence, ask AI to show its supporting records, and keep a human owner for the decision. The same discipline applies for product builders whether you work in Pendo or another AI workspace.

1. Understand a customer problem with real research

Use AI to organize interview notes and research you have permission to analyze. Ask it to identify the task the customer was trying to complete, the obstacle they described, and the evidence behind each finding. A general explanation of a persona can help you prepare interview questions, but it cannot tell you what your own users need.

Keep the original source attached to every theme. If a summary says onboarding is confusing, check which participants said it, which workflow they meant, and whether there are examples that disagree. Use those gaps to choose the next customer conversation.

2. Connect feedback themes with product behavior

AI can help group feedback into themes, but frequency alone does not establish priority. Check who raised the issue, which audience they represent, and what those users actually did in the product. A repeated request from one account should not be counted as the same breadth of demand as requests from many independent accounts.

Use Pendo Listen to bring feedback into one place and examine themes with customer context. Then use Pendo Product Analytics to investigate the related workflow. Define the eligible audience and time window before comparing adoption, drop-off, or repeat use. Feedback helps explain a possible problem; behavior helps you judge its scope.

3. Turn an observed problem into testable options

Once the evidence points to a problem, ask AI for several ways to address it. Include an option that changes the existing experience without adding a feature. For each option, spell out the assumption, the smallest useful test, and the result that would make the team stop.

For example, if new users abandon setup when they reach a permissions step, possible explanations include unfamiliar language, missing access, or a task they do not need yet. An onboarding guide may help with the first explanation but do little for the others. Keep the competing explanations visible until research or an experiment distinguishes them.


4. Compare current competitor evidence

Use AI to summarize a defined set of current competitor product pages, documentation, and release notes. Ask it to return the source URL and review date beside each capability claim. Separate an advertised capability from evidence that the capability is available in the plan or environment your customer would use.

Competitive research can reveal alternatives and questions to investigate. It does not establish demand for your product. Bring the findings back to customer evidence: which problem would a different approach solve, and how would you validate that it matters?

5. Share a decision brief that is candid about any uncertainties

Ask AI to turn the discovery work into a short decision brief: the customer problem, the audience, the evidence, competing explanations, the proposed test, and the owner. Keep the source links and the reason for the recommendation. A concise summary should make the decision easier to inspect, not hide the uncertainty behind a confident sentence.

Tailor the emphasis to the audience. Engineering needs the workflow and technical assumptions; design needs the user’s task and context; leadership needs the decision, expected outcome, and remaining uncertainty. The underlying evidence should stay consistent across versions.

Getting started with AI for product discovery

As you start leveraging AI to bolster your discovery efforts, it’s important to keep some key risks and challenges in mind. These include:

Assuming AI can replace humans: Keep a human owner for the research and the decision. Use AI to organize evidence and suggest questions, then check its conclusions against the underlying records and conversations.

Convincing AI hallucinations: A request not to invent information is not enough. Ask for supporting records, inspect a sample of summaries against their sources, and require the tool to identify missing evidence. Treat an answer without support as a hypothesis to investigate.

Legal and compliance: Follow your organization’s approved AI tools and data-handling policies. Confirm the relevant service terms and settings before submitting customer information, and retain the records needed for review. Do not assume that the same ownership, retention, or training rules apply to every provider or plan.

Unknown depth of knowledge: Ask which sources support the answer and how current they are. Define product-specific terms and acronyms in the prompt. If the tool cannot access the relevant research or product data, state that gap rather than treating a fluent answer as a complete view.

Bias: Check whose feedback is represented and whose is missing. Compare findings across relevant roles, plans, account sizes, and access levels. Use interviews and product behavior to test a pattern before generalizing it to your full customer base.

Start with one discovery question and one decision you need to make:

  • Choose an approved source of feedback or research.
  • Define the audience and the product workflow.
  • Ask AI for source-linked themes and competing explanations.
  • Check the pattern against captured product behavior.
  • Validate the next step with affected users or a focused experiment.
  • Record the result and revisit the hypothesis before committing to a larger build.

Ready to connect what customers say with what they do? Explore Pendo Listen and Pendo product analytics to bring feedback and behavioral evidence into your next discovery decision. Even better if you want it to meet you where you are at use the Pendo MCP!