How we rank notable patents

The two signals

Renewal depth, and why it is not a third score

Maintenance fees fall due at 3.5, 7.5 and 11.5 years from grant, each with six months to pay, and they rise every time. Paying is the company's own repeated decision that a patent is worth keeping, so we read it, store it and state it on every card. We do not average it into the ranking, because it barely separates anything: of the patents in force with at least one window closed, 99.7% have paid every fee due. That is close to a definition rather than a discovery, since not paying is what takes a patent out of force in the first place. Averaging a number that is the same for almost every patent would not rank them, it would just add the same amount to most of them, while penalising design and plant patents, which carry no maintenance fees at all, for a fee record they were never going to have.

Age is the whole problem

Both of those grow with age. A patent granted in 2005 has had twenty years to be cited and to build a family. A patent granted last year has had none of that. So citations and family size are turned into a percentile against patents granted the same year, and nothing is ever compared across years. Without that rule, notable would just mean old, and every company's flagship would be its oldest surviving patent.

The comparison is also kept within a patent's own kind, for the same reason. Design patents are cited far more often than utility patents, because an examiner cites a wall of design prior art on every design application: of the patents we hold granted in 2015, 95% of designs have been cited against 69% of utilities, and the median design has 13 citations against the median utility patent's 2. Pooled together, a company's bottle designs would outrank its formulations on a quirk of how design examination works. Design patents are compared with design patents.

Citations only count once a grant year has had time to accumulate them. Of the patents granted in 2024, 83 in every 100 have never been cited, so the top 5% of that year is a patent cited twice, which is not a finding. A year's citations enter the ranking only once at least half its patents have been cited at all, which takes about seven years. Before that the count is still shown on the patent's own page. We simply do not rank on it.

Putting them together

The two percentiles are averaged with equal weight. Any other weighting would be a claim about how much more one kind of evidence counts than another, and we do not have a basis for that claim.

What we do when a signal is missing

A signal we do not have is left out rather than counted as zero. A patent granted after our citation data ends has no citation count, which is not the same as never being cited. A patent whose grant year is too recent for citations to mean anything keeps its count on its own page and is not ranked on it. A patent missing either signal is not ranked at all, because one number is not a ranking, and that is why the notable list on a company page holds only patents old enough for the public record to say something about them.

One exception, bounded and disclosed on the card itself: a patent granted too recently for a citation percentile to exist yet may still fill a slot the ranking otherwise left empty, chosen on family size and how recently it was granted instead. At most one per company, and only ever a backfill: it never displaces a patent the ranking can actually score. Its card says outright that it is too new for citation comparison.

The ranking decides which three patents are chosen. The order they are shown in is a separate rule: strongest citation standing first, family size to break a tie. A patent chosen by the recent lane above has no citation standing to show and is always listed last.

Which patents are eligible

• Only patents in force today are eligible. An expired patent can be important and is still shown everywhere else on the site; it is not a current holding.
• One patent per family. Without this rule the list fills with siblings of a single invention, which is one finding printed three times.
• Terminal disclaimers, term extensions and confidence tiers change what a patent's dates are, and they are shown wherever those dates appear. They do not move a patent up or down this list.

What this ranking does not tell you

• Citations count US patents only. Most come from one granted patent citing another; some are recorded earlier, against a patent's own pre-grant publication, and are counted the same way once we can tell which patent the publication became. Citations from an application that never grants, from foreign patent offices, and from non-patent literature are real and are not counted here, because they live in different sources we have not ingested.
• A citation is not an endorsement. Examiners add most of them, and a patent can be cited often because it blocks a crowded field rather than because it is admired.
• The comparison population is the patents in our own database granted that year, not every US patent granted that year. Where a grant year holds too few patents for a percentile to mean anything, no band is shown.
• This ranking says which patents look most significant on the public record. It is not a valuation, not a validity opinion, and not advice.