I built an AI tool that parsed examiner's reports and generated CIPO and USPTO-compliant trademark amendments automatically.
When a company files a trademark application, the government (CIPO in Canada, USPTO in the US) reviews it and sometimes sends back an examiner's report with a list of objections. The mark might be too descriptive, too similar to an existing one, or the goods description might be too vague.
Someone then has to read that report, identify the objection type, and draft a formal response with the right legal language, including an amended goods description that shows exactly what changed using strikethrough for removed text and underline for added text. That is what this tool automated.
Pick a sample examiner's report, choose your jurisdiction, and see the amendment get generated.
pick a sample on the left, then click generate.
At Smart & Biggar, every flagged trademark application triggered a manual process. Someone had to read the examiner's report, identify each objection, and draft a full amendment response from scratch. That took 2 to 6 hours per report across 127+ filings.
The objection types were almost always the same. The response format never changed. It had automation written all over it.
"Every single amendment follows the same template. We're just filling in different boxes each time."
I built an internal tool that takes an examiner's report and outputs a fully drafted, tracked-changes amendment in under 20 minutes. It reads the report, classifies the objection type, maps it to the right response template, and formats everything to CIPO or USPTO standards.
For repeat filers, the tool also pulls prior approved goods descriptions and reuses them, keeping language consistent across filings.
Templates over generation. Having AI write responses from scratch introduced too much variability. Legal language has to be precise. Mapping objection types to pre-approved templates and using AI only for classification kept accuracy high.
Jurisdiction awareness from day one. CIPO and USPTO have different section references, formal language, and formatting requirements. Building that in meant one tool could handle both.
Prior filings as a source of truth. Repeat clients file multiple marks over years. The tool indexes prior approved descriptions so new amendments stay consistent and avoid introducing language that could create issues down the line.
The goal was not to replace the lawyer. It was to give them a first draft that was already 90% there, so they could spend their time on judgment, not formatting.
The tool became part of the standard workflow right away. What used to take a full afternoon now took under 20 minutes. 100% CIPO adherence across all 127+ filings managed during the term, with not a single amendment returned for formatting reasons.
Repeatability is the signal. If something follows a pattern more than 80% of the time, it can probably be automated. The trick is noticing the pattern, which only happens if you do the work manually first.
Precision beats intelligence. In legal contexts, a creative AI response is a liability. Structured templates plus smart routing beat open-ended generation every time.
Non-technical adoption is the real finish line. A tool only a developer can use is not a product. It is a proof of concept.
Want to see the full implementation or talk through the approach?