A research tool, not an advisor

"What is the security deposit deadline in Texas?" is a research question and Perplexity will answer it well, with a citation. "My landlord sent an itemised list on day 28 claiming cleaning costs - what do I do?" is not a research question. The rule is one input; the answer depends on the facts, the timing and what you want to happen next.

Most people arrive with the second kind and search with the first, which is why the answer often feels unsatisfying.

There is a clean division of labour here. Use Perplexity to establish what the rule is and where it is written. Use a specialist to work out what it means for your circumstances and to produce whatever you have to send.

Perplexity's citations are its best feature and they are worth understanding precisely, because they establish something narrower than people assume.

A citation tells you where a claim came from. It does not tell you that the source is current, that it applies in your jurisdiction, or that it addresses your circumstances. A well-cited answer about security deposit deadlines might be citing a 2019 article about a different state, and the citation makes it feel more reliable rather than less.

That is not a criticism of the tool - it is doing exactly what it says. It is a caution about how citations change reader behaviour. Sourced answers get checked less often, which is precisely backwards when the source is stale or geographically wrong.

The categories where this bites hardest are the ones with fast-moving thresholds: tax figures that change annually, benefit limits, filing deadlines, and anything altered by recent legislation. For those, the primary source is the only reliable one, which is why every finance and legal page here links to the IRS or the relevant statute rather than to a secondary summary.

Used properly the combination is strong: Perplexity to find where the rule is written, then the primary source to confirm it is current, then a specialist to work out what it means for you.

Almost every real problem has a research half and an action half, and they need different tools.

The research half is answerable from published material: what the deadline is, what the threshold is, what the form is called. Perplexity is very good at this and often better than asking a specialist, because the answer is a fact rather than a judgement.

The action half is not. What do I write, what do I send, what do I say first, what happens if they refuse. Those depend on your specific facts, on what you are trying to achieve, and on the sequence - and none of it is published anywhere because it is about you.

The failure mode is arriving with the second kind of question and searching with the first. You get a correct, well-cited answer about the general rule and are no closer to knowing what to do on Monday, which reads as the tool having failed when it has actually answered the question you asked.

Recognising which half you are in is most of the skill. If you can phrase it as "what is", research it. If it is "what should I do given", that is a different tool - and usually a different kind of answer, ending in a draft rather than a finding.

The thing that most often converts a resolved question into an unresolved one is that nobody wrote the letter.

People find out what their rights are, understand the deadline, know roughly what to say - and then the blank page sits there for three weeks until the window closes. That is not a knowledge problem and no amount of research fixes it.

A draft changes the economics of acting. Editing something wrong is a fifteen-minute job; producing the first version from nothing is the part that gets postponed indefinitely. That applies to a demand letter, a dispute of a medical bill, a response to an IRS notice, a resignation, or the message to a landlord.

Research tools do not produce documents, and that is a reasonable scope decision rather than an oversight. But it means the last and most consequential step is left entirely to you, at exactly the point where the emotional cost of the situation is highest.

If the outcome you need is something sent rather than something understood, that is the difference worth paying attention to. Every domain page shows what the copilot actually hands you rather than describing it.

Worth making concrete, because the research-versus-application distinction sounds like hair-splitting until you see it on one question.

"What is a CP504 notice?" is research, and a good answer comes back with a citation to the IRS: it is a notice of intent to levy, it carries a response window, here is the page it is defined on. Correct and useful.

"I got a CP504 three weeks ago and I cannot pay the full amount" is not research. The answer depends on the date on the notice, the balance, whether you have filed everything, and what you want to happen - and it ends in a decision about which option to take rather than a definition.

A specialist establishes those first, then tells you the specific date your window closes, whether an instalment agreement is close to automatic at your balance, why an offer in compromise is far more restrictive than the advertising suggests, and what to send. The most valuable output is usually the deadline, because it is the only genuinely unrecoverable element and it is stated once, mid-paragraph, in a document written for internal consistency.

Both tools are answering honestly. They are answering different questions, and recognising which one you are actually asking is most of the skill. The finance copilot page walks a full example end to end.

The failure mode with research tools is different from the one with general assistants, and it is easier to miss because the answer looks complete.

You asked what the rule is, you got the rule, and you got a citation. Nothing is wrong. But if what you actually needed was to know whether the deadline on your notice has already passed, or which of two options applies at your balance, the well-sourced answer has moved you no closer - and because it looks authoritative, it is less likely to prompt a second question.

That is the specific risk: a correct, cited, general answer is more satisfying than a hedged one and equally unactionable. Satisfaction without action is how a question gets closed prematurely.

The tell is whether you finished with something to do. If the output is a finding, you were doing research and it worked. If you are still looking at the same envelope wondering what to send, the question was never a research question.

Both tools are honest here. The skill is noticing which half of the problem you are actually in, and switching rather than searching harder.