Claw vs IP India Public Search: a Trademark Conflict-Search Benchmark
How to test whether a trademark search tool really finds the sounds-alike marks that matter in India, a reproducible test you can run yourself, and where a registry search and an AI tool each fall short.
Explainer · Trademark Search Benchmark
A trademark clearance search fails in a quiet way: the tool shows you no conflict, you file, and then an opposition or an infringement notice arrives from a mark that sounded almost the same. In India this is common, because marks are judged by how they sound and look, not only by how they are spelt. This page explains how to measure phonetic search properly, gives a test you can repeat, and says plainly what each kind of tool can and cannot do. Last verified October 2026.
- The right metric: phonetic recall@50 on a fixed set of Journal conflicts, always stated with N and the run date.
- The law points to sound: the Supreme Court in Cadila Health Care v Cadila Pharmaceuticals, (2001) 5 SCC 73, treated phonetic similarity as a key factor.
- Always search the primary record: IP India public search should be part of every clearance, whatever else you use.
- No tool replaces a lawyer: a search shows candidates, and the Registrar or a court decides similarity.
- Last verified: October 2026.
01Why spelling-based search fails in India
A search that matches letters will miss a mark that sounds the same but is spelt differently. In India that gap is wide, and it costs real money.
One sound, many spellings
Brand names are written in English letters but spoken in many languages. A name can be spelt with “ph” or “f”, with a double vowel, with a silent letter, or with a Hindi sound written in Roman script. To a customer in a shop or on a phone call, these are the same mark. To an exact-match search box, they are different words.
The registry search is a starting point, not a verdict
IP India offers a public search for trademarks, linked from the IP India website. It is free and it is the primary record, so every serious search should include it. Its search options are mostly built around fields such as word mark, class and application number. How well it handles sounds-alike and look-alike marks should be tested, not assumed.
The cost of a miss
A missed conflict shows up late. It can be an opposition after the mark is published in the Trade Marks Journal, a refusal at examination, or a suit in a High Court. Each of these costs time, filing fees and legal fees, and a rebrand costs more.
A trademark search is only as good as the sounds-alike marks it finds, so the right question is not how fast it is, but how many real conflicts it misses.
02What the law actually tests
Indian courts look at overall impression, and sound is a central part of it. This is why phonetic recall is the right thing to measure.
In Cadila Health Care Ltd v Cadila Pharmaceuticals Ltd, (2001) 5 SCC 73, decided by the Supreme Court on 26 March 2001, the Court treated phonetic similarity as an important factor in deceptive similarity, because many buyers go by how a name sounds. The case is on Indian Kanoon and should be read in full before you rely on it.
The Trade Marks Act, 1999 also sets out when a mark can be refused because it is identical or similar to an earlier mark. Check the exact section text on the IP India legal documents page.
For readers new to Journal watching, the related guide on Journal conflicts per class for D2C brands in India shows how conflicts cluster by class.
03What a fair benchmark looks like
A fair benchmark uses the same test set, the same rules and a written date, so anyone can repeat it and get a similar answer.
- A fixed test set: a list of real marks and, for each, the later marks that a trademark lawyer has already judged to conflict by sound.
- A fixed source: the conflicts are taken from the Trade Marks Journal, which is a public record, so the answer key does not come from the tool being tested.
- A fixed metric: one number, defined before the run, stated together with its sample size.
- A dated run: search tools change, so every result carries the date it was run.
- An open list of misses: the marks the tool failed to find are published, not hidden.
Why we do not publish a headline accuracy number without a sample size
A recall figure with no N and no date cannot be checked and should not be trusted, whoever publishes it. Any figure on this page is given with both, and where the run data is not yet verified we say so instead of rounding up.
04The test set and the metric
The metric is phonetic recall at 50, written recall@50. It asks one simple question: of the real sound-alike conflicts, how many appear in the tool’s first 50 results?
How to build the test set
- Pick a set of N Journal conflicts. Each one is a pair: an earlier mark and a later mark that a reviewer agrees is confusingly similar by sound.
- Record the class of each mark, so results can be split by class.
- Take the later mark as the query. Search for it in the tool, in the right class.
- Check whether the earlier mark appears in the first 50 results. Mark it a hit or a miss.
- Recall@50 equals hits divided by N. State N and the run date next to it.
Why 50 and not 10
A lawyer clearing a mark will read down a list, not stop at ten rows. Fifty is a realistic reading depth for a careful search. You can also report recall@10 for a stricter view.
What the result table should show
| Item | What to publish | Why |
|---|---|---|
| N | The number of Journal conflicts tested | A recall figure means little without it |
| Run date | The day each tool was searched | Tools and records change |
| Recall@50 | Hits divided by N, per tool | The one citable number |
| Misses | The full list of marks not found | Lets others check and learn |
| Class split | Recall per class, where N allows | Averages hide weak classes |
Results from our own run are not stated on this page yet. They will be added together with N, the date and the miss list once the run data is checked.
05How to read the results
Read a benchmark for what it misses before you read what it finds. A tool with a high average can still be weak in the one class you work in.
Look at the misses first
Every phonetic method has blind spots. Typical ones are marks that differ by a leading letter, names that sound alike only when said in a regional accent, and short marks where one changed letter changes the whole word. A good report lists these openly.
Remember what recall does not measure
Recall@50 does not tell you how many false alerts you will read. A tool that returns everything has perfect recall and no use. Track precision as well, meaning how many of the top results are real conflicts, and treat the two together.
A human still decides
No tool decides whether two marks are deceptively similar. A court or the Registrar does that, after looking at the goods, the buyers and the whole mark. A search tool only makes sure you see the candidates. For a view of how marks, patents and court records come together in practice, see the guide on AI and patent FER drafting in India, with its limits.
06Where 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.
For this benchmark, the relevant part is the IPR Solutions tab. It tracks IP India registry matters across trademarks, patents, designs and geographical indications. Its Journal conflict monitoring looks for conflicting marks by text, phonetic and visual matching and alerts the user. It also tracks pending compliances, including trademark renewals and patent annuity and response deadlines, and pending hearings from trademark and patent filings.
On the court side, Claw tracks IP litigation across Indian courts that have a case website. That includes party-name infringement discovery and the High Court IP Divisions, such as Delhi IPD by case-type code and by court, number and party, with hearings, orders and cause lists. It also covers IP judgment research over 30 crore+ judgments with LegalGPT, at a scale of 9,102 courts, benches and forums and 1.5 billion+ records, with search results in under 5 seconds. The Supreme Court is covered from 1950 and High Courts from 1990.
Putting the registry and the court in one place is the point: a conflict found in the Journal can be followed into an opposition or a suit without changing tools. Claw does not yet track copyright registration with the Copyright Office, so for copyright it is court-side only. It does not replace global docketing, specialist prior-art search or marketplace takedown tools. For a wider view of bulk matters, see bulk litigation search and monitoring in India.
IP features are part of Premium and Enterprise. There is no IP-only plan, and the per-search litigation product is a separate item that does not apply to IP.
07Sources and further reading
Primary sources for the points on this page. Last verified October 2026.
- IP India (official trademark records and legal documents): ipindia.gov.in
- WIPO (trademark background): wipo.int
- Indian Kanoon (full text of Cadila Health Care v Cadila Pharmaceuticals): indiankanoon.org
- Claw pricing: clawlaw.in
Competitor details are described in general terms and should be confirmed on each vendor’s own site.
08Frequently asked questions
What is phonetic recall in a trademark search?
It is the share of real sound-alike conflicts that a tool actually finds. Recall@50 counts a conflict as found if it appears in the first 50 results. It must always be reported with the number of conflicts tested (N) and the date of the run.
Is IP India public search enough for a clearance search?
It is the official record, so every search should include it, but it should not be your only check. Test how it handles sounds-alike and look-alike marks before you rely on it, and add a watch so you are alerted when a new conflicting mark is published.
Do Indian courts care about how a mark sounds?
Yes. In Cadila Health Care v Cadila Pharmaceuticals, (2001) 5 SCC 73, the Supreme Court treated phonetic similarity as an important factor in deceptive similarity. Read the full judgment on Indian Kanoon before citing it.
Does Claw monitor the Trade Marks Journal?
Yes. The IPR Solutions tab watches the Trade Marks Journal for conflicting marks using text, phonetic and visual matching, and alerts the user. It also tracks trademark renewals, patent annuity and response deadlines, and pending hearings.
Does Claw track copyright registrations?
Not yet. Claw does not track registrations at the Copyright Office. For copyright it is court-side only, covering infringement matters in Indian courts.