Every IP team is hearing the same directive: use AI. Almost none of them are being told what for.
Draft a cleaner client email, summarize an application at intake, and you've saved a few minutes. You haven't changed a single prosecution outcome. Outcomes move when AI stops summarizing and starts interrogating the record: the examiner, the art unit, the rejection, and the prosecution history behind all three.
That only works when the model can see the record, and general LLMs can't. They're trained on the open web, not on 20+ years of USPTO prosecution history. Juristat Data Layer (MCP) puts that record inside the AI your team already uses, so the answer is grounded in what actually happened instead of what a model guessed.
Five prompts that put it to work.
A generic response strategy treats every 103 the same. Examiners don't. The same rejection behaves differently depending on who's across the table, the art unit, the claim type, and how prosecution has gone so far. Before you draft, make the model account for all of it, then tie that behavior to your next move: argue, amend, interview, appeal, or prepare the client for another round.
For application [application number], analyze the current non-final office action from Examiner [name or ID].
Focus on the pending 102 and 103 rejections, the cited references, the current claim set, and the prosecution history to date.
Using examiner-level and art-unit-level data, tell me:
End with a recommended strategy and explain what data supports it.
This does more than ask "what should we do?" It gives the model enough context to compare real options, which matters when the wrong move means another office action, an unnecessary RCE, or claim scope you didn't need to give up.
Some interviews change the trajectory of a case. Others just add a line to the bill. The difference is usually the examiner, and it's knowable: Juristat data shows examiner interviews can increase allowance rates by as much as 9 percentage points (Evaluating Outside Counsel, 2026). Whether to pick up the phone is a data question, not a gut call.
Does this examiner allow cases after an interview? Do they respond better before an RCE? Do interviews help in this art unit, or only on certain rejections? Those patterns tell you when an interview is strategy and when it's just cost.
As much as 9 percentage points. How far examiner interviews can move allowance rates, per Juristat data (Evaluating Outside Counsel, 2026).
Evaluate whether we should request an examiner interview for application [application number] before responding to the current office action.
Consider Examiner [name or ID]'s historical interview behavior, including allowance rate after interview, average number of office actions after interview, and outcomes compared with similar cases where no interview occurred.
Also compare this examiner's interview outcomes to the broader art unit.
The office action includes [briefly describe rejection type, such as a 103 rejection over references A and B].
Provide:
Keep the recommendation practical and tied to prosecution data.
This is where AI supports judgment instead of replacing it. The attorney still makes the call. The call just has the record behind it.
Every amendment that clears a rejection can quietly cost you scope you'll want back later. In a crowded art unit or a high-value portfolio, that trade deserves more than a reflex. Test the amendment before you commit to it: not just whether it overcomes the rejection, but what it costs in scope, what it does to continuation strategy, and whether argument alone would have worked.
Review the proposed amendment to independent claim [claim number] in application [application number].
Compare the proposed amendment against the current rejection, the cited prior art, the specification support, and the examiner's historical treatment of similar amendments.
Tell me:
Provide three alternative response options. For each option, include expected allowance likelihood, claim-scope impact, estimated prosecution cost, and risk of another office action.
This fits the way prosecution decisions actually happen. You're not just trying to get allowed. You're trying to get allowed without giving away more than you had to.
Business development doesn't have to start cold. For firms, some of the best openings are already sitting in public prosecution data: repeated RCEs, long pendency, appeal friction, continuations nobody filed. That's a sharper reason to call than "we'd love to learn about your IP needs."
Analyze [target company]'s U.S. patent prosecution activity over the last [3 or 5] years in technology area [technology area].
Identify business development opportunities for a patent prosecution practice based on prosecution performance, portfolio growth, examiner behavior, art unit concentration, pendency, RCE frequency, appeal activity, and competitor benchmarks.
Compare [target company] to [competitor A], [competitor B], and [competitor C] where relevant.
Provide:
Do not make generic claims. Tie every recommendation to a visible prosecution pattern.
This is the difference between selling services and showing relevance. Patent leaders don't need another pitch. They need a reason to take the meeting.
Clients ask the same three questions: how much longer, how much more, and is it still worth it. "It depends" isn't an answer. A forecast built on the prosecution history, the examiner's behavior, and the art unit's benchmarks is.
83% allowance rate. 1.6 office actions. 0.32 RCEs per application. The published USPTO averages a forecast gets measured against (Juristat Patent Portfolio Performance Scorecard, 2026).
Create a prosecution forecast for application [application number] for a client status discussion.
Use the application's prosecution history, current office action status, claim amendments to date, examiner behavior, art unit benchmarks, and comparable application outcomes.
Provide:
Format the answer in two sections: attorney analysis and client-ready summary. The client-ready summary should be direct, defensible, and suitable for a status email.
This turns messy procedural history into a recommendation you can defend. Firms brief clients with more confidence. In-house teams manage budget and timing with fewer surprises.
Most AI prompts fail for one reason: broad question, thin context. Patent prosecution doesn't reward thin context. Examiner behavior, art unit patterns, rejection type, claim history, RCE history, interview outcomes, and client goals all change the answer.
So the useful prompt never sounds like "summarize this office action." It sounds like this: What should we do next, given this examiner, this rejection, this claim set, this portfolio, and this business goal?
General AI has an opinion on that. Juristat has the record.
Juristat Data Layer (MCP) gives your AI the prosecution context to answer with substance, inside the tools your team already uses.
You don't need another AI experiment. You need better prosecution decisions.
See what your AI does with the actual prosecution record.
By making the model account for examiner behavior, art unit outcomes, response options, and cost, then recommend the path most likely to reach allowance. The output is only as good as the prosecution data behind the prompt.
Specific context: the application number, examiner, rejection type, claim status, prosecution history, art unit, client goal, and the decision you actually need to make. Broad prompts return broad answers.
Yes. It can compare an examiner's post-interview outcomes to similar cases without one and flag the issues most likely to be productive. Juristat data shows examiner interviews can increase allowance rates by as much as 9 percentage points.
By reading public prosecution data for a target, surfacing the portfolio's pain points, and building outreach tied to real friction instead of generic messaging.
Because general LLMs are trained on the open web, not on USPTO prosecution records. Without the record, the model can describe a rejection but not tell you what has worked against it with this examiner.
The connection that puts Juristat's USPTO prosecution data inside the AI tools attorneys already use, including Claude, Copilot, ChatGPT, and Gemini. Not another AI tool. The data that makes those tools reliable for prosecution work.