High token consumption and unreliable results for professional work
I gave Claude many chances because the AI clearly has potential, but after using it extensively for professional financial writing and TradingView Pine Script development, I decided to cancel my subscription and give it 1 star.
I am a financial writer, so accuracy, logical structure, verified information, and strict adherence to editorial guidelines are essential. I provided Claude with clear outlines, detailed instructions, examples of the structure I wanted, and even reliable, verified sources. Despite this, the final content frequently failed to meet the required standard.
Claude often inserted sentences in inappropriate or completely random sections, changed parts I had not asked it to change, or ignored specific instructions. Even when I explicitly said not to add certain information or not to modify a particular section, it sometimes did exactly that. Fixing one problem could also create several new ones elsewhere in the article. As a result, I spent too much time checking, correcting, and reorganizing the output myself.
Accuracy was another major concern. Even when I supplied the correct sources and information, the final financial content was not consistently accurate or properly aligned with those sources. For professional financial writing, an answer that sounds good is not enough—it must be accurate, logically structured, and supported by the provided information.
Token consumption was also a major drawback. The usage seemed disproportionately high, even for small tasks or simple revisions. I could have accepted higher consumption if the results were consistently strong, but that was not my experience. Too many tokens were consumed while I was repeatedly trying to correct problems that should not have occurred after providing such detailed instructions.
I also tried using Claude to create a TradingView Pine Script strategy. I provided extremely detailed rules covering level identification, confirmations, entries, trade execution, risk management, fixed risk-to-reward ratios, stop-losses, and take-profit targets. Despite all this detail, the strategy still made serious mistakes. It incorrectly identified or respected levels, failed to execute some trades according to the rules, and closed positions before reaching the fixed take-profit—even though I had clearly requested a fixed risk-to-reward ratio.
The most frustrating part was not simply that errors occurred, but that they continued after repeated explanations and corrections. The written explanations could sound convincing, while the actual code still behaved differently from the strategy rules.
I do not expect any AI to be perfect, but a paid professional tool should reliably follow clear instructions, preserve the requested structure, use supplied sources correctly, and correct errors without repeatedly introducing new ones. Unfortunately, Claude did not provide that reliability for my work.
I hope future updates improve its reasoning, token efficiency, accuracy, coding reliability, and ability to respect detailed user guidelines. A direct integration with TradingView would also be valuable, particularly if it allowed the AI to test and verify how Pine Script strategies actually execute. For now, however, I am cancelling my subscription and moving to ChatGPT.