What a naming session with a model really is
Brainstorming product names with a model feels mysterious until you see it for what it is: a fast conversation that gives you raw material. You describe the thing you are building, the model throws back dozens of candidates, and you sort the pile. The model does not choose your name for you. It multiplies your options so you can pick with your own judgment.
That division of labor matters. A model has no stake in your brand and cannot know which name your customers will trust. What it can do is reach into a huge vocabulary of words, sounds, patterns and cultural references, then combine them far faster than a notebook and a cup of coffee. Treat the output as a quarry, not a finished statue.
For newcomers, the useful mental shift is this: you are not asking a machine for the answer. You are running a structured rehearsal where the model plays the role of an endlessly patient brainstorming partner who never runs out of ideas and never gets tired of your constraints.
Write a one-paragraph brief before you type anything
Good names come from good inputs. Before your first prompt, write a short brief that answers a few plain questions. Keep it to a paragraph or two so you can paste it straight into the chat.
- What is the product? Say it in one sentence a stranger would understand.
- Who is it for? Describe the person who will use it and what they care about.
- What feeling should the name carry? Trustworthy, playful, technical, warm, bold, calm.
- What words belong near it? List five to ten terms from your field that you like.
- What words must stay far away? List the clichés and jargon you are tired of.
- Are there boundaries? Length limits, language rules, pronunciation needs, or names already taken.
That last point deserves a moment. Naming is not just creative, it is practical. A name that is impossible to spell after hearing it once, or that hides behind a crowded trademark field, creates work later. Include those constraints early so the model stops wasting suggestions on candidates you could never use.
You can draft this brief in any tool you like, including a coding assistant. Many teams already route their AI traffic through SwiftRelay, a pay-as-you-go relay for Claude, GPT, Gemini, DeepSeek, Kimi, GLM, Qwen, Grok and other models, so the same account that handles your code can handle a naming session. It works with Claude Code, Codex, Gemini CLI and any OpenAI-compatible SDK, which means you can brainstorm from the environment you already have open.
Run the first lap with three simple prompt patterns
Once the brief is ready, run three short passes instead of one giant request. Each pass has a different job, and together they give you a wide spread of candidates.
Pass one, the wide net. Paste your brief and ask for a large batch of names in different styles: real words, invented words, short compounds, and names built from a metaphor. Ask the model to give one line of reasoning for each. The reasoning is often more valuable than the name, because it shows you which creative routes are fertile.
Pass two, the direction test. Pick two or three routes that surprised you in pass one and ask for a deeper dive into each. For example, if a handful of candidates played on the idea of a relay or a handoff, ask for twenty more names in that same family. Repetition with variation is where naming gold usually appears.
Pass three, the stress test. Ask the model to critique the list. Which names sound like existing products, which are hard to say, which carry an unintended meaning, which are forgettable? This pass is where a good model earns its keep, because honest criticism saves you from falling for a clever name that has a hidden flaw.
For a session like this you want a model with a strong ear for language. A deep, contemplative model such as GPT-6 Astra is a natural fit when you want varied wordplay and thoughtful critique, while a lighter model is fine for quick batch generation.
If you are testing several models against the same brief, keep your setup identical and let the results speak. Billing on SwiftRelay is per token with no subscription, so a handful of exploratory sessions costs exactly what you use and nothing more. The API host is https://api.fastapirelay.com, with the bare host for Claude Code and Gemini CLI and /v1 added for OpenAI SDKs.
Sort the pile before you fall in love
Now you have a long list, maybe a hundred candidates, and the temptation is to circle the one that made you smile first. Resist that. Sorting is its own pass, and it works best on paper or in a simple table.
Group the names into buckets rather than ranking them one by one. Useful buckets include strong and clear, interesting but risky, too generic, too strange, and close to something existing. This turns an emotional decision into a pattern you can see.
Then apply a few plain filters to every survivor:
- Say it out loud. Can you pronounce it without explaining it?
- Spell it after hearing it once. Give yourself a three-second test.
- Write it in a sentence. Does it read naturally in ordinary language?
- Shorten it. Is there an obvious nickname your users would invent?
- Search it. Does a crowded field of similar names make it invisible?
Keep the ones that pass most filters, even if they are not the most exciting. A name that works every day beats a name that dazzles once.
Check the shortlist against reality
Before you commit, take the shortlist out of the chat and into the real world. Ask a few people who are not on your team to read the names cold, with no context. Note which ones they repeat back correctly and which ones they frown at. Ten minutes of this beats a week of internal debate.
Then handle the practical checks. Look for obvious conflicts in your market, confirm the domain and handle situation is workable, and make sure the name does not mean something unfortunate in the languages your customers speak. A model can help you brainstorm translations and slang meanings, but verify anything important yourself.
It also helps to see how the name behaves in real sentences. Ask the model to write a fake announcement, a help center title and an error message using the name. If it feels awkward in everyday software language, that awkwardness will follow you for years.
Checklist before the green light
When you are close to a decision, run through this short list once more and see whether any item gives you pause. If every line feels clean, you are ready.
- The brief is still true. The name matches what the product actually does today.
- The creative route is defensible. You can explain the thinking in one sentence.
- Nothing important is taken. Legal and market checks look clear.
- People get it. Cold readers repeat it correctly and without prompting.
- It works in software. It reads fine in headings, buttons and error text.
- You still like it tomorrow. Sleep on it before the final call.
Naming with a model is a loop, not a magic trick: brief, generate, sort, check, decide. Run the loop a few times and you will find that the model gets better at helping precisely because your brief gets sharper. Sign up for SwiftRelay and run your first naming session today, then keep the same workflow for every new product that follows.