‍

The habit that stays with you longest after leaving the patent office is not the one you expect. It’s not the classification codes or the database shortcuts or the institutional vocabulary—it’s slower than any of those. It’s the habit of reading a claim the way a pathologist reads a microscope slide: element by element, without skipping ahead to the diagnosis. 

Under MPEP § 904, examiners are required to search the prior art as disclosed in patents and other published documents after obtaining a thorough understanding of the invention as claimed—not as described in the abstract, not as the inventor intended, but as the claims actually read.

I left the office over a decade ago. I now advise the people who used to sit on the other side of that exchange: inventors, university researchers, early-stage companies, R&D teams that have built real things and are trying to figure out what protection is actually available to them. 

What I find consistently is that the single most expensive mistake they make has nothing to do with claim drafting or prosecution strategy. It’s that they treat prior art search as a task you complete, rather than a process you maintain.

They do it once. Early. Usually before anyone has thought carefully about what the claims actually need to cover. And then they move on.

The invention, meanwhile, keeps changing.

‍

What Examining Teaches You That Private Practice Confirms

When I trained as an examiner, the methodology was specific enough that it felt almost mechanical: parse the independent claim into its elements, search each element against the prior art, map your findings to the claim language, document what you found and where you found it. 

The goal was not to find a reason to reject. The goal was to determine, with documented evidence, whether the combination claimed was already known. Under 35 U.S.C. § 102, a claim is anticipated when a single prior art reference discloses each and every element of the claim as arranged therein. 

Under 35 U.S.C. § 103, a claim is obvious when the differences between the claimed invention and the prior art are such that the subject matter would have been obvious to a person having ordinary skill in the art. Those two standards—anticipation and obviousness—drive everything. And they drive it at the element level, not the invention level.

That discipline is not natural for most inventors. Researchers think about their work in terms of results: what the technology does, what problem it solves, why it matters. 

Claim language works differently. A claim does not describe a result. It describes a structure, a method, a composition—the specific elements and their relationships. The Federal Circuit, sitting en banc in Phillips v. AWH Corp., confirmed that the ordinary and customary meaning of a claim term is the meaning a person of ordinary skill in the art would ascribe to it at the time of the invention, in the context of the specification and prosecution history. 

‍

‍

The question of patentability lives entirely in that comparison—between the claim elements as properly construed and the prior art disclosure—not in the result the invention produces.

I spent the first year of my private practice having the same conversation in different rooms. A researcher walks in with a thorough literature review, confident. I ask which classification codes she searched. Silence. I ask whether she searched the patent literature or only journals. More silence. I ask her to describe her invention in claim-element terms, without referring to the result it produces. That is usually the moment the conversation shifts.

This is not a criticism of researchers. It describes a gap in how IP work gets introduced to people who generate IP for a living. Most scientists and engineers encounter patent analysis at the moment of disclosure, which is often too late for the analysis to change anything. By then, the technical choices are fixed, the publication timeline is set, and the claims will be shaped around the invention rather than the other way around.

‍

The Problem With Doing it Once

The conventional approach to prior art search treats it as a gate. You run the search before you file. You get an opinion before you launch. Both are necessary, but neither is sufficient.

Here is what the gate model misses: inventions are not static between the gate and the market. Claims narrow during prosecution. Features get added, dropped, combined. The commercial application shifts. A product that started as a laboratory tool gets repositioned as a consumer device and suddenly reads on a patent class nobody searched. A start-up pivots and its core technology, which had a clean freedom to operate (FTO) in its original use case, now operates in a space with three active patent families nobody reviewed.

I advised a company in 2022 that had clean IP clearance at seed stage. By the time they were preparing their Series A materials, a competitor had filed two continuation patents in the 18 months since the original FTO. Both continuations were narrow and specific to the exact product feature that had become the company's main commercial differentiator. Nobody had been watching. The original opinion was in a folder. The company spent four months on a design-around that could have been caught in a quarterly monitoring pass. 

MPEP § 904.03 is explicit that when a response is received from an applicant, an update of the original search shall be made to ensure its continued completeness. The principle applies with equal force in private practice, where no procedural rule compels the update—which is precisely why teams fail to make it.

Four months of engineering time, plus outside counsel fees, plus the distraction at the worst possible moment in a financing cycle. The cost was not theoretical.

There is a documentation problem underneath the timing problem. Most one-off searches are under-documented in a specific way: they record findings but not scope. Which databases were searched. Which classification codes were selected and why. Which synonyms and language variants were included. Which adjacent technology spaces were excluded. 

When the analysis needs to be revisited—and it always does—there is nothing to work from. You cannot update a search that did not record its own assumptions. You have to start over.

‍

What a Workflow Actually Looks Like

The word "workflow" makes this sound more bureaucratic than it needs to be. What I mean is simpler: instead of a search that produces a report and ends, you build a process that produces a report, records its assumptions, and knows when to run again.

In practice, a prior art monitoring protocol opens at the same time a new technical project opens. It is not a full analysis. It is a structured record: the technology domain, the relevant IPC and CPC codes, the competitor set, the claim elements in scope, the key date anchors, and the conditions that will trigger a refresh or an escalation. Setting it up properly takes a few hours. What it provides over the life of the project is disproportionate to that investment.

The initial findings document is more specific than a standard search report. Not "no anticipating references found" but "no anticipating references found for the following elements, within the following scope, using the following search logic, as of this date." 

Every claim element either maps to evidence in the prior art or to a documented statement of absence. The absence statement is actually more useful than most teams realize. It defines the boundary of what was searched. Any analyst who picks up the work later can see exactly where the original search stopped.

‍

‍

The legal stakes behind this are concrete. In Halo Electronics, Inc. v. Pulse Electronics, Inc., the US Supreme Court confirmed that courts may increase patent infringement damages up to three times under 35 USC §284, and that culpability is measured against what the defendant knew at the time of the challenged conduct, not what defenses it assembled for litigation. 

The decision effectively raised the value of documented, contemporaneous FTO work. A company that can show a structured, updated IP clearance process is in a meaningfully different position than one relying on a stale opinion and a post-hoc litigation defense.

Refreshes are triggered by events, not just calendar intervals. A material change in claims. A new competitor patent in the relevant classification. A shift in the commercial use case. The approach of a filing deadline. The escalation rules—when a preliminary review needs to become a formal legal opinion, when outside counsel must be engaged, when a design-around assessment is warranted—get defined at the outset, not when a problem has already appeared. You do not want to be designing the escalation process while you are in the middle of escalating.

‍

Where AI Fits, and Where it Doesn’t

I want to be precise here because the conversation around AI and patent search tends toward either dismissal or overstatement, and neither is useful.

Semantic search is a genuine improvement. The classic failure mode of keyword-based analysis is vocabulary mismatch: the examiner uses one term, the prior art uses another—a synonym, a regional variant, earlier nomenclature for the same concept. 

Semantic search reduces that gap by working from meaning rather than exact text. Relevant references surface that would not have appeared under keyword logic. The WIPO Technology Trends 2019: Artificial Intelligence Report documented that AI-related patent applications had grown rapidly, with over 340,000 filed since AI first emerged—a body of prior art that keyword-only search handles poorly, given the field's shifting terminology. 

The growth in AI-adjacent patent literature across technology domains means that vocabulary-based search failure is increasingly common, not only in AI-specific fields but wherever AI techniques are being applied.

Automated monitoring is also genuinely useful. Watching a defined classification space and flagging new publications does not require an analyst to run the same search by hand every quarter. The cost of maintaining a live prior art process has dropped considerably. For smaller teams, for university technology transfer offices with limited IP budgets, that change is practically significant.

‍

‍

What AI cannot do is the legal work. Whether a reference anticipates a claim requires interpreting both the claim and the reference under the applicable claim construction standard—the ordinary and customary meaning of a claim term as a person of ordinary skill in the art would understand it, as Phillips confirmed—then deciding whether each element reads on the disclosure. 

That is legal analysis. AI retrieves the reference. It does not determine whether the reference, properly construed and mapped against the specific claim language at issue, is anticipating. A person has to do that, and that person has to be accountable for the conclusion.

‍

The Five-Point Framework That Turns Prior Art Search Into a Living Process

I have set up versions of this framework with R&D teams at public universities, with start-ups preparing for due diligence, with companies entering markets where IP clearance is a regulatory or licensing prerequisite. What I describe below is not theoretical. It’s what I have found, through repeated failure and adjustment, actually works.

‍

1. Scope Documentation Before the Search Starts

Someone writes down what is being searched and why: the invention as currently described, the claim elements in scope, the technology domain, the competitor set, the classification codes selected, and the rationale for any exclusions. 

This record is the baseline. It is what makes scope drift visible later and what allows the work to be picked up and continued rather than restarted. The MPEP's requirement that examiners document their field of search exists for exactly this reason: reproducibility and auditability. 

Private practice needs the same discipline, without the procedural mandate to enforce it.

‍

2. A Decision Log That Records Conclusions and Their Uncertainty 

A file that simply records, "The reference does not anticipate element three…" is useful. 

A file that adds, "...but the claim construction on that element is not settled and this conclusion should be revisited once prosecution history develops," is honest. 

Most files stop at the first sentence. That is a problem waiting to become expensive.

‍

3. Event-Based Refresh Triggers Alongside Any Scheduled Review 

The list of triggering events gets agreed on at the outset: a material claim change, a new competitor filing in the relevant CPC subclass, a product feature modification, a change in the commercial scope of the activity being cleared. 

Different projects have different lists. What matters is that the list exists and someone watches it.

‍

4. Role Clarity (Harder to Achieve Than it Sounds) 

The most common breakdown I see isn’t a missing search or a missed reference, it’s confusion about who is responsible for what judgment. 

The inventor describes the technical solution in claim-element terms. The analyst searches within defined scope. The patent professional makes the legal determination. The business leader decides what level of risk is acceptable. 

These roles overlap at the edges. They should not collapse into each other.

‍

5. Escalation Rules Defined Before They Are Needed

Without pre-defined thresholds, teams either commission formal opinions for every preliminary finding—expensive and slow—or they never formalize conclusions that carry real legal weight. 

In the second case, organizations end up carrying undocumented oral FTO understandings that nobody can reconstruct when a transaction or a dispute requires documentation. Given that Halo ties enhanced damages exposure to what the defendant knew at the time of infringement, the absence of a documented clearance process isn’t a neutral fact. It’s potential evidence of recklessness.

‍

What This Framework Does For Risk

‍

‍

I want to be honest about what this framework can and cannot do.

No prior art analysis eliminates the possibility that a relevant reference exists somewhere in the global patent literature. No FTO opinion guarantees a product will never face an infringement claim. Those guarantees are not available. Anyone who implies otherwise is misrepresenting what legal analysis delivers.

What a documented, maintained workflow provides is a different thing: it makes the uncertainty visible, records the assumptions underlying each conclusion, and gives the organization a mechanism for incorporating new information as it arrives. That changes the organization's relationship with the risk it is carrying. The risk does not disappear, but becomes manageable in a way that undocumented risk is not.

The due diligence scenario is the one I return to most often with clients. A potential acquirer will not only ask whether an FTO was obtained. They will ask when it was last updated. They will ask what the team did when the competitor's continuation patents were published in the year before the transaction. 

A team with a documented refresh history can answer those questions and can show that business decisions made in response to new information were deliberate, made with legal context, by the right people. A team without that history has to explain why they cannot.

In litigation, the evidentiary picture follows from Halo: culpability is measured against the knowledge of the actor at the time of the challenged conduct. A party that can produce a documented prior art review—with stated scope, documented assumptions, refresh history, and records of the decisions made in response—is in a different position than a party that can only show a single analysis performed before launch. 

The workflow does not prevent litigation. It changes the quality of the record if litigation comes.

‍

For Inventors, Researchers, and the Professionals Who Advise Them

For researchers in university settings, the timing issue matters most. The prior art search that actually helps an inventor make better technical choices is the one that runs while the invention is still being formed—not after the disclosure form has been filed and the technology transfer office is already drafting claims. 

That requires a shift in practice culture. It also requires technology transfer professionals willing to engage research teams before the invention is finished, which is earlier than most TTO workflows begin. The strategic value is there. So is the awkwardness of having IP conversations before researchers think they need them.

For patent professionals, the shift toward workflow architecture changes the engagement model. The durable service is no longer the one-time search or the single FTO opinion, though those remain necessary. It is the design and maintenance of the decision infrastructure—helping clients build habits that make IP risk management continuous rather than reactive. This is a different service with different deliverables. I find it more interesting than the transactional model. Most clients, once they have experienced both, agree.

For R&D directors and business leaders, the question is resource allocation and organizational honesty. A well-maintained prior art workflow has an upfront cost: time to design the scope, define the roles, agree on escalation logic. That cost is real. The return is also real. 

The cost of discovering a blocking patent at product launch, or explaining to an acquirer why the IP landscape was last reviewed two years before the transaction, is almost always larger. The math is not complicated. The barrier is usually inertia.

‍

Conclusion

I want to end with something I genuinely believe, not a summary of what I have already said.

The examiner's discipline—reading a claim slowly, mapping each element, documenting what you found and where, recording what you excluded and why—isn’t a relic of bureaucratic procedure. It’s the only honest way to do this work. It’s honest because it’s explicit about what was looked at and what was not. It’s honest about the conclusions it can support and the ones it cannot. It’s honest about uncertainty in a field that is structurally full of it.

Most IP work in private practice does not have that quality. Not because the practitioners are careless (most are not), but because the engagements were never designed for it. A one-time search does not know what it missed. A static FTO does not flag itself as eighteen months old when you open it. A decision made on the basis of prior art analysis that predates three competitor continuations does not announce that it was made without that information. 

The obviousness framework the US Supreme Court elaborated in Graham v. John Deere and refined in KSR International Co. v. Teleflex Inc. requires factual inquiry into the scope and content of the prior art at the time of invention. That inquiry cannot be performed adequately with stale data and undocumented assumptions.

AI tools help. They lower the cost of coverage, improve recall, and make continuous monitoring feasible for teams that could not have sustained it before. They do not change the core requirement: a practitioner who can read a claim, understand the technology, compare one to the other with evidentiary discipline, and commit to a conclusion.

The inventions that get protected are not always the most creative ones. More often, they belong to the teams that understood the landscape early enough to maneuver within it—before the claims were locked, before the product launched, before the competitor's continuation published and changed the picture entirely.

That understanding doesn’t come from a single search. It comes from a system.

‍