AI Product Research: Compare Markets With Checked Evidence

AI product research uses an assistant to organize evidence about customers, competitors, and possible products, while a person checks the sources and decides what the evidence supports. Start with a specific question, collect original sources, and ask AI to compare only what those sources establish. A fluent market summary is a starting point for investigation, not proof of demand.

Tools like fenbs keep checked reference links and product decisions outside a single chat, so another assistant can continue the research without losing its evidence trail. The useful record says what you know, where it came from, and what remains uncertain.

How do you start with a question you can actually answer?

1. Choose a user, a recurring problem, and the decision the research must inform. For example: should a small studio test a client-approval tool for independent designers?
2. List the claims that would change that decision: how approval happens today, where it breaks down, and what someone has already tried.
3. Collect original sources and record their scope. Keep customer interviews separate from vendor descriptions and population data.
4. Ask AI for a comparison with a source beside each factual claim. Mark missing evidence explicitly.
5. Choose a small next test, name its owner, and record what result would make you continue, change direction, or stop.

Make the question smaller than “What should we build?” Ask whether a particular group experiences a particular problem often enough to investigate. You can research that question without pretending to know the size of an entire market or the future success of a product.

Set an initial research boundary: a user group, geography, use case, and period covered by the evidence. If you expand that boundary later, say why. Otherwise the assistant may combine enterprise buying behavior with solo-business habits and present the mixture as one coherent customer.

Which sources belong in the evidence log?

Use a source for the question it can answer. A vendor help page can establish how a documented feature works; it cannot establish that buyers love it. An interview can describe one person’s experience; it cannot establish how common that experience is across the United States.

For US demographic and economic context, the Census Bureau’s data resources↗ (opens in a new tab) are a useful place to find original datasets. For public-company disclosures, SEC filing search↗ (opens in a new tab) provides access to company reports. Check the specific table or filing before using a number, and retain its date and definition.

For each source, save a descriptive title, URL, publication or update date when available, access date, and the narrow claim it supports. Note whether you actually opened the underlying material. A search preview or an AI-generated citation is a lead until you have checked the page.

How can AI compare competitors without filling the gaps?

Define the comparison columns before collecting answers. Use observable questions such as whether a documented workflow supports a reviewer returning changes. Avoid vague ratings such as “best collaboration” unless you have defined and tested what that means.

Give the assistant a restricted set of sources and tell it to distinguish documented, observed, and unknown. If a feature is absent from the pages you read, write “not found in these sources.” Do not turn that into “the product cannot do it.” That stronger claim requires more evidence.

• Keep each factual cell attached to the page or observation that supports it.
• Compare the same task across products rather than combining unrelated marketing claims.
• Leave current commercial terms behind a link to the vendor’s own page.
• Flag conflicting documentation for review instead of asking AI to choose whichever sounds more plausible.

Open every source used for an important conclusion. NIST’s generative AI profile↗ (opens in a new tab) identifies confidently false output as a risk and recommends checking sources and citations. Your comparison needs that check even when its prose sounds cautious.

Keeping the checked URLs in fenbs Links and listings lets the next assistant see which evidence informed the project, rather than reconstructing a competitor comparison from memory.

How do you separate a finding from an interpretation?

Write the observation first, then your interpretation, then the remaining question. In a hypothetical studio interview, a participant might describe collecting approvals through several email threads. That supports a finding about this workflow. The idea that a shared approval page would solve it is a product hypothesis.

Ask AI to produce alternative explanations. Perhaps the problem is unclear authority rather than missing software. Perhaps a current tool would work with a simpler process. Research becomes more useful when it can disprove your preferred solution, especially before you spend time building it.

Keep counts honest. If you interviewed four people, record what each said and how they were recruited. Do not let an assistant convert those conversations into a percentage of all designers. Preserve contrary examples because they may reveal a narrower audience or a more important distinction.

What turns research into a product decision?

A useful decision names an action, the evidence behind it, the alternatives considered, and the person choosing. It also names what could change the choice. “Test a paper prototype with independent designers before building account setup” is actionable; “the market looks promising” is not.

Choose the smallest test that resolves the largest uncertainty. If you do not know whether people understand the proposed workflow, show a prototype. If you do not know whether they experience the problem, return to observation. A more polished AI report will not answer a question that requires contact with users.

The wider AI brain guide explains why this record should survive the conversation. Keep the research outcome with the work it changes, including the assumptions you rejected and the evidence you still need.

How to do this with fenbs

1. Create a free account, then add a project named for the opportunity you are investigating.
2. Add checked sources to Links and listings. Use AI context notes for the question, scope, findings, and explicit limits of the evidence.
3. Create tasks for interviews, source checks, and the next experiment. Keep each task’s plan specific enough for someone else to continue.
4. Record the person’s chosen action in Decisions and rules. Leave unresolved choices open; assistants can record a person’s decision but cannot decide it.
5. Before stopping, save Where we left off with the checked findings, outstanding questions, and the next task. Ask the next connected assistant to read that context first.

fenbs is free for one person, no credit card. It does not conduct interviews or verify web claims for you, and it is not the right fit as a repository for confidential interview recordings. Keep sensitive source material in an approved system and save only the project context you are authorized to share.

For more practical patterns, browse AI at work and how to use AI at work. Start with one checked research question and one next experiment; expand the record when the work produces something worth keeping.