Why Analysis Takes Time — and Why That's About to Change
If you've used the Text Analyzer, you've noticed it takes a while. Not broken-slow — but not instant either. A thorough analysis of a medium-length text can take 30 to 60 seconds. For a tool built around AI, that might feel surprising. Here's what's actually happening, and why it's going to get much better very quickly.
What a "deep analysis" actually involves
When you submit a text to TellDear's analyzer, it doesn't just run a keyword search against a list of fallacy names. It calls a large language model — currently Claude Sonnet — and asks it to reason carefully about the text across hundreds of distinct analytical dimensions.
The model reads the argument structure. It identifies rhetorical moves. It considers what's implied, what's omitted, and what emotional mechanisms are being triggered. It cross-references these observations against the taxonomy. Then it structures its findings into a ranked, explained report.
This is genuinely complex reasoning. A good human analyst with expertise in logic, rhetoric, and cognitive psychology might take 20 minutes to produce an equivalent analysis. The AI does it in under a minute — but that minute is real compute.
Why it can't (yet) be instant
Current large language models generate text sequentially, token by token. Longer, more complex reasoning takes more tokens. More tokens take more time. There's no shortcut: the quality of the analysis is directly tied to how much reasoning the model does.
TellDear's analysis is also deliberately thorough. We could make it faster by asking shallower questions — but a shallow analysis that misses the key manipulation technique is worse than a slow one that catches it.
What's changing
Inference speed for large language models has improved by roughly 10x in the past two years, and that trend is accelerating. The same analysis that required a powerful server cluster in 2023 runs on consumer hardware in 2025. Models are getting faster without getting less capable — in many cases, while getting more capable.
Concretely: the analysis that currently takes 45 seconds will likely take under 5 seconds within 12 months, and under 2 seconds within 24. Not because we're cutting corners — because the underlying infrastructure is improving that rapidly.
We're also exploring parallel analysis pipelines: splitting the taxonomy into independent sections that can be analyzed simultaneously, then merging results. This could compress analysis time by another factor of 3–5x independent of raw inference speed improvements.
The honest trade-off
For now, TellDear is a tool for deliberate analysis — not for quick reactive checking of a tweet. You submit a text when you want to understand it deeply, not when you're in the middle of a heated argument. That's actually the right use case: slowing down, examining carefully, naming what you're seeing.
The speed will come. The depth is already there.