Customer research methods: what each one answers, and where public conversations fit

One row won't get you there by itself. Good research pairs at least one "why" method with one "how many" method, then checks both against what people say in public.

This is a 30 to 45 minute conversation with someone who recently made a decision. Ask about the last time they tried to solve the problem: what they used, what went wrong and what made them choose. Past behaviour tells you more than predictions about what they'll do.

Use interviews when you need the reasons behind a real decision, the order things happened in and the words the buyer used. They break down when you ask leading questions, when you only talk to happy customers, or when you treat five interviews as a market.

A survey sends one fixed set of questions to many people. Use it for sizing: what share of your audience has a problem, which feature matters most, which channel they use.

It works when you already know the likely answers and need to count them. It breaks when you wrote every answer option yourself, because people pick from the list even when the right answer isn't on it.

You watch someone try to complete a task in your product or prototype and see confusion nobody would mention in an interview.

Use it when you're changing a flow and need to know where it fails. It breaks when the task is artificial: a person told to "find the export button" behaves differently from one with a deadline.

Event data shows what people click, skip and abandon. It's the most honest record of behaviour you have.

Use it to find where the funnel leaks. It breaks when you read motive into a number: a drop-off tells you where people left, never why.

Read support tickets, chat logs and call recordings. People describe problems in their own words when they're annoyed enough to write them down.

This gives you a ranked list of recurring problems. It fails when you only hear from the loudest customers, since people who quietly left don't write tickets.

These are interviews or notes on why recent deals were won or lost. For the short version, see win-loss analysis. Of all the methods, it gets closest to a direct answer to "why do people choose us or someone else".

Use it once you have closed enough deals to see a pattern. It breaks when the person who ran the deal also runs the review, because buyers soften their answers to them.

People ask each other for help on Reddit and X every day, in their own words, with no researcher in the room. Reading those threads on purpose is a research method. Its one advantage over every other method is that nobody is performing for you.

Use it when you need the words buyers use, the problems they raise first, the alternatives they name, and what a bad day looks like for them. It breaks when you treat a thread as a statistical sample. Posts come from people who write posts, and that is only a slice of your market.

1. Define one question, for example: "What do bookkeeping clients ask for when they look for help?" Without a question, a study turns into scrolling.
2. Pick three to five communities where your buyers ask for help. Our subreddit rules tool shows what each one allows.
3. Collect 30 to 50 posts. Search with the phrases people use, not your own. The Reddit search guide and the X search operators cheat sheet show how.
4. Copy the exact sentences. Save the line that shows the need, with the date and the community, and remove usernames.
5. Group them into themes: problems, triggers, alternatives mentioned, and constraints like budget or team size.
6. Record the pattern, not the anecdote. "Eleven of 40 posts mention reconciling payouts" is useful. One vivid quote on its own isn't.
7. Check it against another method. Take your three biggest themes to interviews or the support inbox and see whether they hold.

In our own scans on 16 October 2026 (public Reddit posts from the previous 12 months, up to 100 judged per niche), people asking for software described their needs in sentences like these, paraphrased and with identifying details removed:

• "I need a tool that automatically detects and removes personal information from documents before sharing them."
• "I need software that automatically fills repetitive job application information across different websites."
• "I need a tool that tracks my subscriptions and helps me cancel unwanted services."

Each has a verb for the job ("detects", "tracks", "fills"), a condition ("before sharing", "across different websites") and an outcome. When you see that structure, it names the job people hire a product to do, which is what jobs to be done research is after.

A post being public doesn't mean using it has no consequences. These rules keep the method honest:

• Remove names and handles from anything you share internally or publish.
• Skip sensitive topics. HearBuyers doesn't publish posts that reveal health, sexuality, religion, politics, immigration, crime, death or family crises, and a study you share shouldn't either.
• A public post isn't consent to contact the person who wrote it. Reading to learn is one thing; messaging people is another.
• Label the sample for what it is. "Forty public posts from four communities in October 2026" is honest. "What customers think" is not.
• Expect change. Reddit's public data access is changing between 31 October 2026 and March 2027, so some tools that read Reddit may cover less during that window. Searching Reddit by hand still works.

Ask three questions, in this order:

1. Do I need the reason, the count or the wording? Interviews give you reasons. Surveys and analytics give you counts. Public conversations and support tickets give you wording.
2. How fast do I need it? Reading public conversations or reviewing tickets takes days. Interviews can take weeks just to schedule.
3. What would change my mind? Pick the method whose result could contradict what you believe.

For a list of software that supports each method, see our guide to the best customer research tools. To turn what you read into a ranked list of problems, see customer pain points.

Early-stage founders often run interviews as customer discovery calls, before there is a product to test.

Surveys and interviews are the most common. Plenty of teams also learn a lot from support tickets and sales calls, even if they don't call it research.

There's no fixed number. Keep going until new interviews stop teaching you anything about your question, then check what you heard with a second method.

Yes, as one input. It shows you the language and problems of the people who post. People who stay silent won't show up, so check it against interviews or analytics.

You can, but keep the two jobs separate. For research, strip out names and look for patterns. For outreach, reply in the thread where the person asked.

Write down one research question your analytics can't answer. This week, pull 30 public posts about it, sort the sentences into themes, and bring the top three to your next customer call.