Design the say

Hosts talk about a stream as a performance: energy, flow, chemistry with the room. All real, none of it measurable. A producer talks about the same stream as a run sheet: at minute 4 we pin the serum, at minute 9 we drop the price, at minute 11 we ask. That version is measurable, because every minute of every Lumen Glow stream has been labelled with what the host was doing in it, and every minute has a room size and an order count. Divide orders by the room and you have the value of that kind of minute, in orders per thousand viewer-minutes, comparable across a Wednesday with 1,579 people and a Monday with 498.

Once a script is a sequence of minutes with values, three things follow. You can see which minutes pay and which are the price of admission. You can see how long a cue keeps paying before it turns, which is the urgency finding. And you can hand a host a card instead of a mood: this minute's job is this, and here is what it is worth when it is done.

The nine segment labels. Every minute of a stream is one of these, and the value table in Part 1 is built on them.

• Hook - the opening minutes that earn the tap and the stay. Where a giveaway lives, when there is one.
• Product build - the story of the product before it is shown: who it is for, what problem it answers.
• Demo - the product in use, on camera. What session 4 measures from the picture side.
• Price - the anchor: the 29 dollar serum, the 49 dollar two-pack, what the bundle saves.
• CTA - the ask. "Tap the yellow basket" and its many cousins.
• Urgency - the reason to order now rather than later: a flash sale countdown, a coupon that ends, a genuinely limited quantity.
• Q&A - answering the room's comments by name.
• Filler - talk that does none of the above. Chat about the weather, the host's day.
• Dead air - nobody is talking. Usually a technical gap or a lost train of thought.

FABES is TikTok Shop Academy's own order for a product pitch: Features, Advantages, Benefits, Evidence, Scarcity. Read it as a script skeleton: build (features, advantages), demo and price (benefits, evidence), urgency (scarcity). Scarcity comes last and once.

The CTOR-vs-CTR split. What the host says mostly moves CTOR, click-to-order: the ask, the price, the reason to buy now all act on people who have already tapped the product card. What the host shows mostly moves CTR, the tap itself: the product in hand, the close-up, the price on screen. Tonight is the say and CTOR. Session 4 is the show and CTR.

The ladder above is the plain average. Session 2 taught the sharper comparison: minutes against the other minutes of their own stream, so the room, the host and the hour cancel. Measured that way, in points of CTOR against the stream's own ordinary minute, the say looks like this.

Read the table as a budget. Every minute of the night is spent on one of these, and the sum of the effects times the minutes is roughly how the stream converts against its own potential. A script that adds two CTA minutes and removes two filler minutes has moved about 18 points of minute-CTOR from the wrong side of the ledger to the right one without changing anything about the host, the room or the product.

The hook pays 1.98 orders per thousand viewer-minutes and converts 3.18 points below the stream's own average. If the script were only a list of values you would cut it. You would be wrong, because the hook's job is not to sell. The hook's job is entry and stay: it is the minutes when the feed is deciding whether to push the room and the viewer is deciding whether to give the stream nine minutes of their evening. Nobody buys a serum in minute two from a host they met in minute one. What they do in minute two is stay or leave, and everything after depends on which.

This is also where the giveaway lives, when there is one. Session 2 priced it: a giveaway minute converts 2.31 points worse and buys a bigger room. That is a hook doing a hook's job with a stronger tool, at a conversion cost the rest of the night pays. So the hook is in the script at its full weight, and the rule is about length, not presence: the hook's value is measured in who stayed, not in who bought, and the moment the room has arrived the script should move to build.

The Lumen Glow data prices the CTA as a minute. It does not say how the ask should be worded, because the label is the same whether the host says "tap the basket" or "you deserve to wake up to this skin, tap the basket". One published study does address it. Guo, Zhang, Goh and Peng (2025), in Production and Operations Management, report that affective calls to action, ones that speak to how the viewer will feel, raise orders in live streaming commerce, while cognitive calls to action, ones that list reasons and features, do not.

Read that with one flag attached: this is taken from the paper's abstract, not from the full text, so treat the direction as a lead worth testing in your own streams rather than a settled value. The practical script move it suggests is cheap to try: keep the features and evidence in the build and demo blocks where FABES puts them, and write the CTA itself in benefit-and-feeling language. Session 6 is where you would prove it worked on your own hosts.

An urgency run is a stretch of consecutive minutes in which the host is giving the room a reason to order now. In its first minute the click-to-order rate is 15.12 percent; in its second, 15.44. Both are well above the 12.1 percent a Lumen Glow stream averages. In the third minute the rate is 8.30, and it stays around there, 8.33, 8.43, 9.09, for as long as the host keeps going. Within stream, the same pattern: the first two minutes of a run lift CTOR by +4.93 points against the stream's own ordinary minute, and every minute from the third onward lowers it by 1.84 points. The cue does not fade. It flips.

The reason is in what a scarcity cue is. A flash sale with a genuine countdown on screen, a coupon that genuinely expires at the top of the hour, a bundle with a genuinely limited number of units: these are facts about the offer, and a viewer who hears one either acts on it or decides not to. Two minutes is about how long that decision takes. A host who is still saying "almost gone" in minute five is no longer reporting a fact; the room can see that nothing has gone, and it re-reads the whole pitch as theatre, including the parts that were true. That is the 1.84 points, and it is also why the stretched version costs trust that the next flash sale has to buy back.

• Real cue: LIVE Flash Sale with its countdown visible, LIVE Coupon with a stated end, Creator Exclusive Price for this stream only. All three are tools inside LIVE Manager; the platform shows the clock so the host does not have to keep claiming it.
• Two minutes: state the cue, state what it is worth in dollars, ask, and move on to the next block. If the offer is still live at the next cycle, one sentence reminds the room; it does not need a run.
• Stretched cue: anything past the second consecutive minute. Converts at about 8 percent, costs 1.84 points within stream, and from minute four it is the thing the next card describes.

From minute four the stretched run is more than a conversion problem, because the sentences a host reaches for to keep it alive are the sentences the platform and the regulators have written down. The test in every one of these texts is the same: is the claim true. A genuine countdown is fine. A limited-time claim that is not limited is the named offence.

The pattern across seven texts from three jurisdictions and one platform is not subtle. A scarcity claim tied to a real, visible, checkable fact is a selling tool. The same words with nothing behind them are a listed deceptive practice. Minute three of an urgency run is where the fact usually runs out and the words keep going.

Urgency is one family of risky sentence. The other is the product claim, and TikTok Shop's misleading-content guide lists six absolutes it does not allow for a product like a serum: "100% effective", "Guaranteed results", "Works for everyone", "Permanent effect", "Instant results", "No side effects". They tend to appear in the same minutes as stretched urgency, because a host who has run out of true things to say about the offer reaches for stronger things to say about the product.

The builder track, session b4, turns this list into a checker: the stream's audio is transcribed with word-level timestamps, and every minute is scanned for the flagged urgency phrases and the six absolutes, plus the platform's flagged discount wording. A hit becomes a policy-risk flag on that minute, which is exactly the count session 2 quoted for Ren, 6.82 flags per stream. For a leader the useful part is not the code but the shape of the output: a list of minutes, with the sentence and the rule it tripped, ready to hand to the host with the script card. It is much easier to coach "you said 'guaranteed results' at minute 47" than "watch your claims".