How to validate an AI startup idea with public data

To validate an AI startup idea with public data, find the products that already serve the need, measure their traffic, read their pricing, estimate their revenue range and look for gaps. If paying customers clearly exist and you see a way to serve them better, the idea has earned a small test. If nobody is paying today, be cautious.

This method takes a few hours, costs nothing, and often saves months of building.

What public data can and cannot tell you

Public data can show that a market exists, how big the leaders are, and how products price themselves. It cannot show profit, churn, or whether customers are happy. Treat it as a filter that rejects weak ideas early, not as proof that a good idea will work.

Everything below produces estimates. Label them as such in your notes, and confirm with real customers before you commit.

Step 1: write the idea as a job to be done

Start with a sentence in this form: "Help [specific person] do [specific task] faster, cheaper or better."

Examples are "help real-estate agents write listing descriptions" and "help podcasters turn episodes into clips." A narrow sentence is easier to research than "AI for marketing."

Then list three or four phrases a customer might search to solve that job. These become your keywords and competitor searches.

Step 2: map the existing products

Search for those phrases and collect every product that appears, including the unglamorous ones. Add competitors from category pages in the AI startups directory, such as AI marketing or AI video.

Record each product's name, domain, target customer and main promise. Aim for 10 to 20 products.

Competition is not a bad sign. A crowded market with many paying customers proves demand. An empty market often means nobody wants it enough to pay.

Step 3: measure traffic

Look up each domain in the traffic checker. You get modelled monthly visits, top regions, an SEO audit and a tech stack for each. If you have many domains, the bulk checker handles a list at once.

Build a small table:

Product	Monthly visits (modelled)	Top region	Notes
Example A	Tens of thousands	United States	Strong SEO
Example B	Thousands	India	Mostly direct

What to look for:

• Total demand. If the leaders get very little traffic, the market may be small.
• Distribution of traffic. If one product holds most visits, it will be hard to enter head on. If traffic is spread across many small players, there is room.
• Direction. Rising visits over several months suggest growing interest. Falling visits suggest fading novelty.
• Regions. The countries sending traffic show where your first customers may live.

Check the trending traffic rankings to see whether your category contains fast risers.

Step 4: read the pricing pages

Open each competitor's pricing page and capture:

1. The lowest paid plan and its billing period.
2. How limits work: credits, seats or features.
3. What the free plan includes.
4. Whether annual billing is discounted.

This tells you what customers already accept. If every product charges $20 or more, a $5 tool needs enormous volume. If every product has a free tier, you need a plan for free users. The patterns are discussed in pricing page patterns across AI startups.

Step 5: estimate revenue ranges

Combine traffic and price. The simple formula is monthly visits x paying share x lowest paid plan. StartupsIntel's own estimate uses a 3 percent paying share and an 80/20 split of monthly and yearly plans. See the methodology for the details.

Try a low, middle and high case and write the results in your table. The point is the order of magnitude. If the top products appear to earn thousands a month, the market might be small. If they appear to earn hundreds of thousands, you have a real category.

Estimates can be wrong, particularly for enterprise products and tools with off-site usage. We discuss the limits in estimating revenue from traffic. The monthly revenue rankings give a ready-made comparison across many AI products.

Step 6: look for evidence that people pay

Traffic is interest and revenue is intent. Look for more direct signs of paying customers:

• Public customer logos or testimonials on competitor sites.
• Active reviews and complaints, which show real users.
• Job posts, which show a company is hiring because it is growing.
• Businesses listed for sale with verified revenue. The Buy a startup marketplace shows what small AI businesses actually earn and what they sell for.
• Community threads where people ask for a solution and name prices.

Complaints are especially valuable. A competitor with many unhappy users suggests a gap you can fill with better quality, price or support.

Step 7: find the wedge

A wedge is the specific reason a customer would pick you over incumbents. Good wedges for a new AI product include:

• A niche audience the leaders ignore, such as a profession or language.
• A better workflow, not just a better model.
• Lower price with a sustainable margin. Check inference costs first, as explained in unit economics for AI products.
• Unique data or distribution that a model alone cannot give.
• Regional focus, where a global tool serves local needs poorly.

Be wary of a wedge that depends only on a model feature. Large labs and competitors can copy it fast.

Step 8: run a paid test

Public data narrows the field. The final test is whether strangers will pay. Before building the full product:

1. Create a one-page site that describes the outcome and shows a price.
2. Offer a pre-order, waitlist with deposit, or a manual version of the service.
3. Drive a small amount of targeted traffic, from communities or modest ads.
4. Count how many people click, sign up and pay.

Even a handful of paying customers is stronger evidence than any spreadsheet. If nobody pays after a fair attempt, revise the idea or move on.

A simple scorecard

Rate each factor from 1 to 5 and add them up.

• Demand: do competitors have meaningful traffic?
• Willingness to pay: do products charge and keep customers?
• Gap: is there a clear wedge?
• Reachability: can you reach buyers affordably?
• Margin: does pricing leave healthy gross margin after inference?

A high score does not guarantee success, but a low score is a good reason to stop early.

Common validation mistakes

• Treating traffic as proof of revenue.
• Ignoring that free tools get traffic with few buyers.
• Confusing a trend spike with durable demand.
• Asking friends instead of strangers who would pay.
• Skipping margin math for AI features.

How to keep notes so the research is reusable

Validation work is worth more when you can come back to it. Keep one spreadsheet or document with a row per competitor and a column for each fact you collected, plus the date you checked it. Traffic, pricing and plan limits all change, and a dated record lets you see movement over time.

Add a short written summary at the top with your current conclusion and the three facts that most influence it. When new evidence arrives, update the summary rather than starting again. If you later partner with a co-founder or pitch an investor, this record shows that your decision rests on evidence and not on enthusiasm.

FAQ

Can I validate an idea without building anything?

Yes, to a point. Public traffic, pricing and revenue estimates show whether a market exists, and a pre-order page tests willingness to pay. Only real customers can confirm that your solution works.

Is a crowded market a bad sign?

Usually not. Many competitors with real traffic suggest paying demand. The question is whether you have a specific wedge. An empty market is more worrying than a crowded one.

How reliable are traffic estimates for validation?

They are good for order of magnitude and comparisons, not for exact numbers. Use them to rank products and spot trends, then verify important findings through other signals or direct conversations.

How many competitors should I research?

Ten to twenty is enough to see patterns in pricing, traffic and positioning. Go deeper only on the three or four closest to your idea.