Where AI prediction users come from — and where they pay
Across 17,974 registered users of a live AI prediction product, signup volume barely predicts revenue. The second-largest source of signups converts at 0.24%. The best-converting substantial market — Brazil, at 4.55% — has no translated version of the site. And a fully translated French site produced zero paying customers from 149 signups.
By Zinedine · Founder & Developer · Published 26 August 2026 · Data measured 2026-08-26
Method
MiroFish.us is a hosted AI prediction engine. Users sign up, and each prediction consumes a credit from a paid plan; there is no free tier, so “paying customer” means someone who chose to spend money after seeing what the product does. The observation window runs from the first recorded signup on 2026-03-30 to 2026-08-26.
Country comes from the IP address recorded at signup, truncated to its /24 network before geolocation — accurate to country, but never using or publishing a precise address. Of 126 paying customers, 126 resolved to a country; those counts are exact. Per-country signup volume is projected from a sample of 3,500 of the 12,162 signups carrying an IP, so signup figures and the conversion rates derived from them carry sampling error.
Known limitations: VPN and proxy users are attributed to their exit country. The product is marketed in English, Chinese, Spanish and French, which shapes who arrives in the first place — these are conversion rates for this product’s traffic mix, not a universal measure of national demand. Countries below 100 estimated signups are marked as unreliable, because one customer moves the rate by whole percentage points.
Every country with a paying customer
| Country | Payers | Est. signups | Conversion |
|---|---|---|---|
| United States | 48 | 2,356 | 2.04% |
| Brazil | 15 | 330 | 4.55% |
| Germany | 7 | 466 | 1.50% |
| Mexico | 7 | 368 | 1.90% |
| India | 5 | 2,061 | 0.24% |
| Spain | 5 | 431 | 1.16% |
| Canada | 3 | 222 | 1.35% |
| Austria | 2 | 42 | 4.76%* |
| Saudi Arabia | 2 | 49 | 4.08%* |
| Türkiye | 2 | 52 | 3.85%* |
| Italy | 2 | 219 | 0.91% |
| United Kingdom | 2 | 146 | 1.37% |
| Colombia | 1 | 233 | 0.43% |
| United Arab Emirates | 1 | 136 | 0.74% |
| China | 1 | 146 | 0.68% |
| Malaysia | 1 | 132 | 0.76% |
| Japan | 1 | 101 | 0.99% |
| Singapore | 1 | 97 | 1.03%* |
| Lithuania | 1 | 97 | 1.03%* |
| South Africa | 1 | 63 | 1.59%* |
| Hong Kong | 1 | 52 | 1.92%* |
| Poland | 1 | 52 | 1.92%* |
| Vietnam | 1 | 52 | 1.92%* |
| Finland | 1 | 35 | 2.86%* |
| Romania | 1 | 31 | 3.23%* |
| Russia | 1 | 28 | 3.57%* |
| Estonia | 1 | 21 | 4.76%* |
| Norway | 1 | 21 | 4.76%* |
| New Zealand | 1 | 21 | 4.76%* |
| Afghanistan | 1 | 21 | 4.76%* |
| Uzbekistan | 1 | 21 | 4.76%* |
| Kazakhstan | 1 | 17 | 5.88%* |
| Belize | 1 | 17 | 5.88%* |
| Armenia | 1 | 10 | 10.00%* |
| Panama | 1 | 10 | 10.00%* |
| Greece | 1 | 10 | 10.00%* |
| Luxembourg | 1 | 7 | 14.29%* |
| Cambodia | 1 | 3 | 33.33%* |
* Fewer than 100 estimated signups — the rate is not statistically meaningful and should not be used to make a decision. Cambodia’s 33.33% is one customer, not a market.
The countries that send traffic and no revenue
These nine countries each produced at least 100 signups and not one paying customer between them — 2,279 signups in total.
| Country | Est. signups | Payers |
|---|---|---|
| Indonesia | 1,056 | 0 |
| Morocco | 236 | 0 |
| Australia | 198 | 0 |
| Argentina | 156 | 0 |
| France | 149 | 0 |
| Pakistan | 139 | 0 |
| Chile | 129 | 0 |
| Netherlands | 115 | 0 |
| Bangladesh | 101 | 0 |
What we take from this
Volume is a vanity metric here. India and Indonesia together produced roughly a quarter of all signups and under 4% of paying customers. Any growth strategy optimising for registrations would have pointed almost all of its budget at the two markets least likely to pay.
Translation is not the lever we assumed it was. We built a full French site. It has produced 149 signups and no customers. Brazil, with nothing translated, is our second-best market by customer count and our best by rate. Purchase intent appears to travel across a language barrier better than a translated marketing page creates it.
Small-country rates are traps. Several countries in the table show 5–33% conversion. Every one is a single customer against a handful of signups. Publishing those as headline figures would be flattering and meaningless, which is why they are marked.
What we changed because of it. We stopped advertising globally and now target five countries. If the Brazil result holds through a second cohort, a Portuguese version is the next thing we build — the demand is already there, untranslated.
Questions
Which country has the highest conversion rate for AI prediction tools?
Among markets with enough volume to be meaningful, Brazil converts best in this dataset: 15 paying customers from an estimated 330 signups, or 4.55%. That is more than twice the United States rate of 2.04%, despite the product having no Portuguese-language version.
Does signup volume predict revenue?
No — the relationship is close to inverted at the extremes. India produced 2,061 estimated signups, the second-largest of any country, and 5 paying customers: 0.24%. Indonesia produced 1,056 signups and zero paying customers. Together they account for roughly a quarter of all signups and under 4% of revenue.
Does translating a site increase conversion?
Not on this evidence. French visitors produced 149 signups and zero paying customers despite a fully translated French site, while Brazil converts at 4.55% with no Portuguese version at all. In this dataset, translation tracks conversion far more weakly than purchase intent does.
How was country determined?
From the IP address recorded at signup, truncated to its /24 network before geolocation. Truncation does not affect country accuracy but means no precise address is used or published. VPN and proxy traffic will be attributed to the exit country, which is a known limitation of any IP-based method.
Are these numbers exact?
Paying-customer counts are exact — 126 of 126 payers resolved to a country. Signup counts are projected from a random sample of 3,500 of 12,162 signups carrying an IP, so per-country signup figures carry sampling error and conversion rates derived from them are estimates.
Reuse: these figures may be quoted or charted with attribution to MiroFish and a link to this page. If you need a cut of the data that is not published here, ask. Corrections will be noted on the page rather than silently applied.
This data comes from running the product. See what it produces — a full example report, before you pay anything.
View the example report