The evolution of AI in IP: what has changed

An analysis of practitioner sentiment, adoption patterns, and the governance challenges shaping AI’s role in IP practice

New research from the Clarivate Centre for IP and Innovation Research shows that AI adoption in IP has increased from 57% in 2023 to 85% in 2025, marking a shift from pilots to embedded workflows. Confidence correlates with breadth of exposure, but trust remains conditional on governance and explainability. AI is becoming an operational layer of IP practice; its absence, rather than its presence, increasingly requires justification. Viewed through the data, this pattern signals a maturing discipline.

The relationship between artificial intelligence (AI) and intellectual property (IP) has long been implicit in their shared root: intellect. This connection is more than linguistic; it reflects a deeper truth about the field. IP is one of the few systems designed to recognize, protect, and assign value to human creativity according to specific standards. AI, meanwhile, is a technology that seeks to model and extend the capacities of that creativity. As AI settles into IP workflows, practitioners find themselves at an inflection point; deeper integration must be balanced against governance, the effort required to validate outputs, and emerging questions about where efficiency gains accrue across the wider IP ecosystem.

The latest research from Clarivate, The Evolution of AI in IP: Adoption, impact and readiness, offers a longitudinal view of how AI is reshaping practice. The study surveyed more than 400 IP professionals in 2023 and again in 2025 using identical questionnaires to track shifts in sentiment and experience. Respondents answered 13 questions covering user demographics, current and expected applications, implementation challenges, adoption interest, and views on AI’s longer-term role.

A great deal has changed in two years. Large language models have proliferated across industries, and legal tech has attracted significant investment as models become increasingly fine-tuned and domain-specific.[1] This reflects the shift from traditional natural language processing to transformer-based models, which are capable of generating insights rather than simply analyzing text.

IP law continues to be tested. Judicial decisions, including recent inventorship rulings, reaffirm the necessity of human creative contribution. Patent and Trademark Offices (PTOs) and industry bodies[2] have issued updated guidance on AI use in search, drafting, and prosecution, while several offices have begun integrating AI into search, classification, and examination workflows themselves. On the regulatory front, the EU AI Act, the most comprehensive global framework to date, has now entered into force, imposing strict obligations for deployment, including requirements around organizational AI literacy and the quality of training data.[3] Broader governance standards have also matured, with the OECD AI Principles and emerging ISO standards shaping expectations around transparency and accountability.

This context matters because it clarifies what this research does not attempt. It does not address doctrinal questions of IP law or assess model capability. Instead, it examines something often missing in discussions of AI in IP: practitioner sentiment, benchmarked over time. Sentiment reveals whether AI performs as intended in real workflows.[4] Oversight and involvement are preconditions for trust; systems must reflect the realities of those who rely on them, not just the intentions of those who design them.

Understanding these perspectives helps identify barriers, trust gaps, and unintended effects early. These insights are essential for developing AI that is reliable, explainable, and scalable, particularly as scrutiny intensifies around training-data provenance and the quality and completeness of the patent and scientific corpora on which models depend.

How AI is reshaping IP practice

The findings point to a decisive shift from pilot projects to the embedding of AI within IP practice. 85% of respondents now use AI in their workflows, an increase from 57% in 2023.[5]

Adoption, however, is not a proxy for impact. The value of AI depends on how deeply it is integrated and whether outputs meet the evidentiary and expected standards required in each context. Risk tolerance varies: a prior art search, a trademark availability review, and the preparation of a patent application carry very different consequences.

Behind these adoption rates, a clear pattern emerges: confidence correlates strongly with breadth of use. In other words, trust in AI does not develop abstractly; it accrues through exposure. Teams move from scepticism to support when systems prove they can perform reliably across multiple, diverse workflows. Organizations deploying AI across three or more workflows report significantly higher support for continued adoption. Among non-users, scepticism remains high, with detractor rates approaching 70%, likely a reflection of limited exposure to dependable results in practice.

Trust in AI-assisted IP tools develops through experience, not abstract reassurance. In 2023, discomfort was widespread, particularly among attorneys. Two years later, overall comfort has risen by 21 percentage points, though unevenly. Corporate teams are generally more positive; attorneys remain measured, grounded in professional duty. R&D-adjacent roles are the most optimistic, viewing AI as a Socratic tool to accelerate insight. While causation cannot be confirmed from this survey alone, the breadth of exposure is the strongest predictor of confidence.

Value is real, expectations are uneven

Some perceived benefits have remained consistent since 2023. Respondents cite automation of manual tasks (51%), productivity improvements (42%), and time savings for higher-value work (41%). The stability of these figures suggests that these gains have become baseline expectations, now accompanied by concerns about explainability, defensibility, and governance.

Use cases reflect this maturation. What began with routine automation and advanced semantic search algorithms for more efficient IP research now extends across competitive, technical, and market intelligence (37%), research and discovery (36%), and patentability, clearance, and invalidity analysis (35%). These areas deal with high information volumes and repeatable, evaluable tasks where AI can enhance efficiency without undermining professional judgment.

For practitioners, this clarifies where AI is delivering dependable value and where attention is most warranted. It distinguishes the workflows in which AI is genuinely maturing from those where verification burdens still outweigh the benefits. These distinctions will quietly shape adoption decisions in the years ahead. They are already visible in the differing patterns of use across the profession. Law firms, for instance, show a distinct profile. Their most cited use case is support for drafting and application preparation at 33%. These tasks require context-specific reasoning and benefit from tools that assist rather than replace attorney judgment. Corporations adopt more broadly across the IP lifecycle, prioritising patent search tools at 60%, in-house capabilities at 33% and monitoring and classification tools at 30% and 28%. Law firms adopt more selectively, focusing on patent search (40%), drafting (35%), and trademark search (29%), balancing innovation with client trust and professional standards.

Another divergence is the gap between general-purpose AI and purpose-built IP tools. Nearly 75% of respondents use platforms such as Microsoft Copilot or ChatGPT, reflecting accessibility, cost, and ease of use. Uptake of specialized IP tools is lower, pointing to a mismatch between general AI capability and the precision required for prior art searches, portfolio analytics, and structured decision-making. Adoption, in short, is deliberate rather than indiscriminate.

It also highlights the inherent limitations of general models that are not trained on curated IP corpora or classification schemas. In areas where novelty, distinctiveness, and technical detail are central to legal assessment, the absence of domain-specific grounding can introduce error modes that are difficult for practitioners to detect or explain.

Scaling AI requires managing risk

Privacy, liability, and explainability remain the primary barriers to wider adoption, with 65% of attorneys citing these as the dominant constraints. These concerns are grounded in professional obligation rather than resistance to innovation. Professional competence rules in several jurisdictions now explicitly reference the need to understand the benefits and risks of technologies used in legal work, reinforcing this duty. In this research, transparency was the most cited barrier, followed by security readiness and privacy, especially in workflows involving sensitive disclosures. Around one-third of respondents identify training and implementation gaps, which reflects uneven enablement across roles and regions.

Both corporates and law firms emphasise the need for structured governance: clear processes, standards, and oversight mechanisms that allow practitioners to understand and, where necessary, challenge system behavior. Support for AI increases markedly once teams adopt it across multiple workflows, but scaling responsibly depends less on technical sophistication and more on organizational readiness. Effective adoption requires transparent data lineage, role-specific training, workflow-level guardrails, and continuous monitoring of accuracy, bias, and error patterns.

These efficiency gains are also recalibrating expectations about how legal and IP teams deliver value. The effects are already visible: tighter matter-triage, more consistent drafting, expanded operational support, and more flexible resourcing models. Taken together, they represent the quiet industrialization of specialist practice. Done well, this shift creates the possibility for practitioners to spend more time on higher-quality, more strategic work; however, done hastily, it risks reshaping workloads without materially improving the experience of the individuals doing the work. The task now is to identify where technology can safely carry more of the load. How teams navigate this balance will shape what comes next, and the broader patterns revealed in this research that make the trajectory clearer.

The next chapter

This research confirms familiar trends while adding necessary context on how AI is reshaping the logic of IP practice. The shift is no longer from scepticism to adoption, but from asking whether AI belongs in IP workflows to understanding what kinds of reasoning it can legitimately support.

Adoption is widespread, yet trust remains contingent. Progress depends on validated use cases, strong governance, and deployment practices aligned with IP standards. The rise of large language models has enabled increasingly autonomous workflows and agentic tools, raising new questions about where system autonomy ends and accountability begins.

For some, AI functions as a supportive tool that enhances judgment; for others, its integration reveals structural tensions within the IP landscape. Regardless of viewpoint, growing compliance expectations will shape practice and influence the development of IP policy and law.

What this study consistently shows is that IP, and the legal work that underpins it, maintains a necessary boundary around the tasks that must remain human. That boundary must be minded; it is regulated, but not impermeable, and the increasing fluency of AI risks blurring where assistance ends and responsibility begins. Within it, there is still room for measured innovation and a space for thoughtful experimentation.

Adoption is advancing most rapidly in the layers surrounding these core responsibilities, where AI delivers value in research, analysis, and operational tasks. Conversely, for tasks such as issuing legal opinions or drafting office responses, the evidentiary demands and liability implications exceed what AI can reliably provide. These activities create legal effect and require forms of warranted justification and accountable intent that a statistical system cannot, even in principle, supply.

As one of the most demanding environments for AI deployment, IP offers early insight into how intelligent systems might integrate into other fields. Ultimately, AI will secure its place only when it strengthens the discipline, scrutiny, and judgment that underpin IP practice. The next chapter will turn on what can be demonstrated; in the end, that is the test that endures beyond tools and trends.

[1] Legal tech investment trends and AI adoption: See Law.com, “Clio Acquires Lawyaw, Hits $5B Valuation as Legal Tech Embraces AI” (2024).

[2] The European Patent Institute (epi) has also published guidance on generative AI, emphasizing confidentiality, client consent, and attorney accountability for AI-assisted work. See epi Guidelines on the Use of Generative AI in Patent Practice

[3] Article 4, EU AI Act

[5] Respondents represented a geographically broad but regionally weighted sample: 44% Europe, 26% North America, 21% Asia-Pacific, 8% Middle East/North Africa, and 1% South America. The survey reflected the diversity of the IP ecosystem: 55% focused primarily on patents, 16% on trademarks, and 27% on both. Roles included 75% corporate professionals and 25% law firm practitioners, with 37% attorneys or in-house counsel, 22% executives, 10% R&D professionals, and the remainder in operations, paralegal, analyst, and administrative positions.