Test a product idea against real evidence

There's an idea you keep coming back to: a subscription app that plans a week of meals for people with diabetes, say, starting in India. Before you spend three months on a prototype, you want to know whether people need it, who already does it and what could sink it. You could spend a weekend in search results and app stores and still not know which of your notes to trust. This page shows how to get a first, sourced read on an idea in one check, what that check said about this exact idea, and where it stops and customer conversations have to take over.

Why ideas fail before they get going

Most products that fail weren't impossible to build. CB Insights analyzed 431 venture-backed companies that shut down since 2023. Among those with an identifiable cause, 70% ran out of capital, which the analysis calls the final cause of death rather than the root problem. Poor product-market fit came next, in 43%.

Shutdowns are common enough to justify an afternoon of homework. Carta counted 966 shutdowns among US startups on its platform in 2024, up from 769 in 2023, and its own head of insights said the real number is likely higher.

Neither figure says your idea will fail. They say the most common failure, building something the market doesn't want enough, is the one you can start testing before you write any code.

How an idea check works

Choose Test a product idea and describe the product in a sentence or two. You can add who it's for and where, and that matters: demand, competitors and rules in India can look very different from those in Germany.

CiteJury turns the idea into four questions: is there demand, who competes, what are the risks, and who is it for. It spreads its searches across all four, so competitors and risks get searched as hard as demand. It looks for demand signals such as search trends, forum and community discussion, surveys and spending. It looks for direct and adjacent competitors, with their pricing and reviews, for similar products that shut down and why, and for regulation. Market size is included only where a citable source gives one.

As with every check, the best passages are locked into a sealed evidence pack, Claude, ChatGPT and Grok assess it separately without searching the web, and a final review writes the result. You get an overall signal (Strong, Mixed or Weak) and the same rating for Demand, Competition, Risks and Who it's for, each with findings linked to their sources. At the end comes What to test next: suggested experiments, clearly labeled as suggestions and not findings.

A diabetes meal-plan app, tested

We ran the idea from the top of this page, with India as the market. Here's the scorecard it produced.

“A subscription app that plans weekly meals for people with diabetes (India)”

Mixed signal

Demand

Mixed signal

The need is large: an estimated 101 million people in India live with diabetes.

Competition

Weak signal

Free India-specific apps already give carb counts and diet plans.

Risks

Mixed signal

Free substitutes threaten willingness to pay, and medical-advice limits apply.

Who it's for

Mixed signal

Adults with type 2 diabetes, mostly urban, but no data on who would pay.

What to test next Run a small paid landing-page test and see how many people reach checkout, not just a free list.

The overall verdict was Mixed signal, and the four tiles tell one story.

The need is real. The demand finding, an estimated 101 million people in India living with diabetes, matches the national ICMR-INDIAB study published in The Lancet Diabetes & Endocrinology in 2023 and summarized by The National Medical Journal of India. But a large population with a condition isn't evidence that people will pay for a meal plan, which is why Demand came back Mixed rather than Strong.

Free is the competitor. Competition was the one Weak signal, because free India-specific apps already give carb counts and diet plans. For a subscription product, that's the finding that matters most: free alternatives set a ceiling on what you can charge. The Risks tile makes the same point about willingness to pay, and adds limits on how close a meal planner can get to medical advice.

The buyer is unproven. The likely users are adults with type 2 diabetes, mostly urban, but the check found no data on who would pay.

The suggested next step aims straight at that gap: a small paid landing-page test that counts how many people reach checkout, not just a free list. A waitlist measures curiosity. A checkout page measures intent. If the free apps are good enough for most people, a test like that can show it in weeks, for the cost of some ads, instead of after a launch.

None of this says don't build it. It says the idea as written competes with free, so the version worth building does something the free apps don't, for people who will pay for it. That's a far better brief for a prototype than the one you started with.

Test an idea yourself

1. Write the idea plainly. Say what it does, who it's for and how it makes money, in one or two sentences. "A subscription app" and "a marketplace taking a cut" face different competition.
2. Fill in who it's for and where. The market changes the competitors, the rules and the evidence.
3. Choose a tier. Standard, with Claude, ChatGPT and Grok plus a final review, is a good default. Deep searches wider and lets the AIs run their own searches while sources are gathered, which helps in niche business markets where evidence is scattered. Either way, an estimate is held while the check runs and anything unused comes back.
4. Read each area's findings and open the sources. Start with the competitors. Visit their sites, read their pricing and look at what reviewers complain about.
5. Turn What to test next into experiments. These are suggestions, not findings. Pick the cheapest one that could change your mind.
6. Run the variations. Try a narrower audience, another market or a different business model, and compare the signals. Follow-up questions answer from the same sources for about 1 to 2 credits.

To look harder at the competitors, you can compare them side by side or build a sourced list of companies in the space.

Reading the result

• A big population isn't demand. Evidence of spending, switching or complaining about current options says more than a prevalence figure.
• A weak competition signal usually means crowded, not hopeless. It tells you what you're up against, often free products, and so what you'd have to do differently.
• Look for the ones that closed. A competitor that shut down can teach you more than one that's thriving, if you can find out why.
• Take regulated categories seriously early. Health, money and children's products carry rules that shape the product itself, not just the terms page.
• Decide what would change your mind first. Before you read the result, write down which finding would make you drop or reshape the idea. It keeps you honest about what you read.
• Check the market-size numbers you'll pitch. Figures copied from report teasers and blog posts drift from the original. Trace them to the source before they go in a deck.

What it won't do

• It doesn't replace talking to customers. Web evidence describes markets and people in general. Only your prospective buyers can tell you whether they'd pay for your version, and how much. Treat the result as preparation for those conversations.
• It can't measure willingness to pay. That takes a real price in front of real people, which is why the example's suggested test is a paid landing page.
• Its view of competitors isn't a census. It reflects what the search found, and a new or quiet competitor may not show up.
• It can't read everything. Paywalled market reports, pages behind logins and some sites that block automated access can't be read. The result lists what couldn't be opened.
• It isn't a forecast or advice. A signal is a read of the evidence available today, not a prediction, and nothing in it is legal, financial or medical advice. For a health product like the one above, that matters.
• What to test next is a suggestion. It's labeled that way because it isn't a finding.

An idea check won't tell you whether to build. It gives you a sourced account of what's already known, so your first customer conversations start from the real gaps instead of from questions a search could have answered.

Questions people ask