Back to Blog
startup-strategy5 min read

How to tell if an AI business idea is backed by real evidence

I fact-checked 44 claims from AI market research. A third were fully right and fewer than 1 in 10 were flat wrong. The rest went wrong in the same few places, and those are the places to check before you build.

PaulNovember 16, 2026

Ask any AI tool for business ideas and you'll get a confident list with a market size, a "why now" and a revenue estimate for each. Some of it will be right, and you need to know which parts before you spend your evenings building on them.

I run OppFinder, which does this kind of research with live Google searches and keeps the sources. Last month I took 44 specific claims from its scans (dates, prices, regulations, named companies) and checked every one by hand.

What I foundClaimsShare
Fully right14about a third
Right story, wrong detail18about 40%
Couldn't confirm anywhere9about 20%
Flat wrong3under 1 in 10

Outright invention was rare. Most of the problems were in details you would quote to a customer, and they went wrong in the same few places every time, which makes them easy to turn into a checklist.

1. Check the "why now" against the original source

Every good idea has a trigger: a new regulation, a court ruling, a price change, a deadline. It's usually the most valuable part of the research and the part most worth checking, because it's what you'll lead with when you talk to customers.

In my checks the event was almost always real, while the details around it often weren't. One scan said a consumer opt-out registry was launched by the data regulator. It was launched that month, but by the trade ministry and a different agency. Another put a regulator's directive in July when the sources I could find said August.

Open the gazette, the judgment or the regulator's own announcement and confirm three things: what changed, when, and who did it.

2. Check every price on the vendor's own page

Prices were the single most common problem. Across three different AI models, the claims I couldn't find any support for were mostly specific price points, like "R299 per user per month" or "R899 a month for the solo plan". Two were for products I couldn't find at all.

Before you anchor your pricing to a competitor, open their pricing page. If they don't publish prices, that's useful too: it usually means sales calls and custom quotes, which tells you something about how that market buys.

3. Check the year on every statistic

Several "wrong" claims were right figures from the wrong year. A medical schemes guideline was quoted as 3.3% for 2027, but 3.3% was the 2026 figure and 2027's is 3.8%. A cost figure turned out to be one component of a larger total from a different study.

When you see a statistic, find where it was published and when. Last year's number in a sales email reads as carelessness to a buyer who knows the market.

4. Treat market size as arithmetic and check the inputs

Nobody publishes a market size for a niche as specific as "citation checking for South African litigators". So a good analysis works it out: number of buyers times what each would pay.

That's fine, as long as the inputs are right. In one deep dive the whole market size rested on "roughly 15,000 legal practitioners". The Law Society's own figures put practising attorneys at about 35,000. Every number built on top of it was off by more than half.

When you see a TAM, look for the headcount and the price it was built from and check those two numbers. They matter more than the total.

5. Read the reason not to do it

The most useful line in an idea is often the counter-argument. For the citation checker, the analysis itself flagged that lawyers can already check citations for free on SAFLII. That one sentence decides whether there's a business, and it's cheap to test: time how long the free way takes.

If an idea doesn't come with a credible reason it might fail, find one yourself before going further.

How OppFinder tries to make this easier

I built OppFinder so this checking takes an evening instead of a week. It doesn't remove the need to check.

  • Every idea lists the web sources behind it, so you can open them yourself.
  • It records which sentences in its research Google matched to a source, and tells the analysis to treat the rest as unverified.
  • When it has nothing to base a number on, it says "Unknown - would need..." and names what would settle it.
  • Each deep dive gets a second pass that compares the analysis with its research and lists anything that didn't match, in a box called "Check Before Relying On This".

The details still need checking. This just shows you which ones to check first.

Checking the legal tech idea took one evening, and the price checks were the quickest part.

The one-evening version

If you only have an evening:

  1. Confirm the trigger in the original source (what, when, who).
  2. Open the pricing pages of the two competitors the idea names.
  3. Check the year on the one statistic the idea depends on.
  4. Find the headcount behind the market size.
  5. Spend the rest of the evening on the reason not to do it.

Then talk to five people who would have to pay for it. Research doesn't replace those conversations, but with this list done you'll go into them with facts that hold up.

You can run a scan on your own industry at oppfinder.app and check its sources the same way.


Related Resources:

idea-validationmarket-researchai-business-ideaschecklist