An AI research tool like Vorticel should not be used when the question requires legal judgment rather than a fact drawn from a document, when the material needed is not in your firm's own files, or when the source documents are scanned images without a text layer. In each of those cases the tool either cannot help or will honestly refuse rather than guess — which is the correct behavior, but it means you need a different resource for the task at hand.
When is an AI research tool the wrong tool for the job?
It is the wrong tool whenever the answer does not live inside a document your firm has uploaded. Grounded AI search, by design, only draws on the firm's own files and cites the document and page it pulled from — it does not browse the web, does not know current case law, and does not have an opinion about your matter. If a task is really "what does this contract say" or "which of these twenty leases has this clause," that is squarely in scope. If the task is "is this argument likely to win" or "what does the appellate court in this circuit currently hold on this issue," no document-search tool — grounded or not — is equipped to answer, because the answer isn't sitting in a file on your firm's server.
This distinction matters because the two kinds of questions can look similar on the surface. "What did our engagement letter say about scope" and "is our engagement letter enforceable" both sound like research questions, but only the first is a lookup. The second calls for a lawyer's judgment about your jurisdiction, your facts, and current law — none of which lives in the uploaded document itself.
What kinds of documents won't work with grounded AI search?
Scanned, image-only files without a text layer won't work, because the tool extracts and indexes text — it does not currently run OCR on flat images. A PDF that is really a photograph of a page, with no selectable text underneath, will upload but won't produce useful search results, because there is no text to embed or cite. The same is true for handwritten notes captured as an image, or a fax scan saved as a PDF with no OCR pass ever run on it.
Text-based PDFs, DOCX files, and plain text work well, because the extracted text is what gets split into passages and searched. If you're not sure whether a document has a real text layer, try selecting text in it with your PDF viewer — if you can highlight and copy a sentence, it has one; if the whole page behaves like one image, it doesn't.
This comes up more often than firms expect, because a lot of intake material starts life as a scan: a fax cover sheet, a signed contract photographed on a phone, a decades-old file pulled from paper storage and run through a copier. None of that is unusual, and none of it is a reason to avoid a document-search tool for everything else in the matter — it's a reason to run those specific files through an OCR step first, or to keep them out of the searchable set and handle them manually, rather than assuming the tool silently read them.
When do you need a lawyer's judgment instead of a search tool?
You need judgment, not search, whenever the output is a recommendation, a risk call, or a strategic choice rather than a citation to a document. A tool that only answers from your files is being honest when it refuses to weigh in on those — and that refusal is a feature, not a gap, because a confident-sounding guess on a strategic question is far more dangerous than an honest "that's not in your documents."
- The question asks what to do, not what a document says. "Should we settle" is judgment; "what does the settlement demand letter propose" is a lookup.
- The answer depends on law outside your firm's files. Current statutes, recent case law, or a specific court's local rules aren't in your document library unless you've uploaded them yourself.
- The stakes turn on facts nobody has written down yet. A witness's credibility or a client's real intentions can't be extracted from a PDF.
- The task is drafting original legal argument. Grounded AI can summarize and locate passages; it should not be the author of a brief's reasoning.
- The matter needs privilege or ethics analysis specific to your bar. That's a professional-responsibility judgment call, not a document-retrieval task.
How do you decide if a tool like Vorticel fits a given matter?
Run a quick fit check before you lean on any document-search tool for a matter: is the answer actually inside a document your firm controls, are those documents text-readable, and is the output you need a citation rather than a recommendation. If all three are true, a grounded tool saves real time. If any one is false, plan to use it for the parts of the matter where it does fit and rely on other resources for the rest.
| Situation | Good fit for grounded AI search | Needs something else |
|---|---|---|
| Finding a clause across many contracts | Yes — text lookup with citations | — |
| Scanned, image-only intake forms | — | Needs OCR first, or manual review |
| "What does current case law say" | — | Needs a legal research database or counsel's own knowledge |
| Comparing two contract drafts for changes | Yes — passage-level comparison | — |
| "Should we take this case" | — | Needs a partner's judgment |
A two-week pilot is a low-cost way to find out which side of that table most of your firm's work actually falls on, before you commit a whole matter type to any tool. And if the concern is really about what happens to the documents themselves once they're uploaded, that's a separate — and answerable — question; see where your uploaded documents go for the full data flow.
What should you do instead when the tool isn't the right fit?
When a task falls outside grounded search, the honest answer is to use the resource built for that task: a legal research database for case law outside your files, OCR software before uploading scanned images, and a lawyer's own judgment for strategy and risk calls. None of that is a knock on document-search tools — it's just an accurate description of what a tool that only reads your own files can and can't do. Vorticel is built to be honest about that boundary: when the answer isn't in your documents, it says so rather than filling the gap with a guess, and you can see exactly how that data handling works on the security page or check plan details on pricing.
If your firm's work is mostly the left column of that table — contract review, discovery, lease abstracts, comparing drafts — a grounded tool is worth testing on real files. Start a free trial and run it against a matter you already know well, so you can judge the fit yourself instead of taking anyone's word for it.