Defining your ICP from what a segment already says out loud

The short answer: separate candidate segments by the words they use and the explanations they skip, then keep the one whose urgent problem you can solve completely. Public video research gives you vocabulary, workflow and urgency with reasonable confidence; it does not give you budget authority, so that field stays a hypothesis until someone tells you directly.

Most early ICP documents are demographic fiction: a job title, a company size band, and a list of adjectives. They fail because none of those fields predict whether a person will buy. What predicts buying is whether the problem is urgent in their specific working context — and context is the thing public content is unusually good at revealing.

The three tells that separate segments

When two groups discuss the same underlying problem, three differences show up consistently. Each is easy to spot in a corpus and hard to fake.

Vocabulary. The same concept carries different names in adjacent communities, and people rarely switch registers. A group that says “pipeline” and a group that says “workflow” for the same object are usually two markets with two sets of assumptions.

Assumed knowledge. What a source skips explaining tells you who they think is listening. Content that defines a basic term is aimed at newcomers; content that uses it bare and moves on is aimed at practitioners. The skipped explanations map the segment boundary.

Which step is called hard. Two groups with identical workflows will nominate different bottlenecks, and the nominated bottleneck is where their money goes. This is the single most useful signal in the whole exercise.

The bottom two rows are the honest limit. Writing them into an ICP as if research established them is the most common way these documents turn into confident fiction — the general failure pattern in what AI video research gets wrong. Mark them as hypotheses and test them in conversation instead.

Urgent versus merely annoying

Every segment has a long list of annoyances and a very short list of urgent problems. Products attached to the first list struggle indefinitely; products attached to the second sell themselves. The tells are observable.

The strongest single indicator in that right column is elaborate workarounds. A spreadsheet someone built and maintains by hand is a purchase order with the vendor field left blank — and it is the same signal that makes instructional content such fertile ground, as in finding micro-SaaS ideas in tutorial videos.

Drafting the profile

A useful ICP fits on one page and reads as conditions rather than adjectives. The format that survives contact with reality looks roughly like this: this product is obviously worth buying for someone who does X weekly, currently handles it with Y, considers Z the hard part, and has already spent something on the problem.

Each clause is testable. “Does X weekly” can be confirmed or denied in one interview question. “Considers Z the hard part” can be checked against the corpus and against a call. Compare that to “technical decision-makers at mid-market companies”, which cannot be falsified and therefore cannot guide anything.

Choosing between viable segments

Research often surfaces two or three plausible segments, which feels like good news and is actually the hard decision. Three criteria decide it in practice.

Completeness of the fix. Can you solve their urgent problem end to end, or only the first half? A partial solution to a burning problem loses to a complete one every time, because the buyer still has to keep the workaround alive.

Whether you speak the dialect. Writing to a segment whose vocabulary you have only read is possible but slow, and readers detect it. Prefer the segment whose language you can already write without checking.

Reachability. A segment that congregates somewhere specific is worth more than a larger one that does not, because distribution is the constraint after the product works. If the corpus shows a segment clustered around a handful of channels and communities, that is a distribution plan appearing early.

If two candidates remain genuinely tied after those three, the structured comparison in choosing between two SaaS ideas applies to segments as cleanly as it does to products.

Testing the profile before you build on it

A profile drafted from research is a hypothesis with good sourcing, not a finding. Two cheap tests catch most of the ways it can be wrong, and both can be run in an afternoon.

The falsification test. Write down what you would expect to see if the profile were wrong, then go looking for it specifically. If you claim a segment considers a particular step the hard part, sources who breeze past that step without comment are counter-evidence. Searching for the disconfirming case is the only reliable defence against a corpus assembled to agree with you.

The stranger test. Show the profile to someone who works in that segment and watch which sentence they stop at. Agreement tells you nothing; the sentence that makes them pause and say “well, sort of” is where your framing has drifted from theirs, and that is precisely the sentence that would have gone into your homepage.

A profile that survives both tests is worth building on. One that has been through neither is a description of the sources you happened to watch, and it will produce copy that reads as though it were written from the outside — because it was.

What it takes to run

Budget eight to twelve sources per candidate segment and a day of analysis. As of August 2026 that fits the entry plan at $19 a month for 25 videos and 2 projects; $59 covers 80 videos and 8 projects if you are profiling several markets at once, and $199 covers 250 videos, 20 projects and 3 seats for teams doing this continuously. Full detail is on the pricing page.

Once the profile stabilises, it becomes the filter for everything downstream — which features matter, which objections to answer, which evidence counts. Keeping it current as the segment moves is the recurring job described in monitoring a niche with recurring research.

Closing thought

The strongest ICP statements are the ones a stranger in that segment would read and find obvious. If yours reads as insightful to you and generic to them, the research has not gone deep enough yet.