Official Crafted Creative Intelligence Report: What Makes a Video Sell

hirteen words. That is the length of the opening line that separates a creator video capturing emails at $2.98 from one capturing them at $10.58.

We pulled that number out of our own campaign data. Crafted launched its Creative Intelligence tool last month and analyzed 157 Meta video ads run by CPG brands driving shoppers to retail. Every hook was coded by its opening structure before any performance data was attached, so the coding could not be influenced by knowing which videos had won. Then each hook was measured against its own brand's median rather than a pooled average, which isolates the creative choice from the brand effect.

What came back is a short list of things worth briefing and a shorter list of things you can stop arguing about.

Creative is the variable, and most briefs spend their energy elsewhere

Before getting into what moved, it helps to know how much creative can move at all. NCSolutions, now part of Circana, modeled the drivers of short-term sales lift across nearly 450 CPG campaigns and found creative accounted for 49% of the outcome and targeting for 11%. That study is from 2023 and the ratio has been remarkably stable, sitting at 49% for creative in their earlier 2017 work too.

The economics say the same thing. Motion analyzed 578,750 unique creatives representing $1.29 billion in realized Meta spend between September 2025 and January 2026 and found that roughly 5% of creatives qualify as winners and about half get switched off inside 28 days. Around 55% of total spend concentrates behind the small group that works.

So the brief matters, and the volume of briefs matters. What has been missing is evidence about which parts of the brief actually change the outcome.

Opening length: 13 words is the floor worth briefing

Opening length turned out to be the single strongest structural lever in the dataset, and it moves cost per lead more than anything else we coded.

Hooks with openings under 13 words captured emails at $10.58. Hooks in the 13 to 19 word band captured them at $2.98. Hooks running 20 words or longer came in at $2.94, and picked up 2.7 points of hold rate against their own brand median on top of that. Within brand, the cost per lead rank correlation is −0.368 at p=0.0015, and the hold rate correlation is +0.369 at p=0.0011.

Here is what each band sounds like in practice:

• Nine words, $10.58 per lead. "Avo Fresco is giving away free guacamole. Enter the sweepstakes now."
• Thirteen words, $2.98 per lead. "On my way to secure Tato's next obsession. Package secured from local Walmart."
• Sixteen words, $2.94 per lead and 2.7 points of hold. "Avocado that actually stays green for two weeks after opening. What kind of magic is this?"

The mechanism we think is at work here is premise. A longer opening gives the viewer a reason to stay through the middle of the video, and the middle is where the offer appears. A nine-word opening announces a giveaway and asks for the click before anyone has been given a reason to care. A sixteen-word opening sets up a question the rest of the video answers.

Worth noting: opening length does not move hook rate at all. The correlation there is −0.036 at p=0.76. Longer openings are not stopping more scrolls, they are earning more of the watch that follows.

The brief version: openings run 13 words minimum, and if the story benefits from 20 or more, extend it at no cost.

Let the creator speak from their own experience

The second attribute that resolved was opening framing, and it moves hook rate rather than cost per lead.

Openers framed as the creator's own experience ran 1.4 points above their brand median. Openers that addressed the viewer directly, the "you have to try this" construction, ran 2.8 points below at p=0.0031. Conditional openers, the "if you're a coffee person" construction, ran 3.0 points below at p=0.046. All three figures are deviations from the same reference point, so the gap between an addressed opener and a creator-statement opener is about four points of hook rate.

The same idea, executed either way:

• Instead of "You have to try this product," write "I tried this product and here's what happened."
• Instead of "If you're obsessed with coffee, you need to see this," write "I'm obsessed with coffee and I just found my new favorite."

Nothing about the product claim changes. What changes is who is doing the talking. This is a reframing rather than a prohibition, and the practical instruction is to pass the decision back to the creator, because their natural voice outperforms an imposed instruction mode.

Motion's data points the same direction from a different angle. Across their 578,750 creatives, confession-style hooks hit at 8.74%, second only to offer-led hooks and well ahead of bold claims at 7.19% and storytelling at 6.23%.

Three things you can stop briefing as performance levers

Three attributes were tested specifically because they show up constantly in creative briefs and brand guidelines. Within brand, none of them resolves.

Format. Demonstration, store trip and testimonial perform equivalently. Hook rate p=0.88, hold rate p=0.97, cost per lead p=0.88. Across the pooled dataset demonstrations did look stronger, but four high-cost-per-lead brands primarily brief demonstrations, so what looks like a format effect is a brand effect wearing a format's clothes. Pick the format that fits the product and the creator.

Retailer placement in the opening. Standard guidance says withhold the retailer from the first line so the video does not read as promotional. Hooks that name a retailer in the opening perform the same as hooks that name it later. Hook rate p=0.61, cost per lead p=0.73. If the creator would naturally say "I found it at Target" in the first line, let them.

Point of view. Whether the creator speaks as themselves or narrates about the product shows no measurable difference at any stage. Hook rate p=0.27, hold rate p=0.72, cost per lead p=0.098.

This is not a finding of no effect. It is a conclusion that the current dataset cannot resolve an effect if one exists. What it means practically is that the data does not constrain these three choices, so brief them according to your product, your creator and your brand guidelines, and spend the negotiating energy on the two that do move.

Why the comparison had to be within brand

The methodology choice doing the most work here is comparing every hook against its own brand's median rather than against a pooled average.

CPG brands differ enormously in what a lead and a purchase cost them, based on product positioning, distribution, retailer strength and shopper profile. Without that adjustment, a finding would mostly measure which brands happened to brief which creative format, not whether the creative choice itself works. The demonstration result above is exactly what that looks like when you skip the adjustment.

It also means every figure in this analysis is a deviation, not an absolute. A brand whose leads cost $8 and a brand whose leads cost $1.50 both contribute, and both are measured against themselves.

For external context on where hook rates sit generally, Billo analyzed 80,069 Meta video ads representing $105 million in spend between July and December 2025 and put the Food and Beverage category average at 25.47%, against a cross-industry average of 24.42%. Their broader point is worth keeping: category variance was 8.5 points across verticals while month-to-month seasonal variance was under one point, which is a good argument against chasing a generic benchmark.

What to put in your next creator brief

Four changes, in order of how much they move:

• Ask for openings of at least 13 words. If the story supports 20 or more, extend it. This is where cost per lead moves, and the gap between the short band and the rest is about 70%.

• Ask creators to open with their own experience. Remove direct-address constructions from the brief. Roughly three points of hook rate, and the idea does not have to change, only the construction.

• Leave format, retailer mention and point of view to the creator. No performance advantage attaches to any of them, and every constraint you add is a constraint on the creator's natural voice.

• Set a cost-per-lead baseline before launch and read it at day 14. That is the last stage the video still controls, and it reads within days rather than weeks.

If you want to run this on your own account, hold the offer, landing page, retailer, audience, budget and creative type constant, vary the opening only across three to five variations, and read cost per lead at day 14. If the test arm comes in below your brand baseline, the opening won.

The findings expire, the method does not

This is a September 2026 snapshot of Meta video ads for CPG brands driving to a retailer. Formats fatigue, platform incentives shift, and a 13-word floor is the right floor until the surface that rewards it changes.

What lasts is the approach. Code every creative by structure before you look at its numbers. Measure at every stage rather than at whichever stage the platform surfaces most loudly. Compare each hook against its own brand's median so you are testing the choice and not the brand. Then watch which patterns repeat across creators and quarters, and brief the ones that stick.

The full report, including the per-band tables, the significance testing on all five attributes and the test design for running this in your own account, is available here.