AI OFM research and benchmarks

This section publishes measured experiments about AI influencer production and distribution. Every study states its question and method before its result, and no study is listed with numbers unless those numbers were actually measured.

How we test

One variable is changed at a time against a fixed baseline. Outcomes are measured on the downstream metric that matters - profile visits, link clicks or conversions - rather than on views or engagement. Sample size and time window are declared with the result, and negative results are published alongside positive ones.

Current status

The published case studies below report documented figures. The controlled studies listed after them have not completed data collection, so they are announced with their question and method only.

Published case studies and benchmarks

Broader student evidence

  • Examples of Anonymised AI Uncensored Student Outcomes - Anonymised earnings dashboards and spending panels from individual AI Uncensored student accounts, published exactly as displayed. These are individual examples, not average results, and not attributed to any case-study account.

Planned research

  • Instagram hook test: which opening frame earns profile visits (planned) - Question: Across identical clips, which opening-frame style produces the highest profile-visit rate on an AI influencer account? Method: Same base clip, three hook variants, posted in rotation from one account over a fixed period. Measured on profile visits per view rather than total views. Metrics to be reported: Views, Profile visits, Profile-visit rate, Follows, Link clicks where available.
  • Consistency benchmark: image models across a fixed five-scene test set (planned) - Question: Which image-generation setups hold a persona's identity across indoor selfie, daylight outdoor, low light, mirror shot and full body? Method: One persona dataset, identical prompt skeleton, blind side-by-side identity scoring of outputs from each setup. Metrics to be reported: Generations produced, Generations judged usable, Identity-consistency rate, Failure categories, Scene complexity. Disclosure: AI Uncensored used its standard internal production workflow. Certain implementation details are proprietary to its mentorship programme and are therefore not disclosed, so this study is not fully reproducible from the published method alone.
  • Image-to-video drift benchmark: identity retention over clip length (planned) - Question: How quickly does facial identity drift as generated clip length increases? Method: Fixed source stills animated at increasing durations and motion strengths; final frame compared against the source.
  • Reddit study: which subreddit categories produce profile clicks, not just upvotes (planned) - Question: Where does Reddit traffic actually convert into profile visits for AI personas? Method: Tracked links per subreddit over a fixed posting period, reporting click-through per post alongside upvote counts. Metrics to be reported: Posts, Upvotes, Link clicks, Clicks per post.
  • Production benchmark: what it actually takes to fill a posting week (planned) - Question: How many assets must be generated, and how much time spent, to produce one week of daily posts for a single AI persona? Method: Time and output logged across consecutive production sessions for one persona, counting every generation rather than only the published ones. Metrics to be reported: Total assets generated, Usable assets, Rejection rate, Production time, Posts produced. Disclosure: AI Uncensored used its standard internal production workflow. Certain implementation details are proprietary to its mentorship programme and are therefore not disclosed, so this study is not fully reproducible from the published method alone.
  • Distribution experiment: how posting approach changes downstream traffic (planned) - Question: Across the same persona and content library, how do different distribution approaches affect reach and follower acquisition? Method: Fixed content library distributed under different approaches over matched time windows, compared on downstream traffic rather than reach alone. Metrics to be reported: Reach, Profile visits, Followers acquired, Downstream link traffic.

What we disclose and what we withhold

Every study states what was measured, the sample size, the date range, the scoring method, the definitions used and the limitations. Some studies run on AI Uncensored's internal production workflow, and the implementation details of that workflow are proprietary to our mentorship programme. Where that applies we say so on the study itself rather than claiming full reproducibility.

The workflows these studies test are documented in the AI OFM guide and our methodology.