The 12-word head start
How asserted patents were drafted differently

A patent claim is a list of requirements, and a product infringes only if it meets every one. Applicants mostly overcome rejections by adding requirements, so counting the words claim 1 gained between filing and grant is a rough but honest measure of how much scope a patent gave up in prosecution. It is imperfect: one well-chosen word can narrow a claim more than a 40-word clause. But it can be measured on 7,836 patents without relying on anyone's judgment.

We built two sets of patents. The asserted set is every patent from our litigation benchmark (9,873 patents asserted in U.S. district court from 2020 through 2025) that is not a continuation: 3,949 patents. The control set is one never-asserted patent per asserted patent, drawn at random from the same art unit and the same grant year: 3,887 patents. Continuations are excluded because their claims arrive already shaped by a parent's examination, so their filing-day text is not a clean baseline. That decision matters: continuations are 61% of the asserted benchmark and 30% of controls, and our March report covers them.

For each patent we took claim 1 from the claims document in the USPTO file wrapper, the text as it was filed, and claim 1 from the granted patent. Where the file wrapper is image-only (most filings before 2009) we fell back to the pre-grant publication, which agrees with the as-filed text within two words on three quarters of patents. We took the examination record from the USPTO transaction history: every non-final and final rejection, RCE, notice of appeal, examiner interview, and Track One request, with dates.

On filing day, before any examiner had seen them, asserted patents already had shorter first claims: 103 words for the typical asserted patent, against 116 for the typical control. The broadest independent claim shows the same gap (81 against 93 words on the pre-grant publication). The difference holds in eight of the nine art-unit groups we can segment, and is largest in mechanical and medical devices (106 against 131).

This is the most reliable finding in the report. It holds whichever way we cut the data: by art unit, by grant year, by the size of the filer, by whether the application came in through the international route, and by grant type. It is still there when we compare only patents handled by the same examiner, where the asserted patent's claim 1 was about 8% shorter. We did not find the gap in three groups: B1 grants (patents granted without a pre-grant publication, a small group), patents granted from 2021 onward (the ones asserted within a year or two of grant), and art unit 36xx, business methods.

On every measure we have, asserted patents went through more examination.

The typical patent in each group took the same time to grant, about 41 months, but the shape is different. Asserted patents reached a first office action a little sooner (26 against 27 months, mostly the Track One cases) and then took longer from first action to grant (14 against 13 months; 41 against 35 among patents with four or more actions). One asserted patent in ten took more than 81 months to grant; for controls that line is 71 months. A larger share of asserted patents spent many years in examination.

The Track One and interview differences mostly come from patents asserted by operating companies. Among patents asserted by operating companies, 9% used Track One and 32% to 48% had an interview, depending on era. Among patents asserted by non-practicing entities, 2% used Track One, the same as controls, and interview rates match controls too. Operating companies use Track One and interviews more on the patents they later assert. We did not see those differences in the NPE group.

Overall, only slightly. Claim 1 grew by 45 words in asserted patents and 50 in controls, and because asserted claims started shorter, the proportional growth is the same: 44% against 45%. We expected a bigger difference here.

Comparing patents with the same starting length and the same number of office actions, asserted claims gained 11 to 16 fewer words. Two things hide that in the overall numbers:

• Asserted claims started shorter. A shorter claim has more room to grow, so it tends to gain more words in prosecution.
• Asserted patents went through more office actions. Every action adds words.

Asserted patents gave up fewer words per office action. They went through more office actions, from a shorter start, so the total came out about the same.

The figure below shows the per-round difference. At the same number of office actions, asserted patents added fewer words in every bucket, and the gap widens the longer prosecution runs. After one action, asserted patents added 35 words and controls 40. After three, 60 against 89. After four or more, 87 against 116.

Once you compare patents that started at the same length, went the same number of rounds, and sit in the same art unit and era, asserted patents added 11 to 16 fewer words. Nearly all of that comes as about 7 fewer words per office action, so it only becomes visible in patents that went through a long examination. One caution about the figure: the words are counted once, between filing and grant, so "per office action" is an average across the whole prosecution, not a measurement at each round. A patent that added 90 words across three actions may have conceded all 90 in the last response.

The bigger problem is the claims that gained hundreds of words. Among claims that started under 300 words, 9% of asserted patents were pushed past 300 by grant, against 12% of controls. At 400 words it is 3% against 5%. Controls end up past the 300-word line, the length our July audit flagged as the top decile of enforced claims, 40% more often, and most of that excess was pushed there in prosecution rather than filed there. A claim that went from 115 words to 320 may no longer cover the products its owner cares about, and nothing in the record prompts anyone to check.

Asserted patents also have more claims and lose fewer of them during examination. At grant, asserted patents average 21.7 claims against 17.6; 38% have more than 20 claims against 24%. Between publication and grant, asserted patents lost at least one claim less often (50% against 54%) and gained claims more often (21% against 14%).

Small companies and individuals (small and micro entities at the USPTO) filed a third of the asserted patents but only a fifth of the controls. When we split both groups by the size of the filer, the findings above held for large companies and not for small ones.

Among large-entity filers, asserted patents were drafted 17 words shorter and conceded 12 fewer words. Among small entities, asserted patents were drafted at the same length (a 4-word difference, within noise) and added more words in prosecution.

Here, entity status means the filer's status as recorded at the USPTO when the application was filed. It describes the original filer, whoever later asserted the patent.

A third of asserted patents were filed by small companies, universities, or individuals, against a fifth of the matched controls. That is true whoever is suing: small and micro filers account for 35% of the patents asserted by operating companies (many of which are small companies asserting their own patents) and 26% of the patents asserted by non-practicing entities. Small filers reach litigation two ways: by suing on their own patents and by selling them to others who sue. Our March report found that more than half of asserted patents were acquired and that NPEs buy heavily from private-origin filers, which accounts for the second route. Those small-entity patents were drafted and prosecuted like everyone else's. Our guess is that they were asserted because the claim covers a product, not because of how they were drafted or prosecuted.

The drafting and prosecution patterns in this report are large-company patterns. For small-entity patents, the March report's findings on families and acquisition are the ones that describe how they get asserted.

Each finding below answers a belief we have heard from practitioners or held ourselves.

Asserted patents have the same RCE rate as controls. 91% of asserted patents and 93% of controls had at most one RCE after a final rejection. 31% against 29% had at least one. Counting every RCE, including the ones filed after allowance to submit references, it is 36% against 31%. RCE count is the same in both groups.

An examiner interview shortens the path to allowance and costs claim scope. After an interview, 63% of cases were allowed with no further office action, against 51% of cases that answered their first action without one, and the interviewed cases averaged 0.7 further actions against 1.1. That is what an interview is for: the applicant and examiner agree on an amendment that will be allowed, and the next paper is a notice of allowance. The amendment is the cost. At the same office-action count, interviewed cases added more words to claim 1: 67 against 43 for asserted patents, 72 against 52 for controls. Part of that is selection, since the cases that get an interview are the contested ones, but the direction is the same in both groups. Asserted patents were interviewed 4 to 5 points more often than controls within each era, and that is entirely an operating-company practice.

A first-action allowance usually means the claim was already narrow. First-action allowances are rarer among asserted patents (14% against 18%), and they look the same in both groups: a long claim (143 against 163 words on filing day) that added nothing because there was little left to reject. A claim allowed on the first pass was often already narrow when filed.

Asserted patents took the same time to grant. Same median pendency, 42 against 41 months. More asserted patents took a very long time.

Track One marks the patents that mattered to an operating company. It is an operating-company practice: 9% of operating-company-asserted patents, 2% of NPE-asserted patents, 2% of controls. NPE-asserted patents use it at the same rate as controls.

Appeals add fewer words per office action, in both groups. Appealed patents add fewer words per office action than other patents, in both groups. But appealed asserted patents grew as much as appealed controls (63 against 62 words at four actions), from a shorter start, with longer pendency (84 against 70 months). Appeals do not distinguish asserted patents from controls.

The drafting gap is bigger than the prosecution gap. The prosecution gap is 11 to 16 words comparing like with like, and 5 before that. The drafting gap is 12 words with no adjustment needed.

1. Focus on claim length before you file. Most of the advantage was there on filing day, and prosecution roughly preserved it. The claim that goes in sets what comes out: prosecution will add about 45 words over the next 40 months whatever you start with, so a claim 1 that is 20 words past the asserted median for its art unit on filing day will still be 20 words past it at grant. Filing day is the cheapest place to fix that, and the only one where the fix is entirely in your hands.

2. Fight for claim scope, but not for long. The patents that end up in court took more rejections and more appeals than the ones that did not, and gave up fewer words per round: 60 against 89 after exactly three office actions, 71 against 100 after exactly four. At the fifth the advantage was gone, with both groups at about 96 words added, and 90% of both groups were done within four. A first-action allowance usually means the claim was already narrow, and a fifth office action usually means the claim has already been narrowed.

3. Check whether these findings apply to your portfolio. They describe the assertable patents of large operating companies, which started with shorter claims and added fewer words despite more office actions. A small entity, or a portfolio built by acquisition, should look at the March report's findings on families and acquisition instead.

Population. Asserted set: every patent in the ArcPrime litigation benchmark (9,873 U.S. patents asserted in district court 2020–2025, from public court records) that resolves to a US B1/B2 grant in Google Patents Public Data and is not a continuation, continuation-in-part, or divisional of an earlier U.S. non-provisional application: 3,949 patents. Control set: for each asserted patent, one never-asserted US B1/B2 grant drawn by deterministic hash from the same art unit and grant year (3,385 patents), backfilled from the same art unit across the asserted years where the cell was short (502), with 62 asserted patents unmatched: 3,887 controls. Art unit is missing from the public data for 908 patents (454 per group); those matched on grant year alone. Entity status was not a matching variable and is imbalanced (Chapter 5).

Claim text. Granted claim 1 from Google Patents Public Data (BigQuery), parsed with the same claim splitter as the July audit. As-filed claim 1 from the claims document (CLM) on the filing date in the USPTO Open Data Portal file wrapper, delivered as WIPO ST.96 XML with OCR text, for 5,341 patents; 2,042 filings (mostly 2001–2008) have image-only claims documents and 435 (mostly pre-2001) are absent from the portal. Filings whose filing-date claims document is a preliminary amendment (636, mostly national-stage entries) are excluded from the as-filed measure. The pre-grant publication (A1) claim 1 is used as a sensitivity baseline for all 6,617 patents that have one: it agrees with the as-filed text within two words on 76.5% of the 4,176 patents with both, and is more than 10 words longer on 9.3% (preliminary amendments). Every group comparison has the same sign on both baselines except art unit 16xx, where neither is significant. Claim 1 means the lowest-numbered live claim at each stage.

Examination record. USPTO transaction history (Patent Center / PAIR) for 7,827 of 7,836 patents. Office actions = CTNF + CTFR. RCE after final = RCEX whose most recent disposition was CTFR; RCE after allowance = RCEX following MN/=. with no intervening rejection. Interviews = EXIA, EXAT, EXET (no interview codes exist in the record for any grant before 2011, in either group, so interview comparisons are restricted to grants from 2011). Appeals = N/AP (notice filed, not outcome). Track One = TK1R request or T1GR grant. The after-interview comparison counts CTNF/CTFR events dated after the first interview that follows the first office action (n = 1,798) against CTNF/CTFR events after the first office action for cases with no interview (n = 2,920), grants from 2011 with at least one office action. A second copy of the transaction history, pulled from the USPTO Open Data Portal, agrees on the office-action count for 99.7% of the 5,334 patents both cover. The compact event list is stored per patent.

Statistics. Medians with Mann-Whitney tests for group differences; shares with Fisher tests. The adjusted concession gap is ordinary least squares of words added on group, filing-day words, office-action count, art-unit group, and grant year, with a 300-draw bootstrap interval (−15.7, 95% CI −21.1 to −9.7; n = 4,694). Two independent re-analyses reproduced every headline figure and produced the examiner-fixed-effects null and the entity-status split reported above.

Limits. Word counts measure text, not legal scope. The as-filed sample skews to filings from 2009 onward; year adjustment moves the filing-day gap from 12% to 11%. Assertion is selected on many things this study does not observe, including whether a product practices the claim; nothing here shows that drafting short or conceding less causes assertion, only that the patents that were asserted look this way.