How our AI works

• Your idea is sent to our AI provider. We send the title, industry, problem and description to Groq, Inc.. Not your name, email address or phone number.
• The AI scores five dimensions. Novelty, commercial potential, defensibility, licensing probability and timing, each from 0 to 100 with a written reason.
• A fixed rule turns the scores into a verdict. The AI does not choose the verdict. See the rule below.
• The AI writes your report. A second request produces the eight sections of the report from the scores and your description.
• We compare it with earlier submissions, inside our own database. This returns only counts. No text is sent anywhere for it, and no one else's idea is shown to you.
• We create your report and certificate as PDFs, store them privately, and email them to you. This usually takes under five minutes.

The verdict comes from the scores by a published rule, not from the AI's opinion:

• Do not patent if novelty is below 30, or defensibility is below 20.
• Proceed now if the average score is 65 or more and novelty is 50 or more.
• Refine first in every other case.

• It assesses only what you write. A thin description gets a thin assessment.
• It does not search patent databases. Any similar patents it names come from the model's general knowledge, are unverified, and may not exist. Treat them as leads to check.
• It can be wrong, incomplete or inconsistent. Two similar ideas may get different results.
• A high score does not mean your invention can be patented, and a low score does not mean it can't. Only a registered patent attorney, working from a real prior-art search, can tell you that.

The free first step sends nothing to anyone: what you type there stays in your browser. Full details are in our Privacy Policy and our confidentiality commitment.

Email human@aipatentregister.com with any question about the assessment or your data.