Intelligence Price Index

What this is

We use the Internet Archive’s Wayback Machine to recover historical Fiverr prices and track how the cost of freelance “intelligence” work — design, writing, coding, video, audio — has moved since 2020. The Intelligence Price Index (IPI) is a CPI-style index of that work. The goal: separate the upward pull of inflation from the downward push of AI productivity gains — which is why the chart shows the index deflated by CPI-U by default.

How to read it

The IPI is in index points, not dollars: 2020Q1 = 100, every quarter scaled to it. 141 pts = 41% more expensive than 2020Q1; 80 pts = 20% less. Read levels in points and changes in percent (quarter-over-quarter, year-over-year, or vs. the 2020Q1 base). In the default Real view, “more expensive” means after stripping out general inflation.

Intelligence Price Index (GEKS-Jevons)

Index level in index points (base 2020Q1 = 100) · full history

Every quarter is compared with every other quarter directly, rather than through a chain of quarter-to-quarter links — so the index is free of chain drift. Change over the window: .
Real, not just nominal. Freelance prices are quoted in dollars, and the dollar itself lost value over this window — US consumer prices rose 26.8% between 2020Q1 and 2026Q1. The default Real view divides the index by CPI-U (US city average, all items, seasonally adjusted) so it shows what a gig costs in constant 2020Q1 dollars. On the composite, general inflation accounts for roughly half of the nominal rise: +78.4% nominal becomes +40.7% real. Switch to Nominal to see the undeflated series with CPI-U drawn alongside it for comparison.
Shaded bands — 95% confidence intervals, on every line. From a bootstrap over gigs. We hold this pilot to a stated precision standard — a category should be within ±5% at 95% confidence in the latest quarter — and six of the seven categories do not meet it: translation ±29.2%, coding ±17.1%, audio ±13.9%, video ±11.9%, writing ±8.3%, marketing ±7.7%; only design (±4.8%) clears it. The ±95% column reports each one and shades the misses. The composite does meet the standard (±3.7%) — it is review-weighted and design carries about 71% of that weight, so the basket is far better measured than most of its parts. Read the shaded band, not the line — and treat the category ordering as provisional, since the top of the table is well within its own error bars.
Validated data and method. The underlying prices were independently corroborated against their archived source pages (100% agreement on a random sample of the displayed 2020Q1–present window). The index is estimated with GEKS-Jevons, the standard drift-free multilateral estimator when only posted prices (and no quantities) are observed; our implementation reproduces the PriceIndexCalc reference exactly.
Why not a simple chained index? Gigs are not archived every quarter, so a chained index credits a gig's whole multi-quarter price change to the one quarter it reappears — on top of the rise already chained in from gigs that were observed in between, counting the same increase more than once. On this panel that inflates the composite to 317.7 pts against 144.7 here. Step 5 of the FAQ works through it.

Highlighted = sharpest quarter-over-quarter moves in the composite: price rise price drop

Volume, without a price index

Two measures that involve no price at all · index, 2020 = 100

Why this card exists. Fiverr reports dollars (GMV) and never an order count, so the count in the next card is manufactured by dividing dollars by a price index. Nothing on this card passes through a price. Review accrual is counted off the archived pages; active buyers is a headcount Fiverr publishes.
Review accrual is how fast listings gain reviews, measured within each listing so that gigs arriving and leaving cannot move it, and equal-weighted across the seven categories. It peaks in 2020Q3, steps down in 2021Q3, steps again in 2023Q4, and ends at roughly a third of its 2020 level.
Active buyers is people, not purchases — 4.20M at the 2021 peak against 2.70M in the twelve months to 2026Q2. It is drawn annually because that is how Fiverr reports it.
Neither one is an order count. Accrual is per surviving listing and the archive cannot measure exit, so it will not scale to a platform total. Buyers counts people, and spend per buyer nearly doubled over the window. A real count needs the dated order records still sitting unextracted in the archived pages.

Pooled review-accrual index and Fiverr active buyers, both on a 2020 equals 100 base
Quarter	Review accrual	Year	Active buyers (M)	Buyers index

Transactions — the implied order count

Index, 2020 = 100 · annual · orders = real GMV ÷ real IPI price

The archive contains no transactions. Fiverr Inc. (NYSE: FVRR) reports GMV — every order added up at what the buyer paid — but never an order count. GMV is dollars; splitting it into a count needs a price, and the IPI is a price. So the count below is this project's index doing a job no published figure does, over the IPI's own time range.
The fall in orders is not a fall in dollars. Real GMV is still above its 2020 level. Every bit of the divergence is price: real GMV +11.3% against a real price +35.8%, which leaves implied orders −18.0% against 2020 and −38.6% from the 2021 peak. Behind it is a composition shift Fiverr reports directly — active buyers 4.20M → 2.70M while spend per buyer rose every single year, $119 (2017) to $368. Fewer buyers, each much larger.
Read the count as an upper bound on the decline, not an estimate of it. The divisor is the listed basic package, and buyers almost never buy it: of the order amounts recovered from the archived pages, 99% are above $50 against a listed median of $25–30. If realised price rose faster than the listed index — which Fiverr's upmarket push implies — orders fell by less than this.
The turn predates generative AI. Implied orders peak in 2021, and the archive's own within-gig transaction proxy turns in 2020Q4 — both before ChatGPT, marked on the chart. Buyer growth hit exactly +0.0% in 2022 after +44.7% and +23.5% in 2020–21, an era that fits ChatGPT and post-pandemic normalisation equally well. Nothing here identifies a cause.

Fiverr transaction quantities, 2020 to the twelve months ending 2026Q2
Year	Buyers (M)	Spend / buyer	GMV ($M)	Real GMV	Real price	Implied orders

Transactions by category

Within-gig review accrual, age-adjusted · index, 2020 four-quarter mean = 100 · one panel per category

Fiverr publishes no category split of GMV or buyers, so the platform order count above cannot be broken apart. The only category-level quantity in this project is review accrual per gig — how fast a listing gains reviews — measured within each listing so that gigs entering and leaving the panel cannot move it. Direction is corroborated by Fiverr's own books; magnitude is not.
Six of the seven peak in the same quarter, 2020Q3, and the fall is a step between 2021Q2 and 2021Q3 that happens in all seven at once — roughly 15 to 20 index points each, fifteen months before ChatGPT. Design's 2019Q3 peak sits in the shaded thin-panel stretch and should be read as noise.
ChatGPT lands inside a plateau. From 2022Q1 to 2023Q3 every category sits between 58 and 84, and five of the seven are higher in 2023Q3 than in 2022Q3 — design rises 67.2 to 83.8 across exactly the window an AI story would have it falling. The next real step down is 2023Q4, a year after the launch.
The end-state contradicts AI exposure. On the pre-registered Eloundou ranking the two most exposed categories finish at opposite ends: writing worst at 23.7, translation second-best at 48.5. Least exposed of all seven is audio, at 43.9. An exposure story would have to produce both ends from one mechanism.
What is and is not identified. Within a gig, age and calendar quarter move one-for-one, so the shape of each panel is identified and its trend is not — part of every downward slope is listings getting older. The peak quarter is the statistic that survives. Separately, the archive captures gigs less often late in the window (mean span 1.2 → 1.75 quarters), so some of the 2024 fall is the crawl.

Select categories · expand (▸) for top freelancers All None

		Category	Trend	Δ'20–'26	±95%	Weight	Gigs