Your influencers might be getting you attention and sales. They are pulling your brand apart.

We put twenty-one brands' own advertising and 695 of their influencer posts on the same Hotspex Map, blind, with the same instrument. Nine US sportswear brands, six Unilever brands, six P&G brands. For nineteen of the twenty-one, the influencer content clusters less tightly than the brand's own advertising, sits off it, and lands in the same register whoever the brand is. Nike and Gillette are the sharpest cases. Axe, Old Spice and Pampers hold.

First, the instrument

The fall of Nike

It is not just Nike, and it is not just sportswear

Every brand's creator content, next to its own advertising. Gold is how tightly the brand's ads cluster; red is how tightly its influencer posts do. In sportswear it is all nine; at Unilever and P&G, all but one.

Coherence = mean resultant length of the leg's attribute vectors, 0 scattered to 1 single-minded. Bars show bootstrap 95% intervals on the creator minus ad difference; p is a label-permutation test.

Median degrees off the brand. Sportswear brands are measured from the six-touch-point core (ads against a core that excludes them); Unilever and P&G brands from the advertising centroid (ads leave-one-out), because those brands are not yet in the league.

The creators do not scatter at random; they converge. Sportswear creators read Like you, Active, Youthful, Cool, Friendly; Unilever's read Youthful, Like you, Interesting, Practical, Trendy, Friendly. Eight of the twelve most common creator words are the same across those two studies. P&G's creators read Like you, Practical, Friendly, Witty, Family-oriented, Interesting: five of the sportswear twelve, and the rest are household words, Knowledgeable, Straightforward, Nurturing, that a laundry or diaper creator brings whoever the brand is. The brands say Inspiring, Confident and Reliable in their mission language. The creators deliver Like you.

Distinctive assets do not pull a post back. In sportswear, posts carrying a logo, tagline or signature product sit 49° from the core and posts carrying none sit 48°. At Unilever, 43° against 45°. At P&G, 70° against 50°, a gap the sample cannot tell from zero (p 0.39; it would need 33° to register). The logo buys recognition, not territory. In CPG the pack does the branding: the signature product appears in two thirds of posts and the logo in fewer than one in five.

What the people buying this media get back is delivery and conversion, the same day, and same-day numbers are hard to argue with. A platform can see a sale this week. It cannot see a brand. That gap is the layer this page measures.

The one that gets it right

Axe's ads read Witty, A joker, Interesting, Different. Its creators read Youthful, Witty, Cool, Interesting, A joker, Confident. The two centres are 6° apart, the creator posts cluster as tightly as the ads (0.71 against 0.71), and the median post sits 14° from the advertising, closer than the ads are to each other. Two things are visible in the data: 38 posts come from 22 creators, so voices repeat, and 84% show the signature can or the brand's colour. Whether that comes from a tighter brief we cannot see from here, but Axe is the only one of twenty-one where the creator layer and the advertising are the same thing. Old Spice, selling the same thing with the same joke, comes closest at P&G.

New Balance is the sportswear counter-example: creators 14° off the core at 0.55, Singularity barely moving. Neither is an accident. One voice repeated builds a territory; forty voices once each build the creator's.

How we did it

1. Sportswear: the nine brands' existing Hotspex Map placements from the North American Singularity league, 436 pieces across advertising, website, packaging, product, retail and sponsorship, plus 348 influencer posts and 39 brand statements collected from the open web with a source link each.
2. Unilever: six brands' own advertising from 2024 to 2026, 72 items, plus 209 influencer posts and 27 brand statements, collected the same way.
3. P&G: six brands' own advertising from 2024 to 2026, 64 items, plus 138 influencer posts and 27 brand statements, collected the same way, 18 September 2026. Duplicate entries and items sourced from press releases rather than the post itself were removed before placement was analysed.
4. Brand names, handles and hashtags scrubbed; brands mixed together in batches; the same AI instrument that scores the league placed every item three times, keeping only what two of three agreed on.
5. Measured where each brand's influencer content sits against its own advertising and its core, how tightly it clusters, whether a visible logo changes that, and, for the sportswear brands, what it does to the Singularity score.
6. Had simulated Sharp, Ritson and Binet try to break the sportswear study, and changed what they broke.

How much to trust it

• This is a synthetic read. An AI model placed the content on the Map; no consumer panel saw it.
• Match rate. On the metric most people quote, and on a different Hotspex instrument measured across 6,674 real survey cells, synthetic respondents match real cells 86.5% of the time. Two sets of real people match each other 94.9%. A model that ignores the stimulus entirely matches 83.1%, so a match rate on its own flatters everything.
• Skill. The number that cannot be flattered is agreement on how content moves a brand away from the average. There we have one benchmark, and it is a sibling instrument, not this one: on 197 banking ads, Claude Sonnet 5 scored r +0.161 at zone level against a human ceiling of 0.308 with the brief and the creative, and +0.092 with the creative alone. The placer used here is claude-opus-5 working from descriptions with no brief, and it has not been put through that test. Treat +0.092 to +0.161 as the nearest measured range and this model as unvalidated for these categories, for influencer posts and for brand statements.
• For most posts we could not open the video itself, because the platforms block automated fetches: 342 of 348 sportswear posts, 198 of 209 Unilever posts and 134 of 138 P&G posts were placed from the caption and the coverage about the post. The brands' own ads re-placed from caption-grade text keep 0.64 coherence against 0.72 from full text; influencer posts sit at 0.43. Thin text does not explain the gap.
• What we would bet on: the direction, which repeats nineteen times out of twenty-one, and the register, which is the same for every brand. What we would not bet on: where any single post sits, the size of any one brand's Singularity drop (inside the ±12 band), the stated-brand angles for Dove and Vaseline, whose four to six statements barely agree with each other, or any degree figure for Tide and Dawn, whose own advertising is too scattered to anchor one.
• Three of the P&G posts and eight of its ads were placed from inferred material rather than sourced coverage; the rest of the site's items were placed from a source that described them.
• Nothing here measures whether any of this content sold anything.

What to do about it

• Fewer creators, longer relationships. One voice repeated builds a territory. Forty voices once each build the creator's. Creator agencies serve two masters, keeping the creators happy and reporting on delivery. Holding a territory is nobody's job in that chain unless the brand makes it someone's.
• Make the thousand ads and solve the singularity problem at the same time. Personalisation at scale is available to everyone now. The brand that wins is the one whose thousand pieces still point the same way, because people eventually walk into the store or open the app, and what the content made them feel has to match what they find.
• Brief the territory, not the product. Give creators the three or four words the brand owns and score their posts against those words before they run. None of this is the creators' fault: most were briefed on a product, not a territory, and did what they were asked. Reach is set by interest-based distribution now, not follower counts, and the volume of content is going up, not down. The brief is the part of this you still control.
• Stop leaning on the logo. Distinctive assets get you recognised. They do not move where the post lands. In CPG, make the pack the asset creators cannot leave out.
• Put influencer content on the same map as the ads every quarter and treat it as a seventh touch point. Today it is the one layer nobody measures.

Synthetic read. Placement by claude-opus-5, three passes, two-of-three pooled, brand names scrubbed, norm of 4,110 North American advertising executions; food-and-beauty calibration on the Unilever and P&G runs. No measured accuracy for this placer; nearest benchmark is Claude Sonnet 5 on 197 banking ads, zone-level r +0.092 to +0.161 against a human ceiling of 0.308. Public Nike figures verified 14 September 2026. Disclosure: Hotspex builds the Map used here and sells the measurement the last bullet recommends. Built 18 September 2026 · hotspex.com