Winning creatives cluster by emotion, not mechanic
Tag your winners by the feeling they sell rather than the mechanic they show, and your portfolio looks far less diverse than the concept count suggests.
Pull the top twenty performing creatives from any mature mobile game account and tag them twice: once by the mechanic on screen, once by the feeling the first three seconds sell. The mechanic tags scatter. The emotion tags don't, and in my experience the winners sit in two or three emotional registers no matter whether the footage shows sorting, shooting, merging, decorating or something else entirely.
That observation leads to a claim many creative directors will resist, because it undercuts how most concept libraries organise themselves. The unit of creative performance isn't the mechanic; it's the emotional state on offer to the viewer, and a portfolio that counts its diversity in mechanics measures the wrong thing.
Why the viewer buys a feeling
A studio sees its game as a set of systems, so its creative taxonomy inherits that view: core loop concepts, meta concepts, narrative concepts, fail-state concepts. The viewer has no access to any of that. In the first three seconds they decide whether the promised feeling is one they want right now, which might mean relief, competence, mild dread, smug superiority over an on-screen character, or the specific calm of watching something get tidied.
Network delivery models learn from the responses they see; those responses come from people reacting to a feeling. Two creatives that show completely different mechanics but sell the same feeling will, in the model's eyes, work on overlapping audiences. Two creatives with the same mechanic but different registers, a tense timer versus a serene sort, behave like different products.
AppsFlyer's State of Gaming for Marketers 2026, published last week with Unity and Newzoo, puts top advertisers at somewhere between 2,400 and 2,600 creative variations a quarter. Look at how teams usually generate those variations: swap the mechanic footage, change the fail state, re-skin the character, trim the intro. Most of that volume is mechanic-level variation inside an unchanged emotional register, which amounts to a lot of new files and very little new promise.
The fatigue you cannot see in the concept count
The second-order effect is the one that matters for budgets. Fatigue attaches to the feeling, not the footage.
If your five "distinct" concepts all sell competence through effortless success, they compete for the same viewers and wear out the same appetite. When performance rolls over, it rolls over across all five within a few weeks, and the team concludes the market has gone soft. It hasn't. One emotional register ran dry, and nobody had diversified the portfolio against that risk because nobody was tracking it.
This also explains a pattern buyers have described for years without a clean account of it: a visually unrelated creative, pulled from a different genre's playbook, suddenly outperforms everything in the account. Better craft? Rarely. It was simply the first new register the audience had seen in a quarter.
A UA Ledger piece from last summer, "Creative Hook Variety: Designing Ads for Mobile Games", argued for hook variety as a production discipline. The refinement I'd add is that variety only counts if you score it on the emotion axis.
A two-axis tagging scheme
The fix is administrative and slightly tedious, which is why it doesn't get done. Every concept gets two tags at brief stage; the second one governs portfolio decisions.
The mechanic tag is whatever your team already uses.
The emotion tag comes from a short, fixed list. Keep it under ten entries or it stops being useful. A list that has worked for casual and mid-core alike:
- Relief: chaos resolved, mess cleaned, threat averted.
- Competence: the viewer would obviously do this better than the character on screen.
- Tension: a timer, a near miss, an imminent fail.
- Curiosity: a hidden thing, a reveal withheld.
- Progression: the small base becomes the big base, the number goes up.
- Social standing: beating a rival, being seen to win.
- Comfort: cosy, slow, decorative, no stakes.
Then apply one decision rule. A concept counts as new to the portfolio only if it introduces an emotion tag not currently live at scale; a new mechanic inside a live register is a variant, so budget and test it as a variant rather than as a concept.
A worked illustration
Suppose a team runs eight concepts and believes it has a diversified slate. With purely illustrative labels, the tagging comes back as five in competence, two in relief, one in progression. On the emotion axis that's a three-register portfolio with 62 percent of the concept count, and let us say 80 percent of spend, sitting in one register.
The right move isn't to kill the competence winners. It's to recognise that the next brief must not add a sixth, so the production slot goes to tension or curiosity even if the mechanic footage is the same board the competence ads already use. When the competence register eventually tires, the account has somewhere to shift spend that the network has already learned.
The same audit works in reverse for post-mortems. When a concept fails, check whether its register is already crowded in your own account. A perfectly good relief ad launched into an account running four relief winners isn't a bad ad. It's late.
What this does to AI variation tooling
One more consequence. Most teams feed generative variation tools the winning creative and ask for more of it, and if the input carries a mechanic tag the tools will helpfully produce a hundred mechanic swaps inside the same register, so the account drowns in files while its emotional diversity shrinks.
Tag the register first, then brief the tooling on the register you're missing. The prompt "give me a tension version of this board" produces something the portfolio needs. "Give me twenty variants of this winner" produces the fatigue curve a fortnight early.
