Every IP platform demos AI now, so the demo proves little. These are the questions that actually separate them before you buy. Get the checklist.
Every demo lands the same beat now. The rep opens a matter, types a prompt, and an AI drafts an office action response in nine seconds. The room nods. Somebody says "wow." The deck moves on.
It's a good trick. It's also the wrong thing to be watching.
Here's the problem with the AI demo. Nearly every platform has one now. "Do you have AI" used to separate the field. Now it separates nothing. Ask it and you'll get a yes from everyone in the room, including the vendor whose AI is a wrapper they bolted on last quarter. The demo tells you the AI can produce something. It doesn't tell you what happens to that something next.
And "next" is the whole job.
A draft response is not a filed response. A classified piece of mail is not a docketed deadline. The AI's output is the start of the work, not the end of it. The questions that actually separate one platform from another are about the part the demo skips: where the output goes, who checks it, what it updates, and what happens when it's wrong.
So before you sign anything, ask these.
What happens to the AI's output after it's generated?
This is the question that does the most work, so ask it first.
The AI drafts a response to an office action. Fine. Now what? Does it stay in a chat window you copy and paste from? Does it land in a document you then have to file by hand, re-name, and route to the right matter yourself? Or does the platform attach it to the matter, log it, and move it to the next step on its own?
Watch the rep's hands during the demo. If the impressive part is the generation and the boring part (saving it, naming it, putting it where it belongs) is still manual, you haven't bought AI. You've bought a faster way to create work that someone still has to move. That someone is your paralegal, at 6pm, doing the part the demo didn't show.
The value isn't the draft. It's the draft landing in the right place without a human shepherding it there.
Who reviews the AI's work, and where?
No serious IP team lets an AI touch a deadline unsupervised. You shouldn't, and any vendor who suggests otherwise is selling you risk.
So review isn't optional. The question is whether the platform has a real place for it. When the AI classifies incoming mail or drafts a form, does that output sit in a review queue where a docketer signs off before it's official? Can a paralegal see what the AI proposed, change it, and approve it in one place? Or does "review" mean someone notices the mistake three weeks later when a client asks why their status is wrong?
"Queued for review" is the phrase you want to hear. Not "the AI handles it." The AI proposes. A person decides. The platform should make that handoff obvious, fast, and logged.
Does the AI update your system of record, or a copy of it?
This one's quieter, and it's where a lot of platforms quietly fall down.
When the AI catches a date or classifies a document, where does that land? If it updates your docketing system, your team works from one source of truth. If it updates the platform's own copy of your data, you now have two versions of reality that drift apart the first time someone edits one and not the other. That's not automation. That's a reconciliation problem you didn't have before.
It's also a question your IT lead will ask before anyone else does. Where does the data live, who owns it, and what's the integration that keeps the two systems honest? A platform that holds its own shadow copy of your docket is a second database to secure, back up, and answer for. One that writes back to the system you already run is one less thing for IT to babysit.
Ask it plainly. When your AI does something, does my docketing system know? Or do I have to keep them in sync myself?
The right answer is that the platform works on top of the system you already run, not next to it. You keep your docketing system. No rip-and-replace. The AI does its work and the system you already trust stays the source of truth.
Can you see what the AI did?
A demo shows you the AI doing one thing, perfectly, once. Production is a thousand things a day, some of them wrong.
So you need a record. When the AI re-classified that piece of mail, attached it to a matter, and set a soft date, can you go back and see all three steps? Who can see them, the docketing manager, the supervising attorney, the partner who gets the angry client call? If something goes sideways, can you trace what the AI did and where it went off, or do you just know the outcome was wrong and have no idea why?
An audit trail isn't a nice-to-have for IP work. It's the difference between "we caught it and here's exactly what happened" and "we're not sure how that got missed." It's also the first thing your security review will ask for, along with who can see what. Role-based access isn't a feature on this kind of platform. It's the price of entry. Ask to see the log, not just the magic.
What happens when the AI is wrong?
It will be wrong sometimes. Everyone's is. The mature question isn't "is your AI accurate," because every vendor will say 99-point-something percent and none of them will define the denominator. The question is what the platform does on the bad day.
If the AI misreads a date, does a deadline still get caught by another check, or does one wrong read mean a missed response? If it mis-classifies mail, does it route to a human instead of silently filing it wrong? A docket entry that's 100% accurate and never moves is still a missed deadline. So is one the AI got wrong with nothing behind it.
And remember who answers for it when it goes wrong. When a date gets entered wrong, the client calls the firm, not the vendor. "The AI did it" has never once calmed a client whose deadline got blown, and a malpractice claim doesn't care which system made the mistake. So the platform that assumes its AI will miss, and builds the catch, isn't being cautious. It's being honest about where the exposure sits.
Ask where the catch is. If there isn't one, you've found your answer.
The demo was never the decision
Here's what years of watching these rollouts teaches you. The flashy demo is the permission slip, not the decision. It gets AI onto the shortlist. It doesn't tell you which platform your team will still be thanking you for in year three.
The platforms worth buying aren't the ones with the best nine-second draft. They're the ones where the draft goes somewhere, gets reviewed by the right person, updates the system you already trust, leaves a record, and gets caught when it's wrong. AI that produces is table stakes. AI whose output actually moves into the work is the difference.
That's the bar we built PracticeLink to clear. It works on top of the docketing system you already run, so the AI's output lands on the matter, sits in a review queue for a person to sign off, and updates the docket your team already trusts. Five of the top ten US patent filing firms run their operations on it. They kept the docketing systems they already had.
That's a harder thing to show in a demo. Which is exactly why it's the thing to ask about.
If you're evaluating platforms right now, we wrote a fuller version of this as a buyer's checklist. It's the list we'd want in front of us before sitting through one more "wow" moment. No sign-up, just the questions.