Comparison · Independent
A general chatbot is the wrong tool for a client file.
General assistants are useful for plenty of legal work. Pasting a client's documents into one is where it goes wrong, and the reported sanctions cases all share a shape.
Three differences that have produced real sanctions
Every reported fabricated-citation case has the same cause.
Citations recalled, not verified
A general model produces citations from training data. They look right because looking right is what the model optimises for. HIGHCOURT checks each against the authority's actual text.
The document leaves
Pasting a client document into a hosted assistant is a disclosure to a third party, whatever the retention setting says.
No provenance
A chat answer cannot show which page produced it. Every finding here links back to the source, so checking is a click.
One document at a time
A chat window works on what is pasted into it. Whole-matter analysis is a different task at a different scale.
What stays true on every matter
These do not change by practice area, firm size, or the question asked.
Independent comparison. HIGHCOURT LAW LLC is not affiliated with, endorsed by, or sponsored by OpenAI. ChatGPT is a trademark of OpenAI and is used here only to identify the product being compared. Statements reflect publicly available information at the time of writing and may change — verify current details with the vendor.
Ask both for authority on the same point.
Then check every citation. That single exercise settles the question faster than any argument about it.
Request a BriefingQuestions firms ask
The answers below are specific to this page’s subject.
Is ChatGPT useful for lawyers at all?
Yes — for drafting non-confidential correspondence, explaining unfamiliar concepts and general writing. The problem is specifically client material and citations.
What about enterprise ChatGPT with no training on our data?
That addresses training. It does not change the fact that a third party receives and processes the material, which is the disclosure question.
Does HIGHCOURT use the same underlying technology?
It uses language models, run locally, constrained by verification against a real corpus and by mandatory human review. The technology is related; the system around it is the product.
Could we just use a local open-source model ourselves?
You could run a model locally, and some firms do. What is harder to assemble is the authority corpus, the verification, the whole-file pipeline and the review trail — which is most of the work.
Related
Other pages covering the same ground from a different angle.
See it run on a real matter.
A briefing walks the whole path — intake, workup, the Adjudicator's pass, and the work product that comes out the other side.
Request a Briefing