We gather scattered reviews and count how many recommended it and how many regretted it. Sponsored posts removed, every number linked to its original.
DEMO Counted from sample reviews for practice. The counting method is the same as with real data.
“With our second we only used the lightweight one. Folding it one-handed into the boot is what matters.” ▶ YouTube comment · view original
You can. But what it reads and what we read are different things. When the answers diverge, that is why — we counted something else.
Whatever a crawler can fetch — blog reviews and comparison articles. In Korea a large share of that layer is sponsored and seeded content.
YouTube comments and replies — not in the page HTML, so crawlers never see them. We fetch them through the official API.
Naver cafe posts — they sit behind a login.
Reviews left here — first-party data that exists nowhere else.
The sponsorship is in the post. The honest review is in the comments underneath it.
YouTube comments and replies, Naver blog and cafe posts — official APIs only. We do not scrape.
Anything judged paid-for is excluded from the counts, and we show you how many were removed.
“9 recommend, 1 regret” per item. No rounding into percentages, and the dissent stays visible.
Open the reviews behind any count, filtered by source, each linking to the original.
A service about counting reviews honestly shouldn't oversell its own status. So here it is.
The counting engine Collection, sponsored filtering, item extraction, tallying and the result page all run end to end.
Your own review Takes ten seconds, enters the count immediately, and shows as its own source.
Real collection Every number you see here is practice data. With real API keys, the same code counts real reviews.
English reviews The stance rules are Korean, so English is counted via the LLM path. The code is ready but has not been run live yet.
The questions that arrive from search and the questions an assistant gets asked are nearly the same. So the answers are written here to be quoted either way.
ReviewSaid gathers public reviews scattered across YouTube comments and Korean blogs and forums, removes the ones written for payment, and counts what is left. The answer comes back as people — “9 recommend, 2 regret” — and every review behind a number keeps a link to its original.
We do not guess; we only read what the post discloses. We look for explicit wording — “sponsored”, “paid partnership”, “#ad”, “sent this product for free”, “review campaign” — and for affiliate links from known programmes and review-campaign platforms. A negated mention like “not sponsored” is not excluded, and a YouTube comment complaining “too many ads” is not treated as a disclosure. The number we removed is printed on the result.
Yes, and that is the point. YouTube comments and replies are not in the page HTML, so search crawlers never see them and most AI assistants cannot read them either. ReviewSaid pulls them through the official API. The sponsorship is in the video and the blog post; what people who actually used the thing said is usually in the comments underneath.
A percentage hides how many were counted. “82% satisfied” could be 9 of 11 or 900 of 1,100. Counting people shows how thick the evidence is at a glance, and dissent does not get rounded away.
Revenue comes from affiliate commission on purchase links added after the count, and the commission rate is never an input to the count or the ordering. Options that pay nothing are shown the same way, every link discloses whether it earns a commission, and there are no ads inside a result.
Not yet — it has not opened. The counting engine runs end to end, from collection to the result page, but every number on this page is practice data and real collection is still being set up. Search is free by default, and when it opens we will email the address you leave below, once.
One email, nothing else. Never used for anything but this. Tell us what you're deciding on and we'll have that search counted first.
Every number on this page is demo data, not a real count.