One article in this set publishes more than sixty income figures and tells you it draws on "real-world data from successful AI freelancers and platform analytics." It names no survey, no platform, and no dataset. There is not a single link.
That is not the worst article here. Eight of the fifteen cite nothing whatsoever for any income figure they print.
What I actually did
I took the page-one results for four variations of "make money with AI", deduped them to 20 distinct articles, and read each one with the same question: list every income figure on the page, and say whether a source is named at the point of the claim.
"At the point of the claim" is the whole test. A page that cites Statista for market size and then invents its own earnings ranges has not sourced its earnings ranges. A source counted only if it sat in the same sentence or paragraph as the number: a hyperlink, a named survey, a named report, or named company data.
Of the 20 URLs, I could read 15. Two are dead. Three blocked automated reading, and a 403 describes a bot policy rather than a dead page, so those three exist and I cannot audit them. I did not swap in replacements to reach a round twenty.
The count
| Article | Income figures | Sourced at the claim |
|---|---|---|
| dreamhost.com | 20 | 18 |
| acorns.com | 31 | 3 |
| mercor.com | 6 | 4 |
| fast.io | 50+ | 1 |
| naoma.ai | 30+ | 1 |
| localaimaster.com | 60+ | 0 |
| theinterviewguys.com | 22 | 0 |
| financialbinder.com | 18 | 0 |
| nealschaffer.com | 16 | 0 |
| danmartell.com | 7 | 0 |
| murf.ai | 4 | 0 |
| raenest.com | 4 | 0 |
| wscubetech.com | 1 | 0 |
| ironhack.com | 0 | n/a |
| monexa.it.com | 0 | n/a |
More than 250 income figures across 13 articles. Twenty-seven carry a source. Eighteen of those 27 are in one article.
Take that article out and the picture is this: twelve articles publish roughly 250 income figures between them and cite independent data for five of them.
Where I have written 50+ or 60+, the page restates the same figures across a table, a body section, and a summary box often enough that an exact count would be false precision. Everything at 31 and below is an exact count.
Eight articles, 150 numbers, nothing behind any of them
The eight zero-citation articles are not fringe pages. They rank. Between them they print more than 150 income figures with no survey, report, platform dataset, or link attached to any of them.
The pattern is consistent enough to be a format rather than a lapse. A method gets a heading, the heading gets a dollar range, and the range arrives with the confidence of a measurement and the provenance of a guess.
The clearest example is localaimaster.com, which prints more than sixty figures and includes the line that 85 percent of students earn their first 100 dollars within seven days. That is a precise claim about a specific cohort with a specific deadline, and no cohort, no sample size, and no measurement window appears anywhere on the page.
theinterviewguys.com is the near miss. It names nine external sources and cites not one of them for an income figure. The sourcing apparatus is there. It is pointed at everything except the numbers readers came for.
The best-sourced article is a web host selling hosting
DreamHost cites something for 18 of its 20 income figures. It was the only article in the set that did.
Its sources are not strong. They include ZipRecruiter job postings, a Udemy instructor forum post, PayScale data from 2022, and a 2021 Business Insider survey of twelve writers. Twelve writers in 2021 is thin evidence about AI-assisted income in 2026, and ZipRecruiter advertises what employers offer rather than what anyone takes home.
The point is that I can say all of that. A weak named source can be checked and argued with. An unnamed one cannot be checked at all, and nine articles here give you nothing to check.
The inverse correlation is the finding I did not expect. The articles printing the most income figures cite the fewest sources. One figure and no source is a small claim poorly supported. Sixty figures and no source is a different kind of document.
Not one article cites a rate survey
Zero of fifteen.
Published rate benchmarks exist. The Editorial Freelancers Association's 2026 chart is built from more than 1,100 members reporting their 2025 rates. HeyOrca's 2026 social media pricing benchmark surveyed almost 100 agencies and freelancers. Neither is perfect and neither is specific to AI work. Both are real surveys with stated sample sizes, freely readable, and published this year.
None of the fifteen articles cites either. The sources that do appear point to job postings, another blog, a forum post, and a company's own job listings.
That is the actual state of this niche. It is not that the data is hard to find. It is that a category built on income promises has not gone looking.
The one statistic everyone cites is two years old
Three articles cite Upwork for the AI hourly premium. They give three different numbers.
| Article | Figure cited to Upwork |
|---|---|
| fast.io | 44 percent more per hour |
| naoma.ai | 40 percent more per hour |
| ironhack.com | No number given |
Upwork's current figure is about 34 percent. The 44 percent comes from Upwork's Q4 2024 earnings call. Axios described the 2025 study as over 40 percent.
So the single most-repeated statistic in this niche is circulating in three versions, the highest of which is roughly two years stale, and the gap between the loudest version and the current one is ten percentage points. Nobody rechecked it. I only caught it because I had cited the 34 percent figure in one of my own posts four days earlier and recognised that 44 did not match.
The most honest article has no numbers in it
Ironhack publishes no income figure at all. It says AI-skilled freelancers earn more per hour, names Upwork, and declines to put a range on it.
By doing less it became the only article in the set that does not overstate what is known. Monexa also prints no figures, though it reads more like it ran out of content than like it made a decision.
Ironhack sells a bootcamp. It had every commercial reason to print a number and did not.
Thirteen of fifteen are selling something
| What they sell | Count |
|---|---|
| Own product or platform | 7 |
| Own course or bootcamp | 4 |
| Affiliate links | 3 |
| Nothing visible | 2 |
This is not a scandal and this site runs affiliate links too, which is why the disclosure page exists. It matters only in combination with the first table. An unsourced income figure attached to a product that promises to deliver that income is doing a different job than an unsourced income figure in a neutral listicle.
My own posts had the same defect
Two of mine carried monthly ranges under headings that read "Realistic income". Nothing about those figures was realistic in the sense of measured. They were my read of the market printed in the font of a finding.
I fixed both on 29 October 2026, before writing this. The headings now read "Reported range (estimate)". Where published data exists the figures are anchored to it, and where it does not the posts say so outright. Both carry dated correction notes.
One correction went up rather than down. My social media management figure of 300 to 800 dollars a month was low. HeyOrca's survey found only 31 percent of agencies and freelancers charging under 1,000 dollars a month, so I had been pitching my own estimate at the bottom of a market I had not checked. The error in this category runs in both directions, which is what you would expect from guesses rather than from bias.
My posts are not in the audited fifteen. Auditing yourself inside your own sample is not a method.
The thirty second check
When you hit an income figure anywhere, including here:
Look in the same paragraph for a name. Not "studies show" or "data suggests" or "top freelancers report". A survey, a report, or a company with a dataset. Eleven of these fifteen articles fail at this step.
Check the date on the source. The 44 percent figure above was cited with no year attached in an article published this year. Two years is a long time in this market.
Ask what the number measures. Job postings are not earnings. Listing prices are not take-home. Project rates are not monthly income. Several of the sourced figures in this audit are real numbers measuring something other than what the sentence claims.
Check who benefits. If the figure is the promise of the product on the same page, weight it accordingly.
Three of those four take seconds. The fourth takes a glance at the top of the page.
Sourcing
Twenty URLs from page-one results for four variations of "make money with AI", read on 29 October 2026. Every page was read with an identical extraction question so the counts are comparable: list every income or earnings figure stated, and for each one say whether the page cites a source at the point of the claim.
Read and counted (15): dreamhost.com, acorns.com, mercor.com, fast.io, naoma.ai, localaimaster.com, theinterviewguys.com, financialbinder.com, nealschaffer.com, danmartell.com, murf.ai, raenest.com, wscubetech.com, ironhack.com, monexa.it.com.
Could not be read (5): upwork.com and vocal.media returned 403. aibusiness.vc returned 429 on three attempts. godberrystudios.com and a beehiiv newsletter page both returned 404 and appear to be gone. The three that returned 403 and 429 are live pages behind bot walls, not dead ones, and they are excluded from every count here rather than guessed at.
Benchmarks referenced: the Editorial Freelancers Association 2026 rate chart, from more than 1,100 members reporting 2025 rates. HeyOrca's 2026 social media pricing benchmark, from almost 100 agencies and freelancers, published 20 October 2026. Upwork's 34 percent AI hourly premium, with the 44 percent figure traced to its Q4 2024 earnings call. Upwork's own pages return 403 to automated readers, so its release text was verified through the GlobeNewswire distribution instead.
On the counts. Figures marked 50+, 60+ and 30+ are floors rather than totals, because those pages repeat the same figures across tables and body text often enough that an exact number would be invented precision. Counts of 31 and below are exact. Where a page restated one figure in a table and again in prose, I counted it once.
Related reading: 7 real ways to make money with AI tools and using ChatGPT to make money online, both of which were corrected on 29 October 2026 for exactly the defect audited above.
Nine articles on page one of a query about income tell you what you can earn and give you no way to check it. One tells you nothing and is the only one that does not overstate. The gap between those two is a named source, and this week it cost me two corrections to notice that I had been on the wrong side of it.
