9 Studies, One Site: What Our Own Freelance Data Actually Found in 2026

Over the past few months we've published nine posts built from our own internal data files rather than surveys or scraped third-party numbers — a freelance price index, three studies off a 323-profile freelancer catalog, a DIY-vs-hire cost index, a niche pay index, a platform opportunity index, an AI-task cost-gap ranking, and a transparency piece about an illustrative Fiverr gig-gallery file. Every one of those posts links to a couple of the others in its "Related Reading" section. None of them has been read as a set.

That's what this post does. We're not re-running any numbers — every figure below is pulled directly from the nine posts as already published, with a link back to the original for anyone who wants the full methodology, the full table, or the exact caveat. What we're adding is the cross-post read: the patterns that only show up once you put a 137-row services file next to a 10-row niches file and a 7-row platforms file and notice they're all pointing the same direction on the same question.

This post does no new data pulling of its own. Every statistic below was already published in one of the nine source posts, computed there from a live internal data file as of the date that post was written. What this post adds is comparison across posts — reading a finding from the 323-profile freelancer file next to a finding from the 10-row niches file and asking whether they reinforce, complicate, or simply coexist with each other.

That means this synthesis inherits every caveat the individual posts already disclosed, and adds one of its own: mixing a finding from a 323-row dataset with a finding from a 7-row dataset in the same paragraph does not make them equally reliable. A pattern that shows up in three separate 323-profile analyses is a much stronger signal than a pattern that shows up once in a 10-niche file. We've tried to be explicit below about which kind of claim is which.

Before the cross-post patterns, here's the fast lookup: what each post is built from, its single biggest finding, and the one caveat it insists on. Use this table to jump straight to the original post for full methodology, full tables, and the FAQ.

No single post set out to prove this. It shows up because four unrelated files keep landing on the same answer when you ask, in four different ways, "where's the money and the opportunity?"

None of these four posts cite each other's specific numbers to make this point — they were built from four different files at different times, for different purposes (a buyer-side price ranking, a seller-side niche guide, an AI-cost hub, a platform-choice guide). That they all point the same direction independently is a stronger signal than any one of them making the claim alone. It's also a genuinely different conclusion from what any single post argues: the price index never claims technical work is the "least crowded"; the niche index never claims it's the biggest DIY-vs-hire gap; only reading all four together produces "the price ceiling, the least crowded niche math, the biggest AI-cost gap, and the hardest platform to get into are, again and again, the same kind of work."

Two posts, pulled from two entirely different files, ask a version of the same question — "does climbing a status ladder mean a higher price?" — and get opposite-shaped answers depending on whether the ladder is within one platform or between platforms.

Does Seller Level Mean Higher Prices? found Fiverr's own internal ladder (New Seller → Level 1 → Level 2 → Top Rated) is not monotonic on price at all — New Seller profiles in our data average $95.52, and Level 1, one rung up, averages just $30.83. The only field that climbs cleanly with Fiverr level is review count (a tenure proxy: 19 → 70 → 171 → 259), not price and not rating (flat at 4.90–4.92 across every rung).

Best Platform to Start Freelancing On found something structurally different one layer up: Toptal — a separate marketplace, not a badge within Fiverr — carries a flat $2,927.78 average starting price across all 18 Toptal profiles in our data, and it's the only platform in the 7-platform file rated "hard" to get into at all, alongside being the only one with a $5,000+ monthly earnings floor.

Put together: within a single marketplace, a level badge is a weak, non-monotonic price signal in our data — it mostly tracks how long you've been active, not what you're worth. Between marketplaces, which one accepted you is a much cleaner signal — Toptal's roughly 3% acceptance rate (noted in both the seller-level and platform posts) does more pricing work than four rungs of Fiverr's own ladder combined. If you're hiring and trying to use "level" as a shortcut for quality or price, our data suggests the marketplace someone got into tells you more than the internal badge they've climbed to once they're in it.

The 137-service price index and the 323-profile delivery-time study never reference each other's specific numbers, but they describe the exact same structural mechanism from two different angles.

The price index's spread finding: the widest low-to-high price spreads on the whole site (Mobile App Developers at 250x, several other software-dev roles tied at 200x) are all services priced by unknown project scope rather than a fixed rate. The narrowest spreads (Cloud Architects at 1.7x, several Architecture & Tech Leadership roles at 2.0x) are all hourly-billed senior specialists — you're paying for a rate, not absorbing unknown scope, so the band stays tight.

The delivery-time study's finding: median price climbs with delivery window ($10 at 1-day, $400 at 21-day) — but the two correlation measures disagree sharply (Spearman 0.648 vs. Pearson 0.227), and the reason is that slow-delivery categories (VTuber rigging, blockchain development, SaaS/mobile builds) are inherently high-effort, specialized work that also happens to take longer — not that buyers pay a premium for patience. Within a single category, the study found the opposite of a rush fee: the priciest listing in both SaaS Developers and Mobile App Developers was also the fastest.

This one isn't a finding about the freelance market — it's a finding about how Memvers' own catalog is shaped, and it shows up as a recurring methodological decision across four of the nine posts.

Four separate posts, built at different times from different files, all ran into the same shape of problem: a small cluster of high-ticket, low-count enterprise-tier listings (Toptal engagements, monthly Squad retainers) sitting in the tail of an otherwise much cheaper, much more numerous catalog. Every time, the fix was the same — report the median as the "typical" figure, show the mean too, and name the outliers rather than silently averaging them in or quietly dropping them. That's not a coincidence of four authors happening to think alike; it's what you'd expect any honest analysis of this specific catalog to run into, because the underlying shape (mostly small gigs, a handful of enterprise services) doesn't change between files.

This pattern is worth a specific honesty flag: it comes from two posts, but both pull from the same 323-profile freelancer file, so it's one dataset confirmed twice with different methods — not two independent samples agreeing.

The Rating Illusion computed coefficient of variation across three fields in the 294 rated profiles: rating varies by just 1.4% relative to its own mean, review count varies by 140.2% — a hundred times more relative movement. Does Seller Level Mean Higher Prices? asked a completely different question (does level track price?) on the same file and, as a side finding, discovered review count is the one field that climbs in a clean, monotonic staircase with Fiverr level (19 → 70 → 171 → 259), while rating stays flat (4.90–4.92) at every single rung.

Two different questions, two different statistical methods (coefficient of variation vs. a level-by-level breakdown), the same underlying file, and the same answer both times: rating barely moves no matter how you slice this data, and review count is the field that actually differentiates one profile from another. That's about as solid as a single-file finding gets on this site — it isn't a second independent dataset, but it is a finding that survived being tested two different ways without changing.

The niche pay index (10 rows) and the platform opportunity index (7 rows) are the two smallest files in this set, built at different times for different hub pages — and independently, both posts made the identical methodological call: refuse to compute a mean, median, or correlation on a sample this small, and use a plain cross-tab (demand vs. competition; barrier vs. earnings ceiling) as the primary analytical device instead.

Both posts also flag their own core fields — demandLevel/competition on the niche side, difficulty/averageEarnings on the platform side — as editorial judgment calls with no linked scoring formula, job-posting count, or survey behind them, in nearly identical language. Neither post borrowed this framing from the other; it's the same discipline arrived at twice because it's the right amount of rigor for a file this size, not because one copied the other.

That's a useful contrast with the three posts built from the 323-profile file, which do reach for Pearson/Spearman correlation coefficients and coefficient-of-variation comparisons — appropriately, because 323 rows can support that machinery in a way 7 or 10 rows can't. Matching the statistical tool to the sample size, rather than using the same toolkit regardless of n, is itself a pattern worth naming across this set.

This isn't a pattern so much as a warning against manufacturing one. Two of the nine posts investigate visually similar symptoms — a rating field that barely varies — and land on completely different, non-overlapping explanations, because they're looking at two fundamentally different kinds of files.

If you do technical/build work and have the skills for it, this catalog's own data agrees with itself across four unrelated files that this is where the price ceiling, the lighter competition, and the platform opportunity all sit together. That's a real, repeated pattern in our data — not a guarantee any specific technical gig pays well, but a consistent direction worth weighing if you're choosing where to specialize.

If you're chasing a level badge within one platform expecting it to lift your price, our seller-level study's finding (price actually dips at the very first rung above New Seller on Fiverr) suggests that instinct isn't well supported in this data — review count climbing with level tracks tenure, not pricing power. If a bigger jump matters more to you than another badge, our platform study found the clearer earnings-ceiling move in this dataset is between marketplaces (applying to a harder-vetted platform), not up a single marketplace's own ladder.

And if you're in one of the "crowded despite good demand" niches our niche index identified (graphic design, logo design, video editing, web development, virtual assistance), differentiating on portfolio process and specific, detailed reviews matters more than price competition — with every rating on this site clustering in essentially the same narrow band, the aggregate star number has very little room left to do that work for you.

Every finding above traces back to one of these nine posts. Read the original for the full methodology, the full tables, and each post's own FAQ and citation info.