USPTO Raises the Bar for AI-Assisted PTAB Expert Declarations
On July 2, the U.S. Patent and Trademark Office issued a memorandum aimed squarely at expert testimony in PTAB proceedings. The immediate point is not that AI tools are forbidden. It is that expert declarations containing claim charts, technical comparisons, infringement-style mappings, or data extrapolations assisted by large language models and similar tools must now be accompanied by a sworn statement of independent verification.
The memorandum matters because it reframes AI use as an evidentiary reliability issue, not merely a drafting issue. If the extent of AI involvement is not candidly disclosed, or if the expert cannot show meaningful human review of the analysis that made its way into the declaration, the PTAB may treat those portions as hearsay and exclude them. For parties preparing IPR and other post-grant records, that changes how expert evidence must be built, documented, and defended.
This is less about disclosure etiquette and more about admissibility
A fair number of practitioners will first read this as another AI disclosure rule. That is too narrow. The sharper question raised by the memorandum is whether the declaration still qualifies as the expert’s own testimony when critical analytical steps were generated, structured, or heavily shaped by a model. Once that line blurs, the problem is no longer optics. It becomes a threshold evidentiary problem.
That is why the independent-verification affidavit matters. The PTAB is not asking parties to add a cosmetic sentence saying the expert reviewed the work. It is asking them to convert human review into something visible in the record. If they cannot do that, the other side has an opening to argue that what appears in the declaration is effectively a repackaged machine statement rather than expert opinion grounded in the witness’s own analysis. In a hard-fought post-grant proceeding, that can be far more damaging than a routine credibility attack.
The next fight will focus on how the analysis was produced
PTAB battles over expert declarations have traditionally centered on obviousness reasoning, the completeness of a motivation-to-combine theory, or whether the witness truly reflects the perspective of a person of ordinary skill in the art. Those fights are not going away. But there is now a new line of attack that will be difficult to avoid: show us the workflow. If claim charts, terminology mappings, experimental inferences, or prior-art summaries bear the marks of generative tooling, opposing counsel will want to know what tool was used, what it was asked to do, how much the output was changed, and which underlying materials the expert personally reviewed.
That is a meaningful shift in pressure. An expert who merely “looked over” an AI-assisted draft may no longer be enough. The safer position is one in which the witness can explain, in practical detail, which parts were independently prepared, which parts involved AI-assisted organization or drafting help, and how the expert checked every factual citation and inferential step before adopting the final language. AI can remain a tool. What it cannot be, at least not safely, is an invisible co-author hiding inside the declaration.
For IPR strategy, the impact lands first on timing, cost, and record design
The memorandum may not change case outcomes overnight, but it is likely to change preparation discipline almost immediately. Many teams have grown comfortable using AI systems to accelerate charting, summarize references, organize technical comparisons, and generate early analytical scaffolding before experts and lawyers tighten the draft. That workflow is not necessarily dead. It is simply harder to defend if it leaves no clean audit trail of what the expert actually verified.
That means more work moves upstream. Experts may need to be involved earlier. Counsel will need to think sooner about what kinds of internal notes, checklists, and review records should exist if the declaration is later challenged. Trying to add an independent-verification statement at the end of the process, after the file has already been built in a black box, is the kind of move that reads fine on paper but breaks down under questioning. The practical result is a more expensive front end, but also a more resilient record if done well.
Teams do not need to abandon AI, but they do need new internal rules
The wrong reaction is to assume that any use of generative AI in PTAB-related expert work is now too dangerous. That is neither realistic nor, in many cases, necessary. The better response is to stop treating AI as an invisible productivity layer and start treating it as a governed support tool. Teams should decide now which tasks can be delegated to AI-assisted workflows, which tasks must be performed from scratch by the expert, and which outputs require contemporaneous proof of human verification before they can enter a declaration.
In practical terms, any expert material that may end up in the PTAB record should have a pre-planned disclosure position, a verification protocol, and a document-retention approach that matches the level of risk. Petitioners will need that discipline to protect key evidence from being cut back or excluded. Patent owners, for their part, have gained a sharper way to probe and attack expert submissions that appear polished but thinly sourced. The message from the USPTO is not anti-AI. It is anti-opacity. Efficiency is welcome, but the evidentiary burden still belongs to the party and the witness who submit the declaration.
