What the ad looks like
barely moves the number.

What is actually running in consumer feeds in 2026?

Across 243 consumer ads captured first-hand in 2026, eighteen distinct creative formats appear, and one of them is on three quarters of the library: static image, 182 entries, 74.9% (The Handoff, 243 published teardowns, August 2026). Nine formats carry enough rated pairs to quote a failure rate, and those nine sit inside a band 28.7 points wide.

Every entry was captured by hand: the creative as it ran in feed, and the page it clicked through to, opened signed out from a US IP. 216 of the 243 are pairs carrying a Continuous, Partial or Break rating on whether the page kept the ad's promise. 55 keep it, 110 weaken it, 51 lose it. That is the 74.5% base rate everything below gets measured against.

Format, placement and mechanic are controlled taxonomy fields and they are multi-valued, so an ad can be a demo and a static and a comparison at once. Shares therefore sum well past 100%, and a count is a tally rather than a judgement. The only judgement applied here is the floor: no rate is quoted below 8 rated pairs, and where the sample is thinner the cell says so.

Which ad formats are consumer brands actually running?

Static image at 182 entries is followed by demo or product-in-use at 53, UGC video at 51 and creator-led at 34. After that the tail goes flat fast: no other format clears 25 entries, and eight of the eighteen appear fewer than eight times in the entire library.

Four formats are too rare to list: reply-to-comment and challenge or stunt at 2 entries each, carousel and screen recording at 1. Before / after carries 8 entries but only 7 rated pairs, so it misses the floor by one and gets no number, which is the point of having a floor.

The temptation is to read 74.9% as an endorsement. It is not. Static image is what a brand produces when nobody decided anything: cheap, already in the design queue, and the substrate the other formats get built on. Its failure rate, 72.4% across 170 rated pairs, is two points better than the base rate. A format on three quarters of the library that performs like the library is a non-decision.

Does the format an ad uses predict whether the handoff survives?

Barely. The best format in the library is meme or low-fi at 58.8% failure across 17 rated pairs, the only one under 60, and the worst is comparison static at 87.5% across 16, with 2 Continuous ratings in the entire group. Everything else is stacked between them, which puts the whole format question inside a 28.7 point band.

Compare that to the other two things this library tags. The five placements with usable samples run from 71.4% to 81.8%, a spread of 10.4 points. The nineteen persuasion mechanics with usable samples run from 37.5% to 92.9%, a spread of 55.4 points. What the ad argues separates outcomes roughly twice as hard as what the ad looks like, and where the ad goes decides the severe cases outright: single-product creative sent to a shop rather than a product owns 24 of the 51 outright breaks on its own.

The mechanism behind the meme number is not that jokes convert. A low-fi creative usually carries one idea, and one idea is easy for a page to answer. BiOptimizers ran a line of cream serif on flat orange, "The Hidden Deficiency Behind Your Sleep, Cramps, and Brain Fog", with no bottle, no logo and no price, and the page answers it in its own headline: MAGNESIUM BREAKTHROUGH. Chubbies ran a beach video with no offer and landed on a homepage leading with swim, which is the honest destination for an ad that made no conversion ask.

The comparison static number runs the other way for the same reason. A comparison sets up two claims against a named rival, and 14 of its 16 rated pairs arrive somewhere that restates neither. Boll & Branch printed LINEN SHEETS. COOLER THAN SATEEN. STRONGER THAN BAMBOO. The page answers cooling with a Temperature line and a hot sleepers callout, then never mentions sateen or bamboo, so half the hook is abandoned at the door.

Where do these ads run, and does placement change anything?

This is a Meta library and should be read as one. Instagram stories carries 188 of 243 entries at 77.4%, Meta feed 173 at 71.2%, Meta reels 121 at 49.8%. TikTok is a distant fourth at 30. Across the five placements with 8 or more rated pairs, failure moves 10.4 points end to end, the flattest cut in the dataset.

Placements are multi-valued because most of these creatives run in several at once, which is itself the finding. A brand shipping one asset into feed, stories and reels has already decided placement is a distribution setting rather than a creative one, and the outcome data agrees: 71.4% on feed against 75.7% on reels is not worth a production budget.

TikTok at 81.8% across 22 rated pairs is the one cell worth watching, and it is thin enough that I would not spend against it yet. The plausible cause is not the platform. It is that TikTok creative in this library skews to UGC and creator-led, the two formats that fail at 82.1% and 80.6%, and those formats carry an argument a generic product page was never built to answer.

Which persuasion mechanics survive the click?

The mechanic an ad uses separates outcomes about twice as sharply as its format. Price objection handling fails 13 of its 14 rated pairs, 92.9%, the worst number anywhere in the library. Portfolio concentration, which means running one product hard rather than the catalogue, fails 3 of 8, 37.5%, on a sample sitting exactly on the reporting floor.

Read the top row as a warning about a specific move. Handling a price objection means naming a number, and a number is a debt the page has to settle in the first screen. Everyday Dose priced itself at $27 against a named RYZE pouch at $49, and the page prints $49 as its own preselected BEST DEAL, so the figure the ad taught you to reject is the destination's headline price. Eight Sleep sold Rent as long as you like. Return anytime, and the page opens on Pod 5 at $2,799 with Pay now already selected and Rent monthly beside it unclicked.

The bottom row is the mirror image, on a sample small enough to say so plainly: 8 rated pairs is a hint, not a finding. What it hints at is obvious once stated. A brand running one product hard has one page to keep in sync, so the argument and the destination cannot drift apart. Portfolio breadth is a merchandising asset and a paid media liability, and almost nobody prices it that way.

Which formats get stacked on top of each other?

Format is not a choice between eighteen options. It is a stack, and static image is the plank at the bottom of it. Four of the eight commonest two-format pairings are something bolted onto a static, led by demo plus static at 32 entries, 13.2% of the library.

Two production systems, and the table shows both. A design queue turns a product photograph into a demo, a listicle, a comparison or a fake article. A creator pipeline turns a person into UGC, a talking head or an endorsement. Creator-led plus static image at 13 entries is the seam, where a creator asset is flattened into a still and run as a card.

This is why chasing a format is the wrong unit of work. When a brand says it is moving to UGC, what usually ships is a second stack running beside the first one, aimed at the same generic destination. The formats multiply. The page does not.

Which categories over-index on which formats?

Two skews are strong enough to be conventions rather than noise. Before and after is 3.57 times over-indexed in Beauty and Skincare, where 4 of its 8 entries sit. Meme or low-fi is 3.15 times over-indexed in Food and Beverage, which holds 7 of its 20.

Beauty runs before and after because the result is visible. Food and Beverage runs memes because the purchase is small and personality substitutes for a spec. Supplements runs talking heads and listicles because it is selling a mechanism nobody can see. None of that is wrong. It is the local default, and a default arrived at by consensus is not a strategy a competitor cannot copy.

One tag-level skew is sharper than any of the format ones. 13 of the 23 listicle statics carry an offer in the creative, 56.5%, against 49 of 243 across the library, 20.2%. The listicle is being used as a discount vehicle rather than an education one, which is worth knowing before you brief a fourth one, and worth reading beside what happens to those offers at the click.

What should you change first?

In the order the numbers justify, which is not the order most creative reviews run. Four of these five are about the destination, because a 28.7 point format spread does not survive contact with a 55.4 point mechanic spread.

1. Stop briefing formats and start briefing claims. The format band is 28.7 points wide and the mechanic band is 55.4. Whatever you decide about video versus static matters about half as much as what the ad actually asserts.
2. Put every price, rival and percentage on the destination before the ad goes live. Price objection handling fails 13 of 14 and comparison fails 15 of 17. Both are failures of the page staying silent, not of the creative overreaching.
3. Give each stack its own destination. A demo static and a creator video on the same product are two arguments, and one generic page cannot answer both.
4. Treat placement as a distribution setting. 10.4 points across five placements does not justify cutting a bespoke asset per surface. Spend the hour on the landing page instead.
5. Price the cost of matching your category default. If everyone in Beauty runs before and after and everyone in Food runs memes, the format is not where your advantage comes from.

One item is deliberately absent: produce more creative. The 182 static images and 51 UGC videos here point at destinations written once, for everyone. A nineteenth format aimed at the same unchanged page just multiplies the arguments one fold has to answer, and the fold does not get bigger. If you want the gap read on your own funnel rather than on someone else's, the free store audit does it in a few minutes.

What does separate outcomes is the six-way taxonomy of what breaks at the click, and it is written down rather than left to judgement. Every figure above is recomputed from the same 243 captures, each one open to read with its artefacts attached.

Common questions

The five questions this data gets asked most, answered with the counts rather than with a preference. Every figure comes from the same 243 captures.