The Dead Internet Report · No. 4 · the gig-work contraction

The Fiverr collapse.

Fiverr built a marketplace for the internet's small jobs — a logo, a blog post, a voiceover, five dollars and up. The pitch was that any task could be handed to a human somewhere cheaper. Then a chatbot started doing those same tasks for free, in seconds. The stock didn't crash right away. It waited two years — then fell 71%.

$30.95$8.93 share price · Oct 2022, the month before ChatGPT → Jul 2026 · Yahoo Finance, monthly close

85 months of price history · Jul 2019 → Jul 2026 · monthly closing price · public market data · every figure recomputable

−71%
since the month before ChatGPT
$30.95 (Oct 2022) → $8.93
2 years
the fuse — flat, then the cliff
~$31 held through Dec 2024
−73%
public search interest in Fiverr
2022 → 2025 avg · Wikimedia pageviews
−84%
Upwork, the other gig giant
fell as far before a partial bounce
Isn't this just the 2021 bubble deflating? That's exactly why we throw the peak away. Fiverr hit $270 in the 2021 remote-work mania; by October 2022 — the month before ChatGPT — the bubble was already gone and the stock sat at $30.95. Everything we measure starts there. From that sober post-bubble base, with no mania left to unwind, it still lost 71%. And the shape is the tell: it didn't fall when ChatGPT launched — it held for two years, then broke as the models got good enough to do the work.
Finding 01

A marketplace for tasks a machine now does for free.

Fiverr's promise was arbitrage: any small job — copywriting, translation, a quick logo, a voiceover — handed to a human somewhere who'd do it cheaper. It rode remote work to a $270 peak in early 2021, then gave the whole bubble back. The interesting part isn't the mania. It's what the stock did from its sober 2022 base, once a language model could do those same small jobs for nothing. Every point below is a monthly close.

{"start":"2022-10","unit":"$","values":[30.95,35.17,29.14,37.09,39.61,34.92,36.52,26.19,26.01,30.14,28.15,24.47,21.17,26.09,27.22,26.92,23.6,21.07,20.51,25.23,23.43,26.11,24.87,25.87,29.26,32.61,31.73,30.87,26.89,23.68,25.45,32.42,29.33,22.08,23.55,24.41,22.64,20.93,19.76,16.75,10.83,10.02,11.64,10.99,10.32,8.93],"markers":[{"month":"2022-11","label":"ChatGPT"},{"month":"2024-12","label":"the fuse ends"}],"endLabel":"$8.93"}
Finding 02

The two-year fuse.

The obvious story would be a crash the week ChatGPT launched. That is not what happened. Through all of 2023 and 2024, Fiverr's stock held around $30 — the market wasn't yet convinced a chatbot could really replace a freelancer. Then it broke. As 2024's models learned to write, translate, and design at usable quality — and 2025's agents learned to chain those tasks together — the stock halved, and halved again.

DateCloseWhat happened
Feb 2021$269.97Pandemic peak — we ignore it
Oct 2022$30.95Month before ChatGPT — our baseline
Dec 2024$31.73Two years later — essentially flat
Jul 2026$8.93Down 71% from baseline — the fuse burns out

AI didn't kill the gig economy on launch day. It took two years — the time the models needed to get good.

Finding 03

Gig work was a queue of prompts waiting to happen.

A Fiverr order is, functionally, a prompt: a short instruction — "write me 500 words on X," "translate this," "make a simple logo" — fulfilled by whoever is cheapest. That is the exact interface of a language model, minus the wait, the back-and-forth, and the fee. The high-volume, low-complexity tier of freelancing — the entry-level copy, the basic translation, the template design — is precisely the work models do first and best. Fiverr didn't lose the hard jobs. It lost the easy ones, which were most of the volume.

What it means for you

If you sell hours for simple tasks, price in the machine.

Fiverr is a warning to anyone whose living is a queue of small, well-specified tasks. Abundance came for the bottom of the ladder first. Three lessons.

1 · Move up the complexity curve. The commodity tier — generic copy, basic translation, template work — goes first. Judgment, taste, accountability, and context are what survive.

2 · Sell outcomes, not tasks. A task is a prompt, and a model does prompts. An outcome — a campaign that converts, a brand a human trusts — is a relationship, and relationships don't disintermediate.

3 · Read the two-year fuse. The market didn't react on launch day; it reacted when the capability matured. Don't confuse "AI hasn't hurt us yet" with "AI won't" — the lag is the trap.

The data

Every figure, one table.

MetricValue
Bubble peak (Feb 2021, excluded)$269.97
Post-bubble baseline (Oct 2022)$30.95
Two years later (Dec 2024)$31.73
Latest monthly close (Jul 2026)$8.93
Change since baseline (Oct 2022 → Jul 2026)−71.1%
Change during the flat window (Oct 2022 → Dec 2024)+2.5%
Search-interest change (2022 → 2025 avg)−73%
Upwork (UPWK), peak-to-trough−84%
Months of history measured85 (Jul 2019 – Jul 2026)
SourceYahoo Finance monthly close · Wikimedia pageviews
FAQ

Questions, answered.

Wasn't this just the 2021 bubble? We assume so and discard the peak. The 71% we report is measured from the sober post-bubble base of October 2022, with no mania left to deflate. A bubble can't explain a stock that was flat for two years and then fell.

Why did it fall in 2025–26, not 2023? Because that is when the capability arrived. Late-2024 models crossed the quality bar for everyday writing, translation, and design; 2025's agents chained those steps together. The market re-priced the labor when the substitute became real, not when it was announced.

Didn't Fiverr add AI tools? Yes — like Shutterstock, it shipped AI features and an AI-services category. It didn't stop the slide, because the threat isn't a missing feature; it's that the core unit of work became free.

Can I reproduce these numbers? Yes — monthly closing price for FVRR and UPWK from public market data, and Wikimedia's pageviews API for search interest.

Are you selling tasks — or outcomes AI can't fake?

Fiverr's commodity tier evaporated the moment a model could do it for free. The work that survives is the work AI answers cite and buyers still trust a human to own. Run a free GEO audit and see where you stand.

The defensible work is what AI can't fake at scale: researched, cited, audited — exactly how HarperFlow produces content.

Run the free audit