End a video ad with an ask, not a summary. Only a minority of viewers reach the last frame, and the ones who do are your warmest audience, so the ending should give them a single action with the offer restated rather than recapping what they just watched. That means: one verb, one destination, the terms, held on screen for at least four seconds. Anything else in that slot is spent on people who already decided to stay.
The completion constraint
Start from what actually happens. Our benchmark composite of studies covering more than 12,000 ads puts hook rate (three-second views over impressions) at 28 percent on Meta feed and Reels, 33 percent on TikTok and 22 percent on YouTube in-stream. That is the share still there at second three. The share still there at the final frame is a fraction of that fraction.
Across the accounts we work with, a 15 second feed ad typically finishes with somewhere in the low tens of percent of impressions, and a 30 second one considerably less. Treat those as our experience rather than published figures, and measure your own: what completion rate measures and how to read it covers the definition and its traps, the main one being that platforms count a "completion" differently from each other.
| Placement | Median hook rate | Who is left at the end | What the ending is for |
|---|---|---|---|
| Meta Reels | 28 percent | A small slice of a quarter | The ask, plus the offer repeated |
| Meta feed | 28 percent | Similar, skews warmer audiences | The ask, plus risk removal |
| TikTok | 33 percent | More viewers in, faster exits | The ask, said fast, plus the on-screen card |
| YouTube in-stream | 22 percent | Skippable, so finishers self-selected | The ask, and it can be longer |
The design conclusion follows from the third column. The ending is not a summary slot because the people reading a summary sat through the thing being summarised. It is a conversion slot, and the only question worth answering in it is "what do I do now".
Why summarising wastes the ending
Three reasons, in the order they cost you money.
First, it addresses the wrong audience. A recap is written for someone who missed the middle, and that person is not watching. Second, it occupies the one slot that has a metric of its own. The CTA is judged on conversion rate, not watch time, and a recap produces watch time. If you are unclear which slot owes you which number, how the hook, body and CTA get measured separately is the model we use. Third, it delays the instruction to the point where the ad ends before the viewer knows what to tap.
You can spot the failure in a draft immediately. If the last spoken line starts with "so" or "whether you are", you have a summary. Cut the sentence and start with a verb instead.
The other common waste is the brand logo card with no words. A logo is not an action. If the end frame has your mark and nothing else, you have paid for four seconds of brand recall from an audience that was ready to buy.
Three ending patterns
Pick one per ad. Mixing two is how a four-second ending turns into eight.
1. The direct ask
Verb, destination, terms, in one line. "Start with your two busiest ledgers at ledgerly.com, no card needed." On screen: "CONNECT ONE LEDGER" and underneath "FREE, NO CARD".
Use it as the default. It wins on cold traffic and on anything measured by cost per acquisition. The only real skill is compression: get the terms in without a subordinate clause.
2. The last objection, then the ask
Answer the one thing that stops people, then ask. "Thirty-day returns and we collect the box. Shop the shelf." Or for software: "Two ledgers, one afternoon, cancel whenever. Connect your first one."
This is the strongest ending for retargeting, because you know from the previous ad which objection is live. Delivery, contract length, setup time and returns are the four that come up most. Keep it to one objection: two reads as defensive.
3. The forward promise
Name what happens after they act, in concrete terms. "Tap through, first box is £9 and it is on your doorstep Monday." Or "Book the call, we send the recording and the working file afterwards."
This works when the friction is uncertainty about the next step rather than the product itself. It is also the best ending for a long consideration cycle, since it converts an abstract decision into a scheduled event.
Where the offer really belongs
Not at the end. In the first half, and then again at the end.
Put the offer before the halfway mark (by second 7 of a 15 second ad, by second 13 of a 30) so that everyone who leaves in the middle still heard the trade. Then restate it in the CTA, because the viewer at second 12 may have joined at second 8 and the repetition is their first exposure. Repetition inside a 15 second ad is not a stylistic flaw, it is coverage.
There is a diagnostic benefit too. When the offer appears twice, a weak result is easier to attribute: strong watch time with weak clicks points at the offer itself, whereas clicks that do not convert points at a CTA promising something the landing page does not repeat.
The mechanics that get this wrong in the edit
Endings fail on timing more often than on wording.
- Hold the end card four to six seconds. A URL read once at speed is not remembered.
- Do not fade. A fade to black on a looping placement means the loop restarts before the card registers.
- Keep the card out of the bottom of the frame on 9:16, where the interface sits over it.
- Match the words on the card to the words on the button below the ad. A mismatch reads as a different offer.
- Do not put the price on screen for the first time in the last two seconds. That is a new claim with no time to be believed.
Our notes on CTA clip design go into frame layout, hold times and safe areas. The reason we build the ending as a separate clip rather than the tail of a timeline is swap cost: an ending you can replace on its own can be tested against the same body, and a batch of eight variations can carry two endings without any re-editing.
Genyad, our product, generates each variation as a fresh script with its own ending, shot selection, voiceover and captions rather than a re-cut of one master, exported in 9:16, 4:5, 1:1 or 16:9. Two things it does not do, since they come up here: no predicted score telling you which ending will win, and no publishing into Meta or TikTok on your behalf. You export the files and the platform tells you.
How often to change the ending
Less often than the hook. Hooks decay quickly: our 2026 fatigue benchmark reports CTR falling 15 to 20 percent in the first two weeks, week three landing 45 to 70 percent below launch, and a 45 percent CTR drop after a fourth exposure to the same creative. A working ending is more durable, because the people reaching it are self-selected and the offer is what they are responding to.
Practically: test two endings at a time against a fixed offer, keep the winner, and spend the rest of your variation budget on the first three seconds. When you change the offer, the ending gets retested. Otherwise leave it alone.
Frequently asked questions
How long should the ending of a video ad be?
Three to five seconds of the ad, with the end card held four to six seconds where the format allows. Shorter than three seconds and the instruction does not land; longer than six and you are paying for frames after the decision has been made.
Should a video ad end with a summary of the benefits?
No. The people who reach the end watched the benefits, and the recap costs you the slot where the instruction belongs. Give them one verb, one destination and the terms instead.
Should the offer be at the end of the ad?
The offer belongs before the halfway point and then again in the ending. Stating it only at the end means everyone who left mid-ad never heard the trade you were offering.
Does a logo end card work as an ending?
Not for direct response. A logo with no words gives an audience that was ready to act nothing to act on. If you want brand recall, put the mark on the card alongside the ask rather than instead of it.