In October we ran a small experiment on the platform where journalists and publishers trade questions and quotes. We posted five simple questions to the expert pool — beginner mistakes, breakthrough concepts, daily habits, reliable indicators, combining tools. Fifteen experts answered. Every single answer — 15 out of 15 — was flagged by the platform's own AI-likelihood detector as machine-generated. Most at 100%.
This is not a takedown of any individual (everyone is anonymized here), and not a complaint about the platform — its detector is exactly why we know. It is a snapshot of what "expert commentary" looks like in 2026, why that matters if you read trading advice, and then — because the questions were good — our own honest answers to all five.
Strip the names and the fifteen answers share one texture. The grammar is flawless. The structure is symmetrical — three tidy paragraphs, a bolded platitude, an encouraging close. There are no numbers that couldn't belong to anyone: no account sizes, no pairs, no dates, no "the trade I got wrong was." One answer explained that position sizing "determines how much a mistake can actually cost you" — true, weightless, and impossible to disagree with. That is the tell.
Real experts write like they talk: unevenly, with scars visible. "I lost $9,000 in 2019 holding EURCHF through the SNB floor decision" is a sentence no language model volunteers, because it requires a real person to have done something regrettable. The fifteen submissions contained zero sentences of that kind.
1. No numbers with decimals or dates. Real practitioners say "about 1.5%," "last March," "0.8 lots." Generated text says "a small percentage."
2. No first-person consequences. Every lesson arrives already learned, without the loss that taught it.
3. Perfect balance. Real answers dig into one thing; generated answers give three equal points and a summary.
4. Advice that fits every question. Swap the topic and the paragraph still works — it was never about the topic.
Apply those four checks to your favorite "experts reveal" roundup. The results are uncomfortable.
The tragedy of the fifteen answers is that the questions deserve better. Here are ours — the same five, answered in our own words, with the arithmetic we actually use.
What beginner mistake taught us the most? Sizing by feel. The first account didn't die to a bad strategy; it died to 5-lot "conviction" trades on a $3,000 balance. The fix is boring and absolute: the stop goes first, the lot size follows the risk percentage, never the reverse. One rule, applied for a month, changed everything after it.
The concept that improved performance? Expectancy over win rate. A 45% win rate with winners twice the size of losers beats a 75% win rate that gives it all back on one loser. Once we started tracking (win% × avg win) − (loss% × avg loss) per trade instead of green-day counts, the strategy debates got short and the results got real.
The habit that still pays? The Friday review — 20 minutes, journal open, average loss measured against the planned loss. It is the single habit that catches every other problem early: widened stops, revenge sizing, the setup that looked good in March.
The most reliable indicator? Honesty first: no indicator is "reliable" — all of them are formulas on past price. If forced to keep one: a moving average, purely as a regime filter (trending vs ranging), never as a signal. Everything after that is context.
Combining tools? One setup definition, one confirmation tool, and a higher-timeframe permission — three roles, no overlaps. The moment two tools vote on the same decision, it is one opinion counted twice.
We didn't publish them. Instead, every identifiable expert got one message: the edit pass flags AI-assisted text — send one or two sentences in your own words, a concrete detail, and you'll be featured. The replies that come back in a human hand will appear in a follow-up piece. The ones that don't will answer the experiment's real question: how much of the expert economy is now a costume.
Expert quotes exist to save you time — someone who did the thing, shortening your learning curve. That only works if someone did the thing. Ask for the number, the date, the loss. In 2026, the follow-up question is the only detector that matters.
Terms · Privacy · Editorial Policy · Affiliate Disclosure · Risk Disclaimer · Contact