Most celebrity net worth figures on the internet arrive with no math attached. A site prints $400 million, cites nothing, and other sites quietly copy the number until it hardens into accepted fact. Nobody can check it because there is nothing to check.
We built this site to work the other way around. A figure only goes up when every input and every source behind it can go up with it. Our methodology page holds the current rate tables, the sourcing rules, and the accuracy scorecard. This post is for readers who want more: how the system got built, what broke along the way, and why the rules are what they are.
The one rule everything else follows
Every rate in the model is the median of sourced disclosures. When we need to know what a lead actor earned on a mid-budget film in the 2000s, we take the middle value of the real, cited deals from that era and budget range. Never the high end to make a page more exciting, never the low end to look cautious. The median, every time.
The companion rule matters just as much: outside sources supply individual data points only. A disclosed salary, a court filing, a per-episode fee. No other outlet's net worth estimate is ever an input to our model, a calibration target, or a benchmark we grade ourselves against. If we matched our numbers to theirs, we would inherit their errors and add our own.
Step one was throwing out our own shortcuts
The first version of this site had problems of its own. It multiplied a channel's lifetime views by a flat ad rate, assumed 80% of videos on some channels were sponsored, took a flat 60% haircut for taxes and expenses, and grew everything at a flat 6.5% a year. Each shortcut was defensible alone. Stacked together they produced numbers nobody could trace.
So the rebuild started with a locked rulebook, written before any code. Rates come from cited disclosures. Every line item is labeled by how we know it: disclosed by someone with the real number, measured by us in public data, or modeled from a sourced rate. Each page grades itself from A to C on how much of its figure rests on the first two.
Where the raw numbers come from
For YouTubers, the measurement layer pulls channel and video catalogs from the official YouTube API, detects sponsorships from video descriptions and YouTube's own paid-promotion disclosure flag, and rebuilds each channel's view history from Internet Archive snapshots so income lands in the year it was actually earned. Money earned in 2013 needs to ride 2013's ad rates and 2013's markets, and dating the views is what makes that possible. Measurement kept correcting our assumptions: one top science channel we had assumed was 80% sponsored measured out at 27%.
For actors, the foundation is a table of more than 80 disclosed film deals, each with a citation: the $20 million Black Widow salary Disney confirmed during the 2021 lawsuit, the Gravity deal reported at $20 million against 15% of first-dollar gross, Sylvester Stallone's own account of taking $35,000 plus points for Rocky. Court records, regulator filings, and the UK company register supply the checks. Every cited URL is archived at collection time so the evidence cannot rot.
The engine: one career, one year at a time
The calculator itself is deliberately boring. It walks a career year by year. Income enters at that year's sourced rates. Representation fees come out at the union-capped 10% for agents plus the standard 5% for attorneys. Taxes follow the places the person actually lived, era by era, using documented residence timelines. Spending follows the savings curve economists have measured for households, which rises with income. Whatever survives compounds at that calendar year's actual return of a 60/40 stock and bond portfolio, never a smoothed average.
Where reality is documented, reality overrides the curve. Johnny Depp's managers stated in court that he spent around $2 million a month, so those filings replace the savings curve for the years they cover, and his balance in the model bottoms out in the same years his real finances did. Bruce Willis's reported $90 million divorce settlement enters as a dated liability. The model runs on the same inputs every time and produces the same number every time, which means anyone can re-run it.
What happens when a salary was never published
Most film salaries were never published, and this is where estimation earns its keep. An undisclosed lead role after an actor's breakthrough gets the median of disclosed lead salaries for its era and budget band, capped at 15% of the film's budget, the largest share ever reliably reported. Supporting roles and early-career roles get 2.9% of that median, because across the eleven disclosed deals actors signed before stardom, from Rocky's $35,000 to the $2.5 million upfront on Titanic, the median deal paid 2.9% of its era's going lead rate.
Backend points work as an expected value: about 15% of the disclosed lead deals in our table included points, at a median 2.3% of box office, so undisclosed lead roles carry that expectation against their box office. Endorsements use the median of disclosed ambassador deals, a table that ranges from a $3 million watch contract revealed in a breach-of-contract lawsuit to a $50 million cosmetics deal reported by the trade press. Producer and director credits enter at union-scale floors. Nothing on a page counts zero unless a source verifies no money changed hands, and the few lanes we cannot yet estimate are listed on each page as queued, with the reason.
Rules we learned the hard way
The current rulebook is shaped by specific failures, and publishing them is part of the point.
The first full run of the actor engine printed one action star at $1.78 billion. The cause was compounding: the model let savings grow untaxed at market returns for four decades, which roughly doubles a long career's figure. The fix became two permanent rules. Investment gains are taxed at each era's published capital-gains rates before they compound, and the savings rate is capped at 51%, the highest bracket the underlying household study actually measured rather than extrapolated.
A widely copied $33 million figure for one of Angelina Jolie's films traced back, through layers of re-reporting, to a magazine's estimate of her annual earnings that year. It was never a film fee. That catch produced the original-source rule: we cite the outlet that first reported a figure, treat re-reporting as a pointer rather than evidence, and ban aggregator estimates as inputs outright. When we applied the ban retroactively, 88 rows left the salary table, and the affected films fell back to the budget-based model.
Even small bugs got promoted into rules. Two films sharing a title, like the 2000 and 2019 versions of Shaft, once shared a disclosed salary because the lookup matched on names alone. Disclosures now only attach to a film released within a year of the deal.
How we grade ourselves
A model that publishes its rates also has to publish its misses. The standing test is leave-one-out cross-validation: hide each disclosed salary from the table, rebuild the medians without it, and ask the model to predict the deal it can no longer see. The median miss is 36%, and the model predicts low 63% of the time, so the published figures lean conservative. The worst misses are instructive: an established star who deliberately took union scale for a passion project sits far below any median, and no comparables-based model will see that coming.
Career-level checks come from documents. Depp's former managers put his 2003–2016 earnings near $650 million in court records; our film-only model counts 61% of that for the same window, and his page says which estimated and queued lanes hold the remainder. UK company accounts give hard floors for the people who file them: KSI's register-filed holding company, with him listed as controlling shareholder, is why his page carries an A grade. Every number in the scorecard, including the misses, sits on the methodology page.
Check the math yourself
Every person page shows the full calculation: each income line with its basis, the fee and tax rates applied, the year-by-year balance. The methodology page publishes every rate table with citations. If you think a rate is wrong, you can swap in your own and re-run the arithmetic by hand, and if you can document a better number than one we used, send it over. Corrections with sources beat estimates every time, and the tables get better with each one.
Frequently Asked Questions
How is celebrity net worth actually calculated?
On this site, a celebrity's career is rebuilt one year at a time. Documented income enters as reported, with citations. Undisclosed work is estimated from the median of comparable disclosed deals. We then subtract representation fees at sourced rates, taxes for the places the person actually lived, and spending along the savings curve economists have measured for households, and what remains compounds at each calendar year's real market return. Every rate and every source is published on our methodology page.
How accurate are celebrity net worth estimates?
We test ours and publish the results. When we hide each disclosed salary from the model and ask it to predict that deal from the remaining data, the median miss is 36%, and the model predicts low 63% of the time, so the published figures lean conservative. Career totals documented in court records and company filings give further checks. Every page also carries a grade from A to C showing how much of its figure rests on documented numbers.
Why do celebrity net worth sites disagree so much?
Most sites publish a bare number with no math attached, and other sites copy it, so figures drift apart and harden without any shared evidence. Two sites can also make very different silent assumptions about taxes, spending, and investment returns. Disagreement is inevitable when the work is hidden. Our answer is to publish every input, rate, and citation so a reader can see exactly where any figure comes from and challenge the parts they disagree with.
What counts as a source for a net worth estimate?
Official records first: court filings, regulator documents, company accounts. Then first-party statements on the record, then trade press reporting from the outlet that originally broke the figure. Aggregator sites' own net worth estimates never enter our model as inputs, targets, or accuracy benchmarks. A figure that exists only inside an aggregator is treated as undisclosed and estimated from documented comparables instead.