How Discovery works

NicheIQ research has two phases. Discovery reads public discussion to find recurring problems, the audiences behind them, and a ranked set of candidate products. You then choose one to three exact candidates for one Deep Research run, which investigates that saved scope in more depth. (See How selected ideas get pressure-tested.) This page is about Discovery.

The five Discovery steps

The work runs in order, and each step feeds the next:

1. It frames the niche and the language that distinguishes it.
2. It searches public discussion, including Reddit and Hacker News when relevant and available.
3. It extracts recurring problems and keeps source-linked evidence for them. (See How pain points are gathered.)
4. It maps the audience segments and communities represented in the captured sample.
5. It generates, checks, and ranks a varied set of candidate products. Grounded guidance may help you read the current pool, but it never changes the scores. (See How ideas are generated.)

Understanding your niche

The first step interprets the input rather than simply repeating it. A market such as "home espresso gear" and an audience such as "freelance designers" need different search plans. When the input names an audience, the research can still map the wider market around it so the search does not collapse onto one narrow phrasing.

The framing also identifies signature terms, products, communities, and exclusions that separate the niche from look-alike topics. These anchors help later searches stay relevant. If the input is too broad to support a reliable distinction, the workflow can ask for a clearer scope or proceed with explicit caveats rather than inventing specificity.

The "idea," "audience," and "discovery" entry points are starting hints. The system still evaluates the text itself and stores the resulting framing in the run.

Reading the evidence

Discussion counts, source counts, and charts describe the captured sample, not the whole market. Reddit votes and Hacker News points are platform-specific engagement signals; they are not interchangeable proof of demand. Pain evidence should remain linked to its captured source, while missing or failed-to-load data should be shown as unavailable rather than as a finding of "none."

What it can't do

Public discussion is self-selected and time-sensitive. Quiet buyers, private communities, and offline workflows may be underrepresented. If the run cannot collect enough relevant material, it may stop or return a thinner result with caveats instead of filling gaps with invented evidence. The niche framing is also an interpretation: if it differs from what you meant, rephrase the niche before starting another run.