Attached to every examination report is a list of documents. Each one is compressed to a publication number, a date, a name and sometimes a single letter in the margin. Applicants tend to read that list the way a patient reads a lab result — fast, looking for the word that decides everything — and then spend weeks arguing about the wrong document. The list is not a verdict. It is the examiner's evidence, and evidence can be examined back.
The work of answering it breaks into four steps that anyone can do without training: establish where the references came from, understand how each document is built, check the dates, and compare the reference against your claim one limitation at a time. Nearly every wasted response round in first-time prosecution comes from skipping one of those four.
Provenance · 01
Where the List Comes From
An examiner does not begin with keywords. They begin with classification — a hierarchical scheme that divides all of technology into roughly a quarter of a million subdivisions, from broad sections down to individual mechanisms. The examiner reads your claims, decides which two to five subdivisions the invention belongs in, and searches within them. Keyword strings, citation networks and the reference lists of documents already found are used afterwards, to fill gaps.
That order matters, because it tells you why a search can miss. If the core of your invention sits in a subdivision the examiner never opened, the search covered a neighbouring technology rather than yours. The search notes attached to the report list exactly which areas were consulted and which database strings were run, which makes this checkable in about twenty minutes. Pointing out, factually and without accusation, that the invention is classified elsewhere is a legitimate part of a reply and occasionally reframes an entire case.
The letters in the margin carry real meaning in many systems. A document marked as relevant on its own is asserted to disclose the claim by itself. A document marked as relevant only in combination is being offered as one half of a two-reference argument. A document marked as background is not a ground of rejection at all — it is context, and applicants routinely spend a fortnight rebutting one. Sorting the list by those categories before reading a word of description is the cheapest hour available in the whole process.
Anatomy · 02
What a Cited Document Is Made Of
A published patent document has a fixed architecture. The front page carries numbered bibliographic fields: publication number and date, application number and filing date, any earlier priority claims, applicant and inventor names, classification symbols, an abstract and one representative drawing. Those numbered fields are standardised across offices, so once you have learned to read one front page you can read all of them, including documents in languages you do not speak.
The body then runs in a predictable order: field of the invention, background, a summary of the invention, a detailed description keyed to numbered drawing elements, and finally the claims. Here is the single most useful fact about prior art citations: the claims of a cited document are largely irrelevant to the rejection. What counts is the whole disclosure — including the alternatives its author mentioned and abandoned, the admissions buried in its background section, and every feature drawn but never claimed. A reference can defeat your claim on the strength of a sentence its own applicant never sought to protect.
Citations point into that body in one of two formats: column and line numbers for older typeset documents, or bracketed paragraph numbers such as [0042] for modern ones. Read the cited passage, then read fifteen lines either side of it. Context reverses the meaning of a cited sentence more often than one might guess, particularly where the surrounding text explains that the described arrangement was unsatisfactory. Where the document is a machine translation, treat the wording as approximate and the drawings as authoritative; translation artefacts have sunk more than one confident argument.
Roughly one cited document in ten is not a patent at all. Conference papers, product catalogues, service manuals, university theses and archived web pages all qualify, and each carries its own dating problem. A catalogue with only a year printed on it, an archived page with a capture date, a manual revised three times under the same title — all of these can be challenged on the question of what was actually available to the public, and when.
Dates · 03
The Question That Kills a Reference Outright
Every patent document carries at least two dates, and confusing them is the most common error in reading prior art citations. The filing or priority date records when the document's own applicant staked their claim. The publication date records when the public could first read it. Which of the two governs depends on the system and on the ground of rejection being raised.
The consequence catches nearly everyone. Applications are typically published about eighteen months after their earliest priority date. A document published four months before you filed may therefore have been filed almost two years earlier — and in many systems it counts against you from that earlier date, even though no diligent search on the morning you filed could possibly have found it. This is why a clean search result is never a guarantee, and why an unexpected reference is not evidence that anybody was careless.
The reverse case is the best news available in prosecution. If a reference's effective date falls after your own priority date, the reference is simply out of the case — no argument about mechanisms, no amendment, nothing to concede. Check every date on every cited document before reading anything else. Transcription errors in a report happen, priority chains get misread, and a single wrong year has ended rejections that would otherwise have taken two rounds and a narrowed claim to escape.
Check the dates before you read a word of the description. A reference with the wrong date is not an argument you have to win. It is a reference that was never in the case.
First pass, every time
Mapping · 04
Reading the Citation Against Your Claim
The comparison itself should be done on paper, in two columns. Break your claim into its individual limitations — one per row, including every "wherein" clause and every functional phrase such as "configured to" — and in the right-hand column paste the exact sentence from the reference said to disclose it. Not a paraphrase. The sentence.
Empty cells are the argument. Any row where nothing from the reference can honestly be pasted is a limitation the document does not disclose. Against a single-reference ground, one empty row is decisive.
Half-full cells are the trap. Where the reference discloses something similar but structurally different, the difference has to be stated in terms of mechanism, not adjectives. "The reference uses a spring; the claim requires a fluid damper" is checkable. "The reference is fundamentally different" is not.
Full rows are information, not defeat. A reference that genuinely discloses eight of your ten limitations tells you precisely where the boundary of the known art sits — which is exactly the knowledge needed to decide what to claim next.
Two cautions apply while filling that table. The examiner will read your claim terms at their broadest reasonable meaning, so a word you intended narrowly may have been read widely; explaining what a term means is permitted, but every such explanation joins the permanent record and will be read back later by anyone assessing scope. And a combination ground does not fall merely because neither document discloses everything — it falls when the stated reason for putting them together does not hold up, which is a different fight from this one and worth understanding before drafting. The wider picture, including how the four distinct grounds differ and what a reply should contain, is set out in the guide to reading a first rejection.
The Reply · 05
Answering Without Giving Ground
A reply that works quotes rather than characterises. Reproduce the cited lines, then show what the claim requires that those lines do not describe. Weak submissions assert that a reference "fails to teach the inventive concept"; strong ones identify the element, name the drawing numeral it would have had, and observe that it is absent. Examiners read dozens of replies a month and the difference is visible in the first paragraph.
Resist the reflex to volunteer advantages. Explaining at length why your version is better invites a record in which the invention is defined by those advantages rather than by its structure. Resist, too, the urge to narrow immediately: a limitation added to escape one reference stays in the claim for the whole twenty-year life of the patent, and any competitor reading the file will see exactly which door was left open. Where a conversation is available, use it — most offices allow a scheduled interview of twenty to thirty minutes, and a surprising share of disputes turn out to be two people reading one claim word differently.
None of this is specialist knowledge, and it travels across fields. Accounts of inventor-led design work in the built environment describe the same crowded-field problem that a bracket or fastener inventor meets, and the same discipline resolves it. Longer views of the work, such as four decades spent advocating for independent inventors, tend to report that the cases which grant are simply the ones where somebody read the references properly. It is a long way from the version of invention that reaches the screen, where the difficult middle is a montage.
A cited document is only as strong as the sentence it is being asked to carry. Find the sentence. Read what surrounds it. Then say what is missing.
The method in one line
Sort the list by category. Verify every date. Read the disclosure rather than the claims. Build the two-column table and find the empty rows. That is the whole of it, and it is available to anybody with an afternoon and a printer.
End of report