How Claw's Gazette Matching Works: Text, Phonetic and Visual (Benchmarked)

How-to · Trademark Gazette Matching

Updated for 2026 · Claw Editorial

A conflicting mark in the Trade Marks Journal is easy to miss. It may not spell the same as yours, but it can sound the same, or look the same on a shelf, and the time to oppose it is short. This guide explains how gazette conflict matching works (text, phonetic and visual), why accuracy claims mean little without a stated test set, and how you can run your own benchmark. Last verified October 2026.

01Why gazette matching is hard

Gazette matching means checking every new trademark advertised in the Trade Marks Journal against the marks you own or care about, and flagging the ones that could conflict. It is hard because the journal is large, it keeps coming, and a conflict does not need to be an exact copy.

The clock is short

Under Section 21 of the Trade Marks Act, 1999, a notice of opposition must be filed within three months of the advertisement, with a further period of not more than one month if the Registrar allows it. A missed journal issue can mean a lost chance to oppose.

A conflict is more than spelling

Indian courts look at overall impression. Two marks can differ by a letter or two and still be confusing, because people speak a mark, hear a mark, and remember it imperfectly. A plain spelling check misses these. An exact word search finds almost nothing useful.

Many scripts, many accents

Marks are filed in English, in Hindi, and in other Indian scripts, and are said aloud in many accents. The same brand idea can be spelt many ways in Roman letters. Any matching that works only on letters will have gaps.

A matching tool is only as good as the test set it was measured on. An accuracy number without the number of marks tested and the date is not a measurement.

02How the three matching types work

Good gazette monitoring uses three kinds of matching together, because each one catches conflicts the others miss.

Text matching

Text matching compares the letters of the marks. It finds exact matches, marks that contain your mark, and marks with small spelling changes, added letters or swapped letters. It is fast and easy to explain, but it cannot tell that two different spellings sound alike.

Phonetic matching

Phonetic matching compares how marks sound. It reduces each mark to a sound pattern, so that spellings which are said the same way match. In Indian use this must allow for common transliteration choices and accents. It catches the case where a new mark is said the same as yours but is written differently.

Visual matching

Visual matching compares how marks look, whether as styled words or as logos and devices. It catches look-alike lettering, shapes and layouts that text and sound do not capture. It is the hardest of the three to measure, because "looks similar" has to be judged against a human reviewer.

No single method is enough. The question for any tool is how the three are combined, and how well the combination performs on a test set.

03What the law actually asks

The legal test for a conflict is about confusion, not about a score from a program. A tool can only flag candidates. A person decides whether to oppose.

Section 11 of the Trade Marks Act, 1999 sets out the relative grounds for refusing registration, including identity or similarity with an earlier mark for the same or similar goods or services.

Two Supreme Court decisions are often cited on how similarity is judged. In Amritdhara Pharmacy v Satya Deo Gupta (AIR 1963 SC 449), the Court looked at the overall impression and the way an ordinary buyer with imperfect recollection would see the marks. In Cadila Health Care Ltd v Cadila Pharmaceuticals Ltd ((2001) 5 SCC 73), the Court listed factors such as the nature of the marks, the degree of resemblance (visual, phonetic or otherwise), the kind of buyer, and the nature of the goods, and applied a stricter view for medicines.

These cases explain why the three matching types make sense. The Court itself speaks of resemblance by sight and by sound. Primary texts are available on Indian Kanoon and the journal and search tools are on IP India.

04How to build a reproducible test set

A reproducible test set is a fixed list of marks with answers already decided by people, which you can run through any tool and score the same way every time. Here is a simple method.

1. Pick the journal issues. Choose a stated range of Trade Marks Journal issues, and write down the issue numbers and dates. Anyone should be able to fetch the same issues from IP India.
2. Pick the watch marks. Choose a fixed set of marks to protect (your own, or test marks). Include word marks, device marks and marks in more than one script. Write down how many.
3. Label the truth. Have at least two reviewers, ideally IP professionals, mark which advertised entries are real conflicts for each watch mark. Record where they disagree and how you settled it.
4. Keep the answers hidden. Do not show the tool the labels. Run it blind, then compare.
5. Freeze and date it. Save the test set and write the test date. A rerun on a later date or on a newer tool version is a new test, not an update of the old one.

05What to measure and how to report it

Two numbers matter most: precision and recall, and both need N and a date to mean anything.

• Recall: of all the real conflicts in the test set, how many did the tool find? Low recall means missed oppositions. For a watch service this is the number that matters most.
• Precision: of all the marks the tool flagged, how many were real conflicts? Low precision means a pile of false alerts that your team stops reading.
• Conflicting marks per issue: the average number of flagged marks per journal issue for your watch list. This tells you the review workload.
• By match type: report text, phonetic and visual separately, so you can see which one finds what.

A fair report line reads like this: "On N watch marks and M journal issues, tested on a stated date, recall was X and precision was Y." Without N, M and the date, leave the figure out. A ready-made, dated example of this method is in our trademark conflict search benchmark.

This page does not state a precision or recall figure of its own, because a figure belongs next to its test set and date.

06Common mistakes in benchmarks

Most weak benchmarks fail in the same few ways.

• Testing on the easy cases. A set made of exact copies makes every tool look perfect. Include sound-alike and look-alike cases.
• Tiny N. Ten marks tell you very little. State your N and treat small samples with care.
• No date. Tools change. A figure from last year may not hold now.
• Vendor-chosen test sets. A vendor can choose a set it does well on. Use your own marks and your own reviewers where you can.
• Reporting only recall. A tool that flags everything has perfect recall and is useless. Always show precision too.
• Ignoring devices. Many conflicts are logos. A word-only test hides weak visual matching.

If you are screening names for a different job, such as checking borrowers before a loan, see pre-sanction borrower litigation screening. Matching on person and company names is a related but separate problem. For tracking High Court hearings of IP disputes, see the Bombay High Court cause list tracker.

07Where Claw fits

Claw is an all-in-one legaltech platform for Indian advocates, law firms, and corporate legal teams, combining AI-based case search, an AI legal assistant (Legal GPT), case management, and compliance automation across all Indian courts and tribunals.

In its IPR Solutions tab, Claw tracks IP India registry matters for trademarks, patents, designs and geographical indications. For trademarks, it monitors the Trade Marks Journal for conflicting marks using text, phonetic and visual matching, and alerts the user. It also tracks pending compliances such as trademark renewals and patent annuity and response deadlines, pending hearings from trademark and patent filings, and the related documents, tasks and activities.

On the court side, Claw tracks IP litigation across Indian courts that have a case website, including party-name infringement discovery and the High Court IP Divisions. Delhi IPD matters can be followed by case-type code (for example C.O.(COMM.IPD-TM)) and by court plus number or party, along with IPD hearings, orders and cause lists. IP judgment research runs over 30 crore+ judgments with LegalGPT. Scale is 9,102 courts, benches and forums, with results in under 5 seconds. Supreme Court coverage starts from 1950 and High Court coverage from 1990.

So Claw covers both the registry side and the court side on one platform, which is its main difference for India. Judge this yourself: run the test-set method above on your own marks.

Honest limits: Claw does not yet track copyright registration at the Copyright Office. For copyright it is court-side only. For global multi-jurisdiction docketing and annuities, Anaqua, Clarivate and Dennemeyer are built for that job. For very high-volume docketing in specialist IP firms, Iolite and JSK are options. For USPTO or CIPO-first work, Alt Legal fits. For AI prior-art and structure search, look at PatSnap, Questel and IPRally. For mass marketplace takedowns, Red Points and Corsearch are built for it. For technology-transfer office lifecycle work, see Wellspring and Inteum.

Please confirm current plan details on clawlaw.in.

08Sources and further reading

Primary sources for the law, the journal and the cases, last verified October 2026:

• IP India (Trade Marks Journal, forms, Trade Marks Rules): ipindia.gov.in
• Public trademark search: tmrsearch.ipindia.gov.in
• Indian Kanoon (Amritdhara and Cadila judgments): indiankanoon.org
• Claw: clawlaw.in

This guide is general information, not legal advice. Check statute sections, rules and forms against the primary source before filing.

09Frequently asked questions