What we learned filing 200 trademarks with LEA

We crossed 200 trademark applications filed through LEA last week. That's a small number in the grand scheme of the USPTO, but it's enough volume to start seeing patterns, and enough to be humbled by a few of them.

Here's the honest field report.

The pattern we didn't expect: most "trademark" requests aren't

Roughly a third of incoming requests turn out to be something other than a trademark filing. They're:

• Brand-name vetting before a product launch (no filing intended, just a go/no-go).
• Trade-dress questions disguised as word-mark questions.
• Copyright or design-patent questions that ended up in the trademark queue because the user wasn't sure.

LEA's first job, before drafting anything, is to figure out which of those four worlds the user is actually in. We added an explicit triage skill after the first 30 matters made it obvious. Skip the triage and you ship a beautifully drafted filing for the wrong type of right.

The mistake we kept catching: weak goods/services descriptions

Founders love describing their product the way they'd describe it on a landing page. The USPTO does not.

"AI-powered productivity platform for modern teams."

That is a death sentence for an application. The examiner will issue an office action, the timeline slips by four months, and the founder blames the lawyer.

LEA now rewrites every goods/services description into the ID Manual's preferred phrasing before the application is even shown to counsel. Boring, mechanical, and easily the single highest-ROI thing the system does.

The highest-value skill we ship: famous-mark screening

If you've ever wondered why trademark lawyers seem unreasonably cautious, this is why: filing a mark that conflicts with a famous mark isn't just rejected, it can attract a cease-and-desist before the ink is dry.

We built a dedicated famous-mark screen that runs before a clearance search, not after. It catches:

• Phonetic equivalents ("Kandy" vs. "Candy" in confectionery).
• Foreign-language equivalents (you'd be surprised).
• Marks that are famous in a single class but the founder is filing in an adjacent one.

About 1 in 12 applications get redirected at this step. Every single one of those is a future lawsuit that didn't happen.

What we got wrong

Three things, in order of embarrassment:

1. We over-trusted the model on Nice classifications. Class 9 vs. Class 42 is not a vibes-based decision. We now require a human checkpoint on every multi-class filing.
2. We under-communicated office-action timelines. Founders assumed "filed" meant "done." We now send an explicit 6-month-out and 3-month-out reminder.
3. We let users edit the specimen. Don't. The specimen is evidence; it isn't a draft. We locked that flow down after one near-miss.

What's next

We're working on the harder version of all of this: filings outside the US, where the procedural surface area is much wider and the famous-mark databases are far less complete. Expect a follow-up post when we've earned the right to write it.

If you're filing more than a handful of marks a year and any of the above sounds familiar, get in touch. We learn faster when we have your weird edge cases to chew on.