What Nutrinaut’s Source Question Revealed About Creator Marketing

An influencer campaign leaves records in several places. The published video shows what viewers could see. An email might explain the brief. A platform counter describes attention. Inside Nutrinaut, we also kept answers about where people had heard of the app, with an optional field for a creator’s name.

Coming back to those answers has been one of the more useful parts of reviewing our marketing experiments. They connect a creator’s public work with something submitted inside the product. They also make the weaknesses in our measurement much easier to see.

I checked a snapshot of the retained source logs on 3 November 2026. Here is what I can learn from it, and what I would change before relying on the same question for the next round of decisions.

First, work out what was actually recorded

The snapshot contained 7,391 source-event rows, dated from 15 August 2025 through 3 November 2026. After filtering for the known production app identifiers, the expected device-identifier format and timestamps no later than the snapshot, 7,384 rows remained.

Those are submission events. The same device can appear more than once.

There were 5,069 distinct device-and-app identifier pairs in the filtered records. For the platform comparison below, I retained the latest recorded source choice for each pair. This makes repeated submissions less dominant without pretending the identifier is a person: someone can use more than one device, and identifiers are not a verified record of installs or customers.

YouTube was the largest named platform in these records

“All other” combines the Other option, friends or family, Google, Facebook, Reddit, X and unknown values. The table uses one latest choice per technical pair, rather than adding every repeated event.

YouTube’s presence fits my experience of moving toward longer athlete videos, where a meal, a training session and a later app interaction could appear in one routine. That connection is an interpretation of our experience, rather than something the questionnaire proves on its own.

The table also describes the retained sample, not Nutrinaut’s complete acquisition mix. Logging starts well after our original iOS launch. In the Android checkout code we inspected, logging occurs when an influencer name is submitted, so other Android source choices can be missing. We do not have a complete account of logging behaviour across every deployed app version.

That is a concrete change I would make: record the source choice consistently before the optional creator-name step, and document which app versions do so.

YouTube44.6%

TikTok21.5%

Instagram7.5%

Other26.4%

An optional name produces a separate dataset

The companion creator-name logs contained 2,116 response events, associated with 1,996 distinct device-and-app pairs. These rows add detail to source answers; adding them to the 7,384 source events would count a different step as more acquisition.

Free text brings its own work. The records contained 641 distinct raw creator strings. Case folding and privacy filtering left 536 distinct text values. That does not mean we worked with 536 creators. A handle, a real name, a spelling variation and an unrelated answer can all become separate strings.

For named comparisons, I used conservative matches to identities we could establish from public accounts or our own collaboration records:

These are lower-bound matches within the retained answers, using exact normalized identity aliases rather than guessing at every first name. The device columns should not be added into a total: a pair can name different creators at different times.

Adrian Lee and BB Kickers are particularly important to keep separate. BB Kickers is Adrian Colon’s college-soccer content; Adrian Lee is the MMA fighter. Collapsing both into “Adrian” would damage the comparison before we even began interpreting it.

Hannah and Yash show why the identity work matters

The creator I remember from our motherhood and lifestyle experiment is Hannah Hiatt, @hannah_bhiatt. Our historical planning sheet contains an outreach draft for that handle, including a proposed continuing series of food or activity integrations. A proposal for four videos in a month is still a proposal; it cannot supply four missing publication records.

I also found correspondence with a different person named Hannah. Combining the two because their first names match would create a relationship the records do not establish. The questionnaire contains a historical nurse-Hannah handle too, which needs identity review before it belongs in a named performance table.

Boxing creators Yashua Flemming and Dylan Gibson, known as Yash & Dylan Boxing, illustrate the other side of the problem: a joint account can be remembered through either person’s name. Their own website establishes the names behind the account. The conservative family in the table covers the joint account’s name and Dylan Gibson’s full name, while leaving ambiguous bare first names unresolved. It contains 92 TikTok-associated answers and 21 Instagram-associated answers, dated from 18 August 2025 to 28 May 2026. That does not establish the exact Nutrinaut post, delivery date or financial result. I would keep the public-post gap visible in our campaign register.

What I would keep, and what I would improve

I would keep the question. It catches a type of memory that a link label alone cannot explain: the creator or platform someone chooses to name inside the product.

I would make source logging consistent across platforms and app versions, keep the creator field optional, and preserve an explicit “unknown” category instead of guessing. For confirmed campaigns, I would keep the creator’s stable identity alongside their current handle, the exact published links and the dates of the work. Raw personal text and device identifiers belong in private records, not a public campaign table.

For financial decisions, I would join this evidence to verified subscription outcomes and actual costs through an appropriate measurement process. The source answers alone cannot tell me a creator’s return on spending.

What they can do already is improve the next conversation. Instead of asking whether a creator “worked,” I can ask whether viewers remembered the name, whether we recorded their route consistently, and which part of the evidence still needs to be connected.

---

Method and disclosure: I work on Nutrinaut. These are our first-party, unauthenticated client-submission records, captured on 3 November 2026 at 10:13 UTC. Counts describe the retained questionnaire sample, with the coverage limits stated above. They are not verified people, installs, trials, payments or causal attribution. No raw device identifiers, personal responses or private correspondence are reproduced here.