1.6 Who to Study, and How to Do It Honestly
Benchmarking in this field is usually done badly in one of two ways: admiring the famous accounts whose conditions you cannot reproduce, or copying the nearest competitor's format until your account is indistinguishable from theirs.
There is a third method, and it is the one this chapter teaches. First you classify the models correctly, then you audit them systematically, then you extract the transferable structure and leave the surface behind.
The six models worth studying
Each of these represents a different business, and the lesson is different in each case. The figures are approximate and point-in-time. [UNVERIFIED]
Lu do Magalu — the brand asset. A retailer's synthetic spokesperson, deployed inside a company with an existing audience and media budget. Lesson: a synthetic face is a distribution channel, and its value is the size of the channel it is attached to. Not transferable: everything about the audience, which the retailer already had.
Lil Miquela — the entertainment property. A character with years of narrative continuity, managed as a performance rather than as a content calendar. Lesson: continuity over time is the asset. Transferable: the discipline of a canonical character with a consistent voice and history across years.
Shudu — the visual-first model. A persona whose identity is essentially editorial: a defined aesthetic and a photography register rather than a personality-driven feed. Lesson: a content model can be a legitimate business when the visual register is distinct enough to license. Transferable: how a narrow visual signature creates recognition.
Aitana López — the operator-scale persona. A synthetic model doing brand work at a scale resembling a small studio rather than a media company. Lesson: this is the closest reference for a solo operator, and its economics are closest to a freelancer's. Transferable: the working scale, the brand-facing positioning, and the pace.
Emily Pellegrini — the subscription-fronted persona. A persona operating as the front of a subscription relationship rather than as a media property. Lesson: in the current numbers, this is where the highest revenue per follower sits in the synthetic category, and it changes what "growth" is for. Transferable: the funnel-first structure from chapter 10.1.
neuro-sama — the live-presence character. A virtual performer built on live interaction rather than photography, whose audience relationship rests on behavior. Lesson: the character outperforms the render. Transferable: the recognition that voice and behavior carry more durable attachment than image quality, and the willingness to build for presence rather than for stills.
Read as a set, the six argue one thing: the durable examples are built on continuity and voice, and the visual fidelity is table stakes. Note the tension with the seven chapters of Part IV — face consistency is necessary and it is not the differentiator. The accounts that survive are the ones whose persona would still be recognizable if you described them without a picture.
The reverse teardown: twelve points
Pick three accounts — one above your target size, one at it, one below. Then audit each on these twelve dimensions, and write the answers down. The writing down is the method; scrolling is not research.
1. Cadence. Posts per week, and whether it is consistent or bursty. Note the day-of-week pattern, which usually reflects the operator's schedule rather than an audience insight.
2. Format mix. Proportion of video to still, of talking to silent, of lifestyle to product. Most successful accounts have a dominant format and one or two accents, not an even spread.
3. Framing. Preferred shot sizes and camera heights. Consistent accounts reuse two or three framings; inconsistent ones have none.
4. Lighting. One setup or many? Which one recurs in the highest-performing posts?
5. Hook structure. The first three seconds, transcribed. What promise or question appears before anything else?
6. Length. Distribution of durations, and where the account's outliers sit on that distribution.
7. Caption voice. Sentence length, use of first person, emoji or none, whether questions are asked. This is the cheapest thing to learn and the most commonly skipped.
8. Call to action. How often, what kind, and where. An account with a CTA on every post and one with none are different businesses; identify which you are looking at.
9. Link in bio. What it points to. This single field tells you the account's business model more reliably than anything else on the profile.
10. Comment behavior. Does the account reply? As whom — the character or the operator? Do the comments contain the "is she real?" thread, and how is it answered?
11. Content pillars. Cluster the last fifty posts into three to five themes. If you cannot cluster them, the account does not have pillars and is likely a hobby.
12. Monetization signals. Sponsored posts, Promoting someone else's product for a commission: the cheapest first dollar, the weakest position. patterns, product mentions, subscription platforms, a shop. Note frequency, not merely presence.
Twelve points times three accounts is a couple of hours, and it produces a document more useful than any course in this field. Chapter 8.4 uses the same structure in reverse for planning your own launch.
Teardown versus copying
The distinction, stated as a rule you can apply per item:
Copy structure, never surface.
Structure is transferable: a four-pillar content model, a three-post-per-week cadence, a hook that opens with a question, a link-in-bio that lands on an email list. Structure is how the machine is built.
Surface is not: the persona's face, their wardrobe, their catchphrase, their visual signature, the specific compositions that make their feed recognizable. Surface is what the audience already associates with them.
The test for any element you are about to take: would removing it leave your account looking like theirs? If yes, it is surface. If it would leave your account working the same way while looking like yours, it is structure.
There is also a hard boundary here that a rule of thumb cannot cover: copying a real person's likeness, or a competitor's distinctive character design, is a rights problem, not an etiquette problem. Chapter 11.4 governs it.
Building a reference board without stock aesthetics
The last step, and the one where taste is manufactured.
Collect structure, not images. Your board should be mostly text — cadence tables, hook transcriptions, pillar clusters — with images only where a specific framing or lighting setup is being studied. An image board quietly turns into a mood board, and a mood board reproduces the aesthetic of whoever you collected.
Deliberately include non-synthetic references. If every account on your board is an AI persona, your persona will look like an AI persona. Bring in photographers, film stills, editorial work, and non-influencer sources for the visual register.
Name the three things you are not doing. For each reference, write one sentence about what you are specifically not taking from it. This sounds like a writing exercise; it functions as a brake, and the accounts that skip it are the ones whose feeds converge on one look.
Re-audit quarterly. The reference accounts change — some pivot, some get acquired, some get labeled and lose reach. A board that is not refreshed becomes a description of a market that no longer exists.
Where this goes wrong
Studying only the famous accounts. Their conditions — budget, team, existing audience — are not yours, and their methods often do not transfer.
Mixing up models and copying the brand asset's playbook. Lu do Magalu is not a solo operator template.
Building an image mood board and calling it research. You have collected surface.
Skipping the caption-voice audit. The cheapest, most transferable, most skipped item on the list.
Not writing anything down. Memory of a feed is not a teardown.
Copying surface and calling it inspiration. The convergence is visible to audiences before it is visible to the operator.
The gate
- Which three accounts did you audit, at which size tiers?
- Which of the twelve points produced the most surprising answer?
- For your strongest reference, name three things you are deliberately not taking.
- Does your board contain any non-synthetic visual references?
- Which structural element from the audit is going into your own plan, and where?
What you leave with
Six models classified by the business they actually represent, a twelve-point reverse teardown you can run on any account in twenty minutes, the structure-versus-surface rule with its removal test, and a method for building a reference board that produces a distinctive persona instead of a reproduction of whoever you studied. Part II turns that into a character.