Most leadership teams now ask some version of this:
“Are we showing up in ChatGPT and other LLMs?”
The honest answer is usually: not much.
Not because the models can’t recognize your company.
But because most brands don’t get named in answers, even when a model could name them. There is a huge drop off in mentions in answers.
We call this the The Cliff.
The numbers (the short version)
This analysis comes from a proprietary benchmark from LLMtel.com:
- 1,000 entities
- 17 chatbots/LLMs per entity
- 5,070 total prompts (from 4,950 unique question strings)
- That produced 86,190 total model answers
Here’s the headline:
- The typical entity shows up in about 2.4% of answers.
In other words, half the entities appear less often than that. - 355 entities were never named even once.
That’s 35.5% of the list (355 out of 1,000).
Translation for executives:
Most brands are not fighting for first place in AI answers. They’re fighting to show up at all.
Why this surprises smart people
Leaders often test AI visibility like this:
- “Tell me about Company X.”
That forces the model to talk about you. It’s like asking a reporter to write a profile on your company. Of course your name appears.
But real customer behavior is closer to:
- “What’s the best project management tool for contractors?”
- “What paint brand is best for interiors?”
- “What’s a reliable credit union?”
- “What’s a good payroll provider in Australia?”
In these questions, models often do three things:
- Answer without brands (safe, general guidance)
- Name a few default examples (the usual suspects)
- Avoid niche picks unless the user asks for them
That’s how the cliff forms.
Two different scores: “Do they know you?” vs “Do they say you?”
Most “AI visibility” conversations mash these together. We separated them.
1) Entity Score = recognition
This asks: When the entity is the topic, how many of the 17 models recognize it?
Think of it as: awareness inside the model.
2) Questions Score = mentions in real answers
This asks: Across lots of real questions, how often does the entity name appear in the answers?
Think of it as: being surfaced when it matters.
These two are connected but they are not the same thing.
In the full 1,000-entity dataset, the relationship is strong but not perfect (correlation ≈ 0.653). Plain English: models can know a brand and still not name it in an answer.
The “silence” problem is bigger than the “competition” problem
A brand leader’s default mental model is competitive:
- “Are we winning against competitors in AI answers?”
But your first problem is usually more basic:
- “Are we even present?”
Because in this dataset:
- One in three entities were never named in the answers at all
- And the typical entity was named so rarely that most executives would call it “invisible”
This is the key reframing:
- Most brands are not losing to a competitor. They are losing to generic answers and default lists.
The Cliff: AI attention is not evenly shared
The decline in mentions is not a gentle slope. It’s a steep drop-off.
In our benchmark, the “top of the cliff” is wildly different from the middle and bottom:
- A “top 25%” entity shows up around 10% of the time
- A “top 10%” entity shows up around 40% of the time
- A “top 1%” entity can show up over 80% of the time
That’s why some brands feel “everywhere” in AI answers. They’re not slightly ahead. They’re way above you.
Why the mention cliff happens (the executive version)
1) LLMs aim for “helpful,” not “complete”
Why the mention cliff happens (the executive version)
2) They prefer safe defaults
When unsure, models pick widely-known examples that are hard to argue with.
3) Knowing doesn’t guarantee retrieval
A model can recognize your brand in a direct question, but still fail to pull it into a recommendation answer.
4) Many questions don’t require brand names
If the model can answer with categories (“use a CRM,” “look for X features”), it often will.
5) Naming inconsistency hurts recall
Small changes spelling, abbreviations, punctuation, domain vs brand name can scatter your presence and reduce consistent mentions.
What this means for your business
If you’re not named, you’re not in the funnel
AI answers are becoming a shortlist builder. If your brand isn’t named:
- You aren’t considered
- You aren’t compared
- You aren’t clicked
- You don’t exist in that moment
Your real enemy is “silence”
Most companies are not being outranked. They are being excluded. There is no page 2.
What to do next (practical plan for a C-level team)
Step 1: Measure “mention rate” like a KPI
Stop asking, “Does AI know us?”
Start tracking, “How often do we appear when buyers ask real questions?”
A simple version:
- Pick 30–50 real buyer questions
- Run them across a panel of models (not just one) using a tool like LLMtel.com
- Track: % of answers that name your brand
- Track the same for your top competitors
This becomes your AI equivalent of share of voice.
Step 2: Focus on a few “owned intents”
Trying to be “generally visible” is a trap.
Pick a small set of high-value intents where you want to be the obvious answer:
- One industry
- One use case
- One buyer type
- One region (if relevant)
In AI, clear positioning beats broad messaging.
Step 3: Fix your identity so machines can repeat it
Your goal is to make it easy for models and sources to agree on what you’re called.
- One canonical brand name (the exact same name everywhere)
- Consistent product naming
- Consistent domain usage
- Consistent “what we do” language across credible sources
Step 4: Treat AI visibility as an operating system, not a campaign
- Models change. Data changes. Competitors publish too.
- Measure it monthly or quarterly like pipeline, conversion, and share of voice.
Bottom line
This benchmark doesn’t say, “LLMs don’t know your brand.”
It says something more important: In real answers, most brands barely get named.
And a large slice never gets named at all.
So the winning executive question isn’t: “Does AI know us?”
It’s: “When our buyers ask the questions that drive revenue, do we show up enough to matter?”
Because in AI answers, attention is not evenly shared. It’s concentrated at the top. And don’t forget: If you’re not at the top of the cliff, the default state is silence.