Free tools Windows power users keep installed
One-click scans. No signup required.
iTechGuides is reader-supported. When you buy through links on our site, we may earn an affiliate commission. As an Amazon Associate I earn from qualifying purchases. Learn more
Do not file AI-generated patent claims or citations without independent verification. Treat the model’s text as an unverified draft. For every independent and dependent claim, chart each limitation against the specification, drawings, inventor evidence, and the actual prior-art record. Open the original patent publication or official prosecution file, confirm its identity and dates, and read the cited passage in context. Search independently for uncited prior art, preserve an audit trail, and have a registered U.S. patent attorney or agent resolve uncertain claim scope, inventorship, materiality, or filing issues.
This workflow is for U.S. practice. USPTO guidance says the same inventorship standard applies whether or not AI was used: only natural persons can be inventors, and an AI system is a tool rather than an inventor. Rules, guidance, and prosecution records can change, so confirm them immediately before filing.
Why must AI-generated patent work be checked before filing?
Using AI does not reduce the filing team’s legal duties. The Federal Register states that every individual associated with filing and prosecution has “a duty of candor and good faith in dealing with the USPTO.” Known information material to patentability must be disclosed, and the circumstances can create a duty of reasonable inquiry. Failing to investigate a suspicious assertion can jeopardize validity or lead to sanctions under 37 CFR 11.18(c).
An invented patent number, a passage that says something different from the model’s summary, an unsupported test result, or a claim limitation missing from the specification can therefore become a legal problem, not merely a technical typo. USPTO guidance on AI tools says existing rules govern practice before the Office and recommends technical safeguards and human governance.
#1 Best Overall
The scale of AI-assisted work makes disciplined checking more important. The USPTO Patent Public Advisory Committee’s 2024 annual report recorded annual AI patent applications rising from almost 76,000 in 2018 to just over 101,000 in 2023, a 33% increase. More applications do not change the standard: patentability, disclosure, and inventorship remain claim-specific and evidence-based.
What should I save before changing an AI-written application?
Freeze the exact draft first. If you edit while checking, you can lose the ability to show what the model produced, what a human changed, and why a statement was accepted or rejected.
- Save the complete AI output. Preserve the claims, specification, drawings or drawing instructions, abstracts, cited references, and any generated experimental or legal assertions.
- Record provenance. Note the model and version if available, the date and time, the prompts, attached source files, retrieval settings, and the person who accepted each substantive change.
- Mark everything unverified. Tag every factual assertion, numerical result, quotation, patent number, date, and legal proposition until a reviewer checks it against an authoritative source.
- Separate source material from generated text. Keep inventor notebooks, laboratory records, design files, emails, and earlier drafts in an immutable folder so a reviewer can distinguish original evidence from AI wording.
- Assign ownership. Identify the human reviewer responsible for each claim set and each citation. A generic “AI checked” note is not an accountable review record.
How do I break each claim into verifiable limitations?
Read each independent claim as a checklist, then repeat the process for every dependent claim after adding its new limitations. Do not verify only the claim’s overall idea; a single unsupported element can determine validity, infringement scope, or written-description support.
Classify the limitation
- Structural: components, layers, modules, materials, or chemical entities.
- Functional: what a component is configured to do or what a process step accomplishes.
- Relational: how elements are connected, ordered, positioned, or conditioned relative to one another.
- Parameter-based: ranges, thresholds, timing, dimensions, concentrations, performance values, or mathematical relationships.
Check claim mechanics at the same time
- Confirm antecedent basis: every “the” or “said” element must have been introduced clearly.
- Check dependency: a dependent claim must include every limitation of the claim it depends on plus its added limitation.
- Use terminology consistently between claims, the specification, and drawings.
- Look for accidental alternatives, contradictions, or a parameter range that the specification never teaches how to make or use across its full scope.
Build a limitation chart
Use one row per limitation and identify the exact evidence, not a section-level impression.
| Claim item | What to record | Acceptable evidence | Result |
|---|---|---|---|
| Limitation text | Verbatim claim language and a plain-language interpretation | Claim construction notes approved by the reviewer | Clear, ambiguous, or inconsistent |
| Specification support | Paragraph, page, figure, or example identifier | Disclosure describing the element and its relationship | Full, partial, or none |
| Enablement support | How a skilled person would make and use the claimed scope | Detailed methods, parameters, working examples, or justified routine steps | Supported, uncertain, or unsupported |
| Inventor evidence | Who conceived the feature and when | Dated notebooks, design records, or corroborated testimony | Attributable or unresolved |
| Prior-art evidence | Reference, exact passage, and public-availability date | Original publication or official record read in context | Potentially anticipatory, relevant but incomplete, or not located |
Do not let a polished AI explanation substitute for a pinpoint citation. A limitation is not “supported” merely because the specification uses similar words; the disclosure must support the claimed relationship and scope.
How can I tell whether ChatGPT invented a patent citation?
Start with the official record, not the AI-provided URL, hyperlink, snippet, or quotation. A real publication number can still be paired with the wrong title, jurisdiction, inventor, date, family member, or passage.
- Locate the original record independently. Search an official patent database or the relevant official prosecution system using the publication or application number, title, inventor, and key terms.
- Confirm identity fields. Record the exact title, all listed inventors, publication number, application number when available, filing date, priority date, jurisdiction, and publication date.
- Trace the family. Check whether the cited document is a U.S. publication, an international application, a national-stage filing, or another family member. Do not treat different family members as interchangeable without checking their text and dates.
- Check prosecution status. Determine whether the document is an application publication, issued patent, reexamination, appeal, or another official record. Status affects what a statement about allowance or claim scope means.
- Read the cited passage in context. Capture surrounding paragraphs, definitions, figures, incorporated references, and claim language. Decide whether it supports the exact proposition, only a narrower proposition, or nothing.
- Record the audit result. Save the database name, search date, query, record identifier, page or paragraph location, and your support classification.
Common hallucination signals include a number with a plausible format but no official record, a quotation that cannot be found, a priority date later than the publication it supposedly predates, an inventor who is absent from the record, or a citation that describes a different technology. Mark a failed citation as a failed citation; do not silently replace it with a similar document.
How should I search for prior art the AI did not cite?
A citation’s existence is not proof of novelty or non-obviousness. Search the claims independently, including material the model omitted.
- Extract search concepts. Turn each limitation into synonyms, component names, functions, relationships, numerical ranges, and likely classification terms. Search unusual combinations as well as individual elements.
- Use multiple source types. Search official patent databases and relevant non-patent literature. Include published applications, issued patents, conference papers, standards, technical manuals, product documentation, and other material that may have been publicly available.
- Vary the query. Run exact phrases, synonym combinations, classification searches, inventor or assignee searches, citation chaining, and date-bounded searches. Record searches that produced no results as well as useful results.
- Establish public availability. For each candidate, document the publication date or other evidence that it was publicly accessible before the relevant effective filing or priority date.
- Map, do not merely collect. Place each reference in the limitation chart. Identify whether one reference discloses every limitation, or whether multiple references would be needed for an obviousness analysis.
Your log should state the databases, search strings, classification codes if used, date ranges, search dates, and results reviewed. A later reviewer must be able to reproduce the search and understand why a reference was considered material or not.
What patentability and drafting checks are still required?
AI can suggest language, but it cannot make the legal determination that a claim is patentable. Counsel should test at least novelty, obviousness, written description, enablement, definiteness, and claim construction against the actual record.
Written description and enablement
Ask whether the application shows possession of the claimed invention and teaches a skilled person how to make and use the full claimed scope without undue experimentation. AI often broadens a disclosed example into a genus or adds performance ranges that the inventors never possessed. Each such expansion needs evidence or must be removed or narrowed.
Recommended Free Tools
Definiteness and construction
Define terms that control scope, resolve functional language, and test relative terms such as “substantially,” “rapid,” or “low.” Check that the specification supplies an objective boundary where one is needed and that the claims do not shift meaning between embodiments.
Eligibility for AI-related subject matter
Use the USPTO’s existing MPEP framework. Ask whether the claim recites an abstract idea and, if so, whether it integrates that idea into a practical application. The USPTO’s 2024 update emphasizes a case-by-case analysis; no keyword, including “AI,” decides eligibility.
Experimental and performance assertions
Require laboratory records, protocols, raw data, and a named human responsible for every generated result. If no experiment occurred, remove language that implies one did. A model’s plausible number is not evidence.
Can AI be listed as an inventor?
No. Under the USPTO’s current AI resources and revised guidance, only natural persons can be inventors. The revised guidance rescinded the February 2024 AI-specific inventorship guidance but did not change the human-inventor requirement. An AI system may generate options or combinations; it cannot be named as an inventor.
Windows Errors? Fix Them Before They Spread
Repair common Windows errors and clear accumulated junk for a smoother, more stable PC - no reinstall needed.Free scan · no reinstallOutdated Drivers Are Slowing You Down
One free scan finds every outdated or missing driver and matches the right update for your exact hardware.Free scan · exact hardware match- For each claim, identify the natural person who contributed to conception of the claimed subject matter.
- Distinguish proposing an option from recognizing and developing the claimed solution. Document the human’s technical contribution, selection, modification, or verification.
- Retain dated records showing conception and collaboration, including prompts and outputs when they help explain the sequence of events.
- Have counsel resolve joint-inventorship questions, omissions, or additions before signing the application papers.
Do not infer inventorship from who operated the software, paid for access, or edited prose. The relevant question is the human contribution to conception of the claimed invention.
What disclosure and reasonable-inquiry duties apply when AI was used?
People associated with filing and prosecution must disclose known information material to patentability and deal candidly with the USPTO. When an AI output contains a suspicious citation, unsupported assertion, or unexplained result, the circumstances can require a reasonable inquiry before filing.
- Investigate every reference that could affect novelty, obviousness, enablement, written description, or claim interpretation.
- Do not submit an AI-generated quotation, test result, or technical statement until a human verifies it against primary evidence.
- Explain internally why a located reference was disclosed, distinguished, or determined not to be material; preserve that explanation.
- Coordinate the information-disclosure review with the claim chart so materiality decisions are tied to specific limitations.
- Escalate uncertainty rather than relying on a confidence score or a model’s statement that a citation is “verified.”
The Federal Register describes sanctions under 37 CFR 11.18(c) for improper papers. A defensible process is therefore both a patentability safeguard and evidence that the filing team took its inquiry obligations seriously.
What should a registered patent attorney or agent review?
A final human review should be independent of the model and should occur after the claim chart, source audit, prior-art log, and inventorship record are complete. The reviewer should be able to explain each material decision without asking the AI to recreate it.
Quick wins for a faster PC:
Fix the driver behind crashes, sound loss and screen glitchesFind Drivers →Repair Windows errors before they cause bigger problemsFix Now →Scan for outdated or missing drivers - takes under a minuteDriver Scan →| Review area | Self-check can do | Professional review adds |
|---|---|---|
| Claim-limitation coverage | Organizes limitations and flags missing passages | Construes disputed language and decides whether amendments preserve scope |
| Source and date verification | Finds records and records identifiers | Evaluates family, priority, public availability, and prosecution context |
| Prior-art search depth | Runs documented keyword and classification searches | Designs a strategy, evaluates combinations, and makes materiality judgments |
| Written description and enablement | Maps examples and disclosures | Assesses full-scope support and legal risk of broad language |
| Inventorship evidence | Collects contribution records | Applies the legal conception standard and resolves joint inventorship |
| Disclosure and materiality | Maintains a candidate-reference list | Determines what must be disclosed and signs filings with independent accountability |
| Legal accountability | Documents process decisions | Takes professional responsibility for claim scope, papers, and filing formalities |
USPTO Director Kathi Vidal described the system as one designed to protect “human ingenuity and the investments needed to translate that ingenuity into marketable products and solutions.” The practical implication is straightforward: AI may accelerate drafting and searching, but a qualified human must own the legal judgments.
What records should I retain for an auditable file?
Keep a chronological, access-controlled file that another practitioner can review without relying on the model’s memory.
- Original prompts, model and version information, timestamps, and complete outputs.
- All source documents supplied to the model and their provenance.
- Versioned claims showing who made each substantive change.
- Limitation charts with pinpoint specification, drawing, inventor, and prior-art evidence.
- Patent and literature search strings, databases, classification codes, date ranges, dates searched, and result sets.
- Copies or stable records of the official publications and prosecution documents reviewed.
- Inventor-contribution statements and supporting notebooks, design files, or experiment records.
- Reviewer comments, disclosure decisions, unresolved questions, and the final sign-off by counsel.
Protect confidential source material and personal data under the firm’s security and privilege procedures. If an external AI service was used, confirm that its terms and settings did not expose confidential invention information.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.What are the most common verification failures?
Trusting a plausible number
Failure: A publication number or quotation looks authentic, so it is copied into the application. Fix: Require an official record and a passage-level check before the citation leaves the unverified queue.
Checking only the independent claim
Failure: Dependent claims inherit unnoticed defects or add limitations absent from the specification. Fix: Chart every dependency and every added limitation.
Using one reference as a novelty shortcut
Failure: The presence of a citation is treated as proof that the claim is new or old. Fix: Map all limitations and separately analyze obviousness with counsel.
Letting AI broaden an example
Failure: A specific material, range, or architecture becomes a broad genus without supporting disclosure. Fix: test full-scope written description and enablement, then narrow or add substantiated disclosure.
Confusing editing with inventorship
Failure: The person who prompted or polished the text is listed automatically. Fix: document each natural person’s contribution to conception and obtain a practitioner’s inventorship analysis.
What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
Deleting the audit trail
Failure: Prompts, failed searches, and rejected citations disappear after drafting. Fix: preserve immutable snapshots and decision logs from the first output through filing.
Best Value
How do I decide whether the application is ready to file?
File only when the record supports a reasoned human decision, not when the AI reports high confidence.
- Every claim limitation has a reviewed support entry or has been removed or revised.
- Every citation has an authoritative record, verified identity fields, a public-availability analysis, and a passage-level support result.
- Independent prior-art searches and their logs are complete.
- Novelty, obviousness, written description, enablement, definiteness, construction, and eligibility issues have been evaluated for the actual claims.
- Natural-person inventorship is documented and signed off.
- Known material information and any required disclosure decisions are addressed.
- A registered patent attorney or agent has completed the final review of claim scope, inventorship, materiality, and filing formalities.
If any item remains unresolved, preserve the issue in the file and escalate it before submission. The safe default is to treat the affected claim or citation as unverified, not to assume the model was right.
Frequently Asked Questions
What if an AI citation is real but the quoted passage is wrong?
Keep the official document in the file, mark the quotation as unsupported, and locate the passage that actually bears on the proposition. If the correct passage changes materiality or claim strategy, update the claim chart and disclosure analysis before filing.
How long should I keep prompts and search logs?
Keep them with the patent prosecution file for the period your firm uses for invention records, litigation holds, and professional-responsibility requirements. Your practitioner should set the retention period and access controls because confidentiality, privilege, and preservation duties can overlap.
Can I use an AI tool to prepare an information disclosure statement?
It can help organize references, but a human must verify every bibliographic field and decide what information is material. Do not submit an automatically generated list without checking the original records and the statements made about them.
When should prior-art searches be repeated?
Repeat or update searches when claims change materially, when prosecution introduces new limitations, or when a long delay makes the earlier search stale. Record the trigger, databases, date, and differences in the results so the update is auditable.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
The Tool Desk
Outbyte Driver Updater FREEFix the driver behind crashes, sound loss and screen glitchesFind Drivers →Outbyte PC Repair FREEClear out junk files and repair common Windows errorsFree Scan →
