The video has four million views by the time you see it. The product — a jar organizer, a sunset lamp, a cleaning paste — is selling out somewhere, the comments are a wall of "I need this," and every hour you spend deciding feels like margin evaporating. This is where TikTok product research actually happens: not in a spreadsheet, but in a seventy-two-hour window with incomplete information.
The uncomfortable part is that the view counter cannot answer the only question that matters. A trend that dies in three weeks and a trend that runs for two years look nearly identical in week one. Same growth curve, same breathless comments, same "TikTok made me buy it" compilations.
What separates them is structure, not size. This is a field method for reading that structure: the anatomy of the trend itself, the ledger hidden in its comments, and three durability checks that live outside TikTok entirely.
The lifecycle of a viral product
Viral products move through four phases: discovery, pile-on, saturation, collapse. Discovery is a handful of creators finding something genuinely novel. Pile-on is the imitation wave — the sound spreads, the format gets copied, view counts go vertical. Saturation is when every seller runs the same clip as an ad and the comment tone shifts from "where do I buy this" to "I have this, it broke." Collapse is attention simply moving on.
The whole arc often fits inside sixty to ninety days. That is the trap, because sourcing has its own clock: three to four weeks for inventory to land is normal, longer if you are negotiating a private label run.
Most losses in this game do not come from bad products. They come from good products entered at the wrong phase — a seller sources during the pile-on, stock arrives in saturation, and they sell into the collapse. The product was real. The timing was fatal.
So the real question is never "is this product good." It is "which phase am I looking at, and is there anything underneath the attention."
Spike anatomy vs signal anatomy
A spike has a single load-bearing element. One sound that made the format work. One creator whose delivery is the joke. One moment of novelty where the video is the entertainment and the product is a prop. Remove that element and nothing remains, because nothing else was ever there.
A signal is redundant. Multiple independent creators, posting months apart, framing the product differently, with no shared sound and no shared script. The videos are about a problem the product solves, and the comments are about the problem too. Any single video could vanish and the next one would still get made, because the demand exists outside the format.
The distinction is structural, not aesthetic. A spike can dwarf a signal on raw numbers — a one-sound wonder at ten million views is still more fragile than a boring kitchen tool with forty unconnected videos across a year. The test worth carrying:
If the biggest video disappeared tomorrow, would there be any reason for the next one to exist?
In practice: search the product on TikTok and scan the upload dates. Videos spread across months from unrelated accounts is signal anatomy. A dense cluster inside ten days, all riding one audio, is spike anatomy — however large the cluster is.
Reading the comments like a ledger
Most people read viral comments as hype. Read them as bookkeeping instead — every recurring comment type is an entry with a meaning.
- "Where can I buy this" density. This is raw purchase intent, and its ratio to total comments matters more than its count. When buying questions dominate, demand is outrunning awareness of where to purchase — which is precisely the gap a seller fills.
- Unanswered variant questions. "Does it come in black." "Will this fit a king mattress." "Does it work on pet hair." Each one is unmet demand the original creator is not converting, and together they are a free draft of your product page and your FAQ. If nobody in the thread is answering them, the lane is open.
- Skepticism. "Does this actually work" is not a negative signal — it is a cost estimate. Every skeptical comment previews the education your ads will have to pay for before a stranger converts. A little skepticism is normal; a wall of it means an expensive convincing job, priced into every click.
One entry deserves separate treatment: "I bought this and it broke." That is not skepticism, it is product truth arriving from the saturation phase, and it usually means the cheap version of this product has already burned the audience you were about to sell to.
The durability cross-check
Everything above tells you attention exists. Whether demand exists — the kind that survives the algorithm losing interest — is a question TikTok cannot answer about itself. Three checks, all free, each answerable in ten or fifteen minutes by hand.
Search history
Open Google Trends and set the window to five years. You already know the product is spiking; the question is what previous spikes left behind. If earlier waves of attention left a permanently raised baseline, the trend converted browsers into a durable market — people kept searching after the videos stopped. If the line fell back to zero every time, the category has spiked and died before, and you are watching a rerun. There is a longer treatment of this in reading search demand without fooling yourself.
Ad persistence
Search the product and its category in the Meta ad library and look at launch dates, not creative. An ad that has been running thirty days or more is, in field terms, paying for itself — nobody funds a losing ad for a month. Advertisers still active weeks after the videos peaked means the unit economics work off-platform. Ads that all launched this week means you are looking at the pile-on itself. No ads at all means either you are genuinely early or the product cannot be sold profitably with paid traffic — and you should want to know which. The full method is in reading the Meta ad library.
Retail accumulation
Find the product on Amazon and read review dates like sediment layers. Reviews still arriving steadily after the trend peaked means buying continued once the attention moved on — retail proof that the product outgrew its videos. Reviews that stop cold at the same time the trend did mean the product was content, not commerce. The mechanics of review velocity are covered in what Amazon rank and review velocity actually say about demand.
Timing rules
Two rules cover most situations.
Enter while heat is rising and at least two of the three durability checks agree. Rising heat alone is how people buy inventory for the collapse phase. Any single check can mislead — a raised search baseline with no ads, or persistent ads against a dead search line — but two independent confirmations rarely lie in the same direction at once.
If the peak has already passed, demand must show a second leg before you commit inventory or budget. A second leg looks like the search baseline holding above its pre-trend level, advertisers still paying weeks out, and reviews still accumulating. A second leg is evidence of a market. A first spike is only evidence of a video.
One scan instead of four tabs
Done by hand, this cross-check is forty-five minutes of tab-juggling per product — TikTok comments, Trends, the ad library, Amazon — and in a seventy-two-hour decision window that cost is real. This is the specific chore Kestrel was built to compress: it reads community chatter, Google search demand, live Meta ad library activity, and Amazon retail proof in one pass, then scores the market 0–100 as Hot, Promising, or Weak with the evidence broken out per source. A complete worked example is public on the specimen page, so you can judge the output before trusting it. The method in this post works without it; the tool just runs the same method in minutes instead of an afternoon.
The honest failure mode
Some spikes are genuinely unknowable in week one, and it is worth saying so plainly. A truly new product has no search history to leave a baseline, ads too young to prove persistence, and no review trail. All three durability checks return "too early" — which is an answer, just not the one anyone wants.
The honest response is position sizing, not conviction. A small test order instead of a container. A capped ad budget with a decide-by date. Treat a week-one entry as paid information: you are buying the data the checks could not give you, and the price of that data should be small enough that being wrong is a line item, not an obituary.
The sellers who last in this game are not the ones who call every trend correctly. They are the ones whose wrong calls were small and whose right calls got follow-up capital — because they sized for the uncertainty instead of pretending the chart resolved it. If you want the cross-check without the tab-juggling, the first twenty market scans are free, no card required.
