Reflections from IPWatchdog: Masterclass on AI Prompt Engineering for Patent Workflows

Why we hosted this masterclass

Industry research from Thomson Reuters shows that 80% of legal professionals believe AI will have a high or transformational impact on their work over the next five years, and 53% say they are already seeing ROI. A BCG and Spellbook study found that lawyers using AI scored 94% accuracy compared to 85% for those who did not, with an 81% median time saving across tasks.

At Solve Intelligence, we’ve seen this play out directly. The number of IP professionals using our platform grew by over 560% in 2025. 

The case for AI in patent work is no longer speculative but realising the return on investment requires work. At Solve Intelligence, we’ve supported more than 700+ IP teams across six continents to realize 30% to 60% time savings across patent workflows. 

But the same research that shows those gains also shows that good AI knowledge is the primary driver of whether professionals actually realise them, and organisations where practitioners have strong AI knowledge are 2.8x more likely to benefit. 50% of AI performance improvements come not from better models, but from better prompts.

The panel 

Gene Quinn (President & CEO, IPWatchdog), Johnathon Webb (Partner, Faegre Drinker), and our own Dr. Chris Parsonson (Founder & CEO) and Justin Doop (Founding Member & General Counsel) to focus specifically on the skill that matters most right now: prompt engineering for patent workflows.

What the polls revealed

To get to know the audience, and tailor our conversation, we asked several poll questions, and the results were worth paying attention to. 

1. What is your level of experience?

This is a senior, experienced audience. 35% had 11–20 years of patent experience, and 40% had more than 20. Only 15% had fewer than six years of practice.

2. What work do you do?

Patent application drafting (86%) and office action responses (83%) dominated, as expected. But the breadth was notable, and indicated a large in-house presence: 69% handle portfolio management and invention harvesting, 63% handle FTO and clearance analysis, and 58% handle invalidity analysis. 

3. What workflows are you using AI for?

Summarising documents and research led the field (72%), followed by drafting patent applications (51%) and responding to office actions (44%). 

FTO, invalidity, and claim charting came in at 21%; a number we expect to grow significantly as more teams move beyond drafting and into more complex workflows. The release of Solve Intelligence’s Claim Charts product demonstrates what we see in the market. 15% said they are not currently using specialised AI tools at all in any of the aforementioned workflows. 

4. Is AI approved for use in your organisation?

The results were more positive than most would expect:

• 39% said AI is approved for all or most matters
• 42% said AI is approved for some matters
• Only 2% said AI use is not permitted
• 11% are unsure or still reviewing AI policy, and 6% are still seeking approval

In other words, over 80% of respondents are working in environments where AI is either approved or actively being evaluated. The permission question, for most, is already settled. The question now is how to use AI well and scale the use of AI across teams. 

The 6 tips that drove the conversation

Solve Intelligence speakers walked through the framework we use internally and teach to every IP team we work with. The logic is simple: prompt engineering delivers roughly 90% of the value of AI customisation with around 1% of the effort of fine-tuning a model. For patent professionals, it’s the fastest and most practical way to improve AI output quality.

The 6 tips covered in the masterclass:

1. Be clear and direct. Treat the AI like a smart new colleague with no context on your matter. State the task, the format, the purpose, and the audience. Do not assume it will infer what you need.
   
   
2. Use XML tags. Wrapping instructions, documents, and examples in XML tags prevents the AI from confusing your instructions with the content you are asking it to analyse.
   
   
3. Use examples. Providing 2–5 examples of what a good output looks like, whether that’s a well-structured claim, a correctly formatted chart cell, or a particular drafting style, dramatically improves consistency across responses.
   
   
4. Give the AI a role. Telling the AI it’s a senior patent attorney preparing a §103 response for a medical device client produces materially different output than a generic instruction. The more specific the role, the better the result.
   
   
5. Use chain-of-thought prompting. For complex tasks, such as prior art analysis, claim construction, validity opinions, asking the AI to think step-by-step before reaching a conclusion can double accuracy.
   
   
6. Build and share prompt templates. Once you have iterated on a prompt that works well for a particular client, jurisdiction, or technology area, save it. Share it across your team. 

This is where the flywheel of AI ROI begins: 5–10% of a firm's practitioners typically become power users, and their best prompts can then be distributed to the rest of the organisation.

Jonathan Webb from Faegre Drinker, an active user of Solve Intelligence and Customer Advisory Board member, made a point that stayed with us: the way experienced patent attorneys already think (starting with claims, building figures, constructing the detailed description) maps naturally onto how AI works best when given structured, stepwise instructions. Good patent practice and good prompting turn out to have more in common than most people expect. At Solve Intelligence, we have built an intuitive and iterative platform that focuses on user experience and high quality output.