How many phrases should I start with?

Start smaller than feels ambitious — a dozen well-chosen phrases that cover your core product category, your top two or three use cases, and a couple of direct brand-vs-brand comparisons will tell you more than fifty vague ones. You can always add phrases once you see which ones surface real signal; you can't easily tell which of fifty phrases are actually useful without first seeing results from a focused set.

Do I need API keys for every provider to get started?

No. Depending on your plan, a shared platform-managed key may already be available for supported providers, so you can start tracking without provisioning your own credentials first. Bringing your own provider key is generally an option for advanced/self-managed setups. Check your workspace's provider settings to see what's enabled by default before assuming you need to source API keys yourself.

How is a competitor different from just another phrase mention?

A phrase is the question you're asking a model; a competitor is a brand you're watching for inside the *answer* to that question, alongside your own brand. Defining competitors up front means every run's mention extraction checks for all of them at once, so a single scheduled run tells you not just 'were we mentioned' but 'were we mentioned instead of, alongside, or ahead of X.'

What happens if a scheduled run comes back with zero mentions?

That's a real result, not a failure — a zero-mention run for a given phrase and provider is exactly the kind of signal this whole system exists to surface. It's a prompt to look at whether the phrasing is realistic, whether your public content actually answers that question clearly, or whether it's simply a category where models tend not to name any brand at all. Trend data across multiple runs tells you which explanation fits.

Guides

Setting Up Brand and Competitor Tracking in LLM Glow: A Complete Walkthrough

LLM Glow · 5 min read · July 11, 2026

This is a hands-on walkthrough of setting up your first tracked workspace in LLM Glow — from creating the workspace through reading your first visibility trend. If you’re new to the concept of LLM visibility itself, our complete guide covers the “why” before you get here; this post is the “how.”

1. Create your workspace and define your brand

A workspace is the container for everything: your brand, your competitors, your tracked phrases, your provider settings, and your run history. When you set it up, you’ll define your own brand first — the entity every mention extraction pass checks for by name.

Take care with how you name your brand here. If your product is commonly referred to by a shortened name, an abbreviation, or a former product name that’s still floating around in public content, note that too — mention extraction works from what a model’s response actually says, and a model won’t necessarily use your exact legal or marketing name.

2. Add your competitors

Next, define the competitor brands you want every run to check for alongside your own. This is what turns a single scheduled run into a comparative result: instead of just “were we mentioned,” you get “were we mentioned instead of, alongside, or ahead of competitor X” for the same question in the same response.

A practical starting list is the competitors your sales team already gets asked about, or the brands that show up when you search your own category — not an exhaustive list of every company that’s ever competed with you. You can add more later once you see which comparisons actually come up in real model responses.

3. Write phrases that surface real signal

A phrase is the question a model gets asked, once per provider, on your schedule. This is the single highest-leverage step in setup, and the one worth the most care.

Guidelines that hold up in practice:

Write the way a real prospect would ask, not the way you’d write ad copy. “best tool for tracking engineering deadlines” surfaces far more useful signal than “why choose [Your Brand].”
Cover multiple intents, not just your product’s name. Include category questions (“what’s a good alternative to spreadsheets for sprint planning”), use-case questions (“how do small teams track project deadlines without a dedicated PM”), and direct comparisons (“X vs Y for engineering teams”) — each intent type tends to surface a different mix of brands.
Start with a focused set, not an exhaustive one. A dozen well-chosen phrases across your core category and top use cases will teach you more in the first few scheduled runs than fifty vague ones, and you can expand from there once you see what’s actually producing signal.
Revisit phrases that come back consistently empty. A phrase with zero mentions across several runs and providers is telling you something — either the phrasing doesn’t match how people really ask, or it’s a genuinely brand-agnostic category. Either way, that’s useful information, not a setup mistake to fix reflexively.

4. Configure providers

Runs execute across multiple LLM providers — OpenAI, Anthropic, Google, and others, depending on what’s enabled for your workspace. Running the same phrase set across more than one provider matters because different models are trained differently and can genuinely disagree about which brands to name; measuring a single provider tells you about that provider specifically, not about “LLMs” as a category.

Depending on your plan, a shared platform-managed provider key may already be configured so you can start running phrases immediately, with the option to bring your own provider credentials for more advanced or self-managed setups. Check your workspace’s provider settings before assuming you need to source your own API keys.

5. Let runs execute on schedule

Once your phrases, brand, and competitors are set, runs execute per-phrase across your enabled providers on your configured schedule. Each run’s response goes through mention extraction — identifying every point in the model’s answer where your brand or a defined competitor is named — and that extraction feeds into your visibility score for that phrase and provider.

A single run is a snapshot; the value compounds as runs accumulate into a trend. Resist the urge to judge your visibility off one run’s result — the whole reason this system runs on a schedule, rather than as a one-off check, is that generation is probabilistic and a single answer isn’t representative on its own.

6. Read your trend and set up alerts

Once you’ve got a few weeks of scheduled runs behind you, your trend view is where the real signal lives: is your mention rate for a given phrase climbing, flat, or dropping — and how does it compare to your defined competitors over the same window? That’s the number a single manual chat prompt can never give you, because it has no memory of yesterday’s answer.

From there, set up alerts on the changes that actually matter to you — a meaningful drop in mention rate for a core phrase, or a competitor starting to appear where they hadn’t before — so a real shift reaches the right person before it’s been quietly happening for weeks.

Ready to set this up for your own brand?

If you haven’t created a workspace yet, get started with LLM Glow and walk through the steps above with your own brand, competitors, and phrases. The setup itself takes minutes; the value is in the trend line that builds once your scheduled runs start accumulating real data.

How many phrases should I start with?: Start smaller than feels ambitious — a dozen well-chosen phrases that cover your core product category, your top two or three use cases, and a couple of direct brand-vs-brand comparisons will tell you more than fifty vague ones. You can always add phrases once you see which ones surface real signal; you can't easily tell which of fifty phrases are actually useful without first seeing results from a focused set.
Do I need API keys for every provider to get started?: No. Depending on your plan, a shared platform-managed key may already be available for supported providers, so you can start tracking without provisioning your own credentials first. Bringing your own provider key is generally an option for advanced/self-managed setups. Check your workspace's provider settings to see what's enabled by default before assuming you need to source API keys yourself.
How is a competitor different from just another phrase mention?: A phrase is the question you're asking a model; a competitor is a brand you're watching for inside the *answer* to that question, alongside your own brand. Defining competitors up front means every run's mention extraction checks for all of them at once, so a single scheduled run tells you not just 'were we mentioned' but 'were we mentioned instead of, alongside, or ahead of X.'
What happens if a scheduled run comes back with zero mentions?: That's a real result, not a failure — a zero-mention run for a given phrase and provider is exactly the kind of signal this whole system exists to surface. It's a prompt to look at whether the phrasing is realistic, whether your public content actually answers that question clearly, or whether it's simply a category where models tend not to name any brand at all. Trend data across multiple runs tells you which explanation fits.