A patent citation points from one document to another, and the direction decides what it tells you.
Backward citations are the prior art cited against a patent during examination — what came before.
Forward citations are later patents citing it — what came after, accumulating for years.
Forward counts get used as a value proxy, and the signal is real, weak, and confounded in at least three ways that most analyses ignore.
Reading them well
| Practice | Detail |
|---|---|
| Separate examiner from applicant citations | Different weight |
| Note the grant year | Age drives the count |
| Check in-force status | Filing is not conviction |
| Read the claims | The count does not measure scope |
Where citations come from
| Source | Share |
|---|---|
| Applicant IDS submissions | Frequently the majority |
| Examiner search | The independent signal |
| Third-party submissions | Occasional |
| International search reports | Where a PCT exists |
In one line
Forward citations are a weak, confounded signal — useful for screening, useless as a valuation.
The two directions
| Backward | Forward | |
|---|---|---|
| Points to | Earlier documents | Later documents |
| Fixed at | Grant, mostly | Never — keeps growing |
| Tells you | What the invention built on | What built on the invention |
| Used for | Understanding scope, prosecution history | Value proxy, competitive signals |
| Who creates them | Applicant and examiner | Other applicants and examiners |
Backward citations are largely settled at grant. Forward citations accumulate for the life of the document and beyond.
Citation data is public
| Source | Cost |
|---|---|
| Google Patents "Cited by" | Free |
| USPTO Patent Public Search | Free |
| Espacenet | Free |
| PatentsView bulk data | Free |
Paid tools add normalisation and alerting, not access.
Who added the citation matters
| Source | Meaning |
|---|---|
| Applicant-cited | Art the applicant knew and disclosed under the duty of candour |
| Examiner-added | Art the examiner found and considered relevant |
| Examiner-added is generally weighted higher | Independent judgement of relevance |
| Self-citation | Same owner citing its own earlier work |
Applicants cite defensively. The duty of candour requires disclosing material art you know about, so citation lists include things that are barely relevant, submitted to avoid an inequitable conduct problem.
Examiner citations reflect a search and a judgement, which is why analyses that separate them find a stronger signal.
Self-citations indicate a continuing programme rather than external validation, and most serious analyses strip them out.
Normalising properly
| Step | Method |
|---|---|
| By age | Citations per year since grant |
| By field | Compare within the same CPC area |
| Remove self-citations | Same assignee |
| Separate examiner-added | Where data allows |
| Compare within cohorts | Same grant year |
What confounds the count
| Confound | Effect |
|---|---|
| Patent age | Older patents have had longer to be cited |
| Technology field | High-volume fields cite more, mechanically |
| Applicant citation practice | Some firms cite exhaustively, some minimally |
| Examiner workload and art unit | Varies |
| Self-citation | Inflates counts for large filers |
| Continuation families | Members cite each other |
Age is the biggest one. A 2012 patent and a 2023 patent cannot be compared on raw forward citations, because one has had eleven extra years to accumulate them.
Normalise by cohort and field or the comparison means nothing. That is the step most often skipped.
Use it for what it does well
| Good use | Poor use |
|---|---|
| Screening a large set | Valuing a single patent |
| Finding who works nearby | Concluding infringement |
| Mapping a field | Replacing claim analysis |
| Spotting foundational work | Comparing across fields raw |
What the metric is actually worth
| Claim | Assessment |
|---|---|
| Forward citations correlate with value | Weakly, in the literature |
| High counts mean a valuable patent | Not reliably |
| Zero counts mean a worthless patent | No |
| It is a useful screening signal | Yes, with normalisation |
| It replaces reading the claims | No |
Treat it as one input among several. A valuation resting heavily on citation counts is resting on an age-dependent, field-dependent metric partly generated by examiners.
What actually drives patent value is whether the claims read on products people sell, which citation counts do not measure. See intellectual property valuation.
Backward citations tell you something too
| Reading the backward citations | Reveals |
|---|---|
| What the examiner considered | The art the claims were allowed over |
| How crowded the field was | Density |
| Which companies appear | Who else works there |
| Non-patent literature cited | Where the field's knowledge lives |
| Foreign art cited | International activity |
The examiner's citations are a free prior art search already done by someone with access to the classification system.
They are the starting point for an invalidity search, because the strongest references are frequently what the examiner did not find rather than what they did.
Where to see them, free
| Tool | Shows |
|---|---|
| Google Patents | Forward and backward citations, "Cited by" |
| USPTO Patent Public Search | US citation data |
| Espacenet | Citations across jurisdictions |
| Patent front page | Backward citations as printed |
| USPTO bulk data / PatentsView | Programmatic analysis |
All free. Paid tools add normalisation, visualisation and alerting rather than access to data that is public.
Citation tracking
| Watch | Signal |
|---|---|
| New citations to your own patents | Someone is working nearby |
| Who is citing | Which companies operate in your space |
| Citation clusters | An area heating up |
| Citations from a specific competitor | They have read your patent |
| Citations in continuations | Family activity |
A company citing your patent has read it. That tells you they know about your rights and work in the area — which is a lead, not evidence.
Set an alert on your own patents. It costs nothing and it surfaces the parties most likely to matter to you. See patent monitoring.
Citation counts by patent age
| Grant year | Typical accumulation |
|---|---|
| 15 years ago | Fully accumulated |
| 8 years ago | Substantially accumulated |
| 3 years ago | Early — counts still rising |
| Last year | Near zero, regardless of quality |
A recently granted patent will always look weak on raw counts. That is arithmetic, not an assessment of the invention.
Compare within cohorts or normalise per year, and treat anything younger than about three years as uninformative on this metric.
Citations are not an infringement finding
| Citation tells you | It does not tell you |
|---|---|
| They read your patent | Whether they infringe |
| They work in the area | What their product does |
| They knew about your rights | Whether claims are met |
| Where to look | Anything about claim scope |
Infringement requires every element of a claim to be present in a product. A citation is a document referencing a document.
But it is a good starting list. Companies citing your patent are worth checking first in an infringement search. See patent infringement search.
Worked example: two patents, same field
| Patent A | Patent B | |
|---|---|---|
| Granted | 2012 | 2022 |
| Raw forward citations | 47 | 6 |
| Age-normalised, per year | ~3.6 | ~3.0 |
| Self-citations | 19 | 1 |
| Excluding self-citations | 28 | 5 |
| Examiner-added share | 40% | 55% |
| Claims read on shipping products | No | Yes, two |
What the numbers said versus what mattered
| Reading | Conclusion |
|---|---|
| Raw count | A looks 8× stronger |
| Age-normalised | Roughly comparable |
| Excluding self-citations | Gap narrows further |
| Examiner share | B slightly stronger |
| Evidence of use | B is the valuable asset |
Patent A's count was inflated by age and by nineteen self-citations from its own owner's continuing filings.
Patent B is worth more because its claims read on products people sell, which is the factor citation counts do not capture at all.
What a high count does not mean
| Not implied | Why |
|---|---|
| The patent is valid | Citations are not a validity finding |
| The claims are broad | Counts do not measure scope |
| It is enforceable | Different question entirely |
| Anyone infringes it | The value driver, unmeasured |
| It is worth buying | Depends on the claims |
A heavily cited patent with narrow claims nobody practises is still worth little.
Citing prior art in your own application
| Practice | Why |
|---|---|
| Disclose material art you know | Duty of candour |
| Supplement as you find more | Including foreign counterpart citations |
| Do not withhold to shorten the list | Can render the patent unenforceable |
| You are not obliged to search | The duty is to disclose what you know |
| Cumulative art | May not need repeating |
Failure to disclose material prior art can render a granted patent unenforceable, which is far worse than any rejection. See patent application form.
Using citations in a portfolio review
| Signal | Reading |
|---|---|
| High citations, still in force | Foundational, maintained |
| High citations, lapsed | Field moved on, or owner did |
| Low citations, recent | Uninformative |
| Low citations, old | Possibly peripheral |
| Cited by a specific competitor | Worth investigating |
Pair citations with maintenance behaviour. Filing measures belief; maintaining measures conviction, and the two together say more than either alone.
In landscape analysis
| Use | Detail |
|---|---|
| Citation networks | What a field builds on |
| Highly cited nodes | Foundational work |
| Citation clusters | Sub-areas |
| Cross-field citations | Technology transfer |
| Combined with survival data | Belief versus conviction |
Pair citation counts with in-force status. A heavily cited patent that lapsed tells a different story from one still maintained. See patent landscape analysis.
Patent citations: the checklist
- Backward is prior art; forward is what came after.
- Normalise by age. Older patents accumulate more, mechanically.
- Normalise by field. High-volume fields cite more.
- Strip out self-citations. They indicate a programme, not validation.
- Weight examiner-added citations higher than applicant-cited ones.
- Treat the count as one input, never as a valuation.
- Read the claims. That is what actually determines value.
- Set free alerts on your own patents. Google Patents and Espacenet both do this.
- Use citations as infringement leads, not as evidence.
- Disclose material prior art in your own filings. The duty of candour is unforgiving.