"Where did that number come from?" — How to add sources and reference dates to your market research

You put a market size figure in your business plan. You asked AI, it gave you a number, and the number looked plausible, so in it went.

Then, during the pitch, someone asks a single question.

"Where did that number come from?"

If you get stuck here, the rest of your explanation doesn't land well. Not because one number was wrong, but because the audience starts to think the other numbers were probably produced the same way.

AI is good at producing numbers. It doesn't attach sources

Ask "What's the size of the domestic market?" and you get an answer. The unit looks right and the sentence reads smoothly.

But two things are usually missing: as of when, and who published the figure. Even when an organization name that looks like a source is attached, you can't know whether that organization actually published that number until you check.

This isn't a case of using the tool wrong. Writing sentences and citing sources are different jobs. Attaching sources is still the part a person has to do.

Every number needs four fields

When you write a number, four things should sit right next to it.

1. Value — how much
2. Unit — won, people, cases, or percent
3. Date — as of when (at least the year, and the survey period if possible)
4. Source — who published it (organization name + report title)

Here is the difference.

• Not enough — The domestic market is worth roughly KRW 1 trillion.
• What it should look like — Domestic ○○ market size: KRW 1.2 trillion (as of 2025, △△ Association, "2025 ○○ Industry Survey")
  * The organization, report title, and figures above are fictional examples used only to show the format.

It looks long, but with this one line in place, that earlier question doesn't come up. And if it does, you have an answer.

When one field stays empty

This is where most people stall. You search and can't find a source, or you find one but the figure is several years old.

You have three options.

1. Remove it. The easiest choice. A blank is better than a wrong number. A number without a source isn't neutral; it's a minus, because once one number is shaken, the rest come under suspicion too.

2. Turn it into a range. If you can't give an exact value, write only the range you can verify and its basis. Something like "The exact figure could not be confirmed, but based on publicly available ○○, it is at least ○○."

3. Replace it with a number you produced. This is actually the stronger option in many cases. A number you counted yourself is often more persuasive than someone else's market size. For example, how many of 20 interviewees were already spending money on the problem, or how many people signed up for your waitlist. A small sample is fine. It's small, but the source is certain.

"Just the organization name" is not a source

There's a pattern we see often.

• This doesn't count as evidence — According to the △△ Association, the market is growing rapidly.

There's an organization name, but not which report, from which year. That leaves no way to check it, and what can't be checked isn't evidence.

It helps to sort your numbers into three groups.

• Numbers you produced yourself — interview, sign-up, and payment records. Small samples, but the source is certain.
• Numbers checked against the original — you opened the report and saw the page.
• Numbers you saw somewhere — carried over from articles, blogs, or AI answers. If you couldn't open the original, you can't use the number yet.

Trouble starts when the third kind is used as if it were the second.

How we handle it

We ran into the same problem while making our own content, so we keep one method in place.

We maintain a separate list of numbers we're allowed to use, and any number not on that list stays out of our articles. Only entries with all four fields (value, unit, date, source) filled in go on the list. And under every article we publish, we note the figures we chose not to use. You'll find them under this article too.

It's tedious. But it's worth at least this much: six months later, we don't have to dig around again asking "where did this number come from?"

When you ask AI

Just changing how you ask for numbers changes the result.

"Tell me the size of the ○○ market, but write the value, unit, reference year, and source (organization name and report title) as separate items. For any item where you can't identify a specific source, don't give a number; write 'cannot be verified' instead."

Asking this way makes the blanks visible. Blanks appearing isn't the bad part; moving on without knowing about the blanks was the problem.

One more rule is worth keeping: always open the original before using a source the AI gave you. Sometimes even a convincing report title comes back, but when you open it, the report doesn't exist or the figure is different. Using only what you've opened greatly reduces this risk.

If you do just one thing today

Pick just one number currently in your business plan or proposal, and write out the four fields next to it.

Value ____ / Unit ____ / Date ____ / Source ____

If even one field stays empty, that number is exactly where a question will come during your pitch. Removing or replacing it now is easier than searching for an answer on the spot.

---

Not sure whether the numbers in your materials have evidence behind them?
A 2-minute assessment lays out the one task to tackle first, a plan for starting a 2-week pilot, and the risks a person must check. Feel free to answer only as much as you're comfortable with.

STAR-T AI Business Operations Assessment →

---

Notes

① References and verification date
This article does not cite external statistics or research. Its basis is more than 500 cumulative one-on-one consulting sessions with aspiring founders (per our career-record SSOT) and the in-house review process we use when publishing content. Numbers in the text such as "KRW 1.2 trillion," "2025," and "20 interviewees," as well as organization and report names such as "△△ Association, 2025 ○○ Industry Survey," are all fictional examples used only to show the format; they are neither real survey results nor existing organizations or reports. Verified 2026-08-06.

② Figures not used
We did not use figures such as "X% of AI-generated statistics are inaccurate," because we could not confirm their original source or measurement method. We also did not use satisfaction or performance figures from our lectures and consulting, or figures and direct quotes from internal surveys or customer support responses, in this article.

③ How this was made
The draft was created with AI, then verified and edited by people. Before publication it went through 3 reviews: facts, tone, and legal.