Help us prove a creative intelligence model against public outcomes.
SaliencyLab predicts engagement and click intent on TikTok, YouTube, and Meta from the ad itself, before media spend. The Research Fellows program is how we cross-validate that prediction against real public-platform performance, attribute by attribute, cohort by cohort. Fellows do the work that becomes a co-authored paper.
A measurement tool that won’t show its math is just another vendor.
Most pre-spend creative testing happens behind a six-figure invoice and a 90-page deck. SaliencyLab does the opposite: published methodology, open KPIs, public-outcome validation, and a benchmark dataset that will eventually ship under CC-BY-NC. We need help building the validation layer. That is what fellows do.
Not coffee runs. Actual validation studies.
You will score real ads through the pipeline, code attribute presence, cross-check LLM judgments against ground truth, and help build the test-retest reliability dataset that goes into our methodology paper.
Your name on the methodology paper.
Top-contributing fellows are invited as co-authors on the validation paper (target: Journal of Advertising, Marketing Science) or the dataset paper (Scientific Data). Honorary authorship is explicitly forbidden, contribution is logged.
Direct contact with the ad-tech research stack.
You will learn how Kantar Link AI, System1, and Zappi structure their constructs, and how a small lab can validate against TikTok Creative Center, Meta Ad Library, and Google Ads Transparency without a panel budget.
Six weeks. Four deliverables. One published artifact.
The program is structured. You know what is due, when, and why it matters to the research line. We protect your time, you protect the science.
We hire for curiosity and rigor, not pedigree.
You do not need a PhD. You do need to like the kind of work that does not look glamorous on a slide: reading methodology papers, arguing about construct validity, manually coding 50 TikToks because the LLM got something wrong.
Strong fit
- Undergrad, masters, or PhD in marketing science, behavioral science, statistics, ML, cognitive science, or media studies.
- You have written at least one piece of work (paper, thesis chapter, blog) where the conclusion was "the data did not support the hypothesis."
- Comfort with basic stats: correlation, percentile, confidence intervals. Python or R helpful, not required.
- You can commit ~8 hours a week, on your own schedule, for six weeks.
- You are willing to have your contribution logged publicly and your name attached to the work.
Not a fit
- Looking for a paid summer intern role. Fellows are unpaid, contribution-based, with named credit and reference letters.
- Want a generic "research experience" line for a CV without doing the actual coding/curation work.
- Cannot commit eight hours a week reliably. We would rather you decline than ghost mid-cohort.
- Hoping to pivot the program toward your unrelated thesis. The work is the validation pipeline, not your topic.
Tell us who you are and why you want in.
Six fields. No CV upload. We read every application personally and respond within 5–7 business days.
Frequently asked.
Is the fellowship paid?
No. Fellows are unpaid contributors. What you get instead: named credit on a validation dataset and methodology paper, reference letters, and a direct working relationship with the founder. We are honest about this up front because we hate the trick where an unpaid role is dressed up as a “learning opportunity.”
Will my name actually be on the paper?
Only if you contribute paper-grade work. We follow ICMJE-style authorship criteria. Fellows whose contribution does not meet the bar are credited in the acknowledgements section with a description of what they built, not falsely listed as authors. Honorary authorship is explicitly forbidden.
Do I need to be a PhD student?
No. The current cohort includes a final-year undergrad in cognitive science, a masters student in management sciences, and a PhD candidate in marketing. Fit is judged on the "the data did not support the hypothesis" criterion above, not the title on your degree.
What if I cannot finish the six weeks?
Tell us early. We would rather you withdraw at week 2 than disappear at week 5. Withdrawing in good faith is logged neutrally and does not affect future cohorts.
What is the validation paper's target journal?
The flagship validation paper is targeting Journal of Advertising or Marketing Science. The dataset paper is targeting Scientific Data. The LLM-as-judge inter-rater reliability paper is targeting an information-systems venue.
Who is the principal investigator?
Oussama Nakhil (founder, ex-L'Oreal Global Consumer Insights Analyst, ex-NielsenIQ) leads the validation work and is the corresponding author. Nissrine Khadir (Mohammed V University Rabat, management sciences) is the academic co-author on the dataset and flagship methodology papers.