"We got 4,000 mentions this month." Congratulations. How many of them were a person deciding whether to spend money?
Usually: about six. And those six are worth more than the other 3,994 put together, because they are simultaneously the thing answer engines quote and the raw material for understanding why people do not buy from you.
So we classify every thread rather than counting them. Buying intent, comparison, problem, complaint, or plain mention. Then we pull the actual question or objection out verbatim, because a summary of an objection is useless and the exact wording is gold.
The classification is word lists, not a model, and we show you the raw title next to every label so you can see what we saw and disagree with it. A confident classifier you cannot inspect is how you end up with a report full of plausible nonsense.
Two filters do most of the work, and both exist because of specific embarrassments.
Relevance, which requires the brand or category to be in focus rather than mentioned in passing. That is what keeps a thread titled "A (Very Brief) History of Racism In K-Pop" out of an earplug report. Yes, that happened.
And market, which is a separate test from relevance. A thread can be entirely about your category and still be useless: two of the most active communities we found for one brand were Philippine, and the customer sells in the US, UK and Denmark. On one run, 15 of 36 threads were dropped, each with the reason written down.
A thread ledger with no memory is a reminder, not a process. So threads persist across runs, and you can see which are new, which you have already dealt with, and which are still open. Otherwise you get the same three conversations presented as news every fortnight.
Around this sit the other places the category gets discussed: press via RSS, roughly 100 hits a run, YouTube, TikTok, and podcasts.
On podcasts, the limit needs stating in the first paragraph rather than the last. We read titles and show notes. We do not transcribe audio. Podscan does. That means a sponsorship read that appears nowhere in the notes is invisible to us, so our count is a floor, not a ceiling. If podcast tracking is the whole reason you are shopping, buy the tool that transcribes.
One more thing about the video sources, because it explains why the numbers used to be nonsense. The most-viewed video for one brand we track turned out to be a Minecraft creator who happened to share the brand's name. Another brand's top video was a prom transition with 27.9 million views and nothing to do with the product. Name matching alone produces garbage at scale, and the filter that fixes it is unglamorous and absolutely load-bearing.
One thing to do next: search your product category on Reddit, sort by new, and read ten threads.
Not for mentions. For the questions people ask before they buy.
Pick one competitor. See what we find.
Your first benchmark is free. One domain, measured against you, no card and no call.