G&S AI Similarity: See How Close the Goods and Services Really Are

Consider the following scenario; You search LEICON and get back 700+ marks. They all sound about right, spread across every industry there is. But you're clearing a camera brand, and almost none of them are about cameras. So, you read specifications one at a time, sorting the real conflicts from the paint, pet food and insurance. An hour has gone by. 

That last step, judging how related the goods are, is where the real clearance work hides. Two marks can look and sound almost identical and still coexist, because they cover goods that never meet. The name gives you a list of namesakes. The goods decide what matters. 

ProSearch's G&S AI similarity does that part with you. Describe the goods your new mark will cover, and it scores every result in % on how close its goods are to yours, then you can drop the ones that aren't close enough. The 700+ LEICON hits can be swiftly reduced to the handful that touch cameras. 

It matches meaning, not words 

It reads what the goods are, not the words used for them. "Cameras" and "photographic apparatus" share no words and score high on similarity. "Apple juice" and "Apple computers" share a word and score low. A keyword filter gets both backwards. 

Each description becomes a set of numbers standing for its meaning, and the system measures how close two of those sit. That's why it's AI matching, and why it can't hand you a written reason for any single score the way a chatbot would. It measures similarity. It doesn't form an opinion you can argue with. 

Reading the score 

Each surviving result carries the closest match between any of your descriptions and any single term in that mark's specification. A rough guide: 

• 90 to 100: the same goods, or nearly so
• 70 to 89: related in meaning, might compete in similar markets
• 50 to 69: broadly related, or sharing some traits
• below 50: little real connection 

Treat it as a guide, not a cut-off. What survives depends on more than the raw score. 

Why chocolate shows up in a cigar search 

Unexpected %-matching does not automatically point to faulty logic.  Search for cigars and chocolate and results with G&S for the near identical “cigars” will return a high percentage of similarity. However, results with G&S, including candles and coffee, might land in the low 70s, alongside the near-identical "cigars" at 92%, but nowhere near noise.  The reasons being; those goods share real traits: premium positioning, a sensory experience, the same gift-giving moments, neighboring Nice classes, the same shelves in the same shops. A lifestyle brand really can stretch from cigars to chocolate to candles, so a mark in one is a fair worry for a mark in another. That's the extension risk a careful attorney flags by hand. 

Get better results 

Describe your goods in a few plain words, not one. "Prescription eyewear and sunglasses" beats "glasses"; "tablet computers and accessories" beats "tablets." A lone word like "chips" sits between snacks and semiconductors and matches both weakly. Three to ten words also helps the system pick the right class, in case you do not want/cannot provide it. 

Keep one product area per box. You can enter up to three categories, and they work as OR, not AND, so adding categories widens your search rather than narrowing it. Put genuinely different goods, like clothing and jewelry, in separate boxes. 

It doesn’t add, it doesn't reorder 

Activating the G&S AI Similarity model never adds results. Every retrieved mark will be scored, because it was considered relevant from a name-perspective by the underlying search. If you search based on Classes and simultaneously add AI similarity scoring on the list of goods and services, you should retrieve a G&S similarity score that is less wide. If you, however, choose to omit specific classes, you may expect a broader range of similarity scores.  

It never reorders the results set, either. The rows keep the normal name-relevance order, so the G&S similarity column won't read top to bottom; a 90 can sit below a 60. To put the closest goods matches on top, click the column header. This way you can sort by descending and ascending order.  

vs

See what drove the match 

The matched wording is highlighted inside the Goods and Services text, and hovering shows the score for that phrase. Behind the headline number sits each matched phrase with its own score, strongest first. That's the difference between "this looks relevant" and knowing exactly which goods pull the two marks together, the line you can quote in a clearance memo or office-action response. 

FAQ 

Does my search data go to an outside AI company? No. The model runs on our own servers. Your search terms, goods’ descriptions, and results aren’t sent to any third party. The whole comparison happens in-house. 

Will the same mark always get the same score? The score depends on how you frame your description, so two people searching for what they'd both call "the same thing" can see slightly different numbers. Frame the goods as they'd read on a specification, and the scores stay steady. 

How do I turn it on, and why might I not see it? Open the Goods and Services dialog, choose AI matching, and describe your goods. If the option isn't there, your account doesn't have it enabled yet. Your account manager can help you out. 

Every market, and the API 

Highlighting, although available for all markets, will become visible only when G&S are translated to English. The same scoring is available through the documented API, returning the score and per-phrase breakdown for eligible accounts, so automated clearance and watch workflows can rank and filter the same way the interface does.