Say hello to the new Saazy! See what’s new ✨

From “Ignored” to “Named”: What Makes AI Actually Mention an Entity?

Date

Author

In the LLMtel 1,000 entity study, one result should make every executive pause:

In the data, 319 entities fall into a bucket we’ve been calling “Ignored”:

This companion article answers the next executive question:

The hidden problem: “AI knows you exist” is not the same as “AI recommends you”Dynamic Content

The Ignored bucket proves a hard truth:

The models can “know” you in the abstract and still avoid naming you when asked:

In the study, the question set is mostly practical and action-oriented. Many prompts start with words like How, What, Where, Who, Any. That’s “help me decide” language.
So the model is not behaving like an encyclopedia.
It is behaving like a recommendation engine under uncertainty.
And recommendation engines avoid risk.

A quick story from the Ignored bucket

One example from the study makes the gap obvious:

One example from the study makes the gap obvious:

The “Named” Levers: what makes an entity show up in answers

One example from the study makes the gap obvious:

Wikipedia helps with the first gate as our previous post and analysis clearly showed. But “Named” depends heavily on the other two. Below are the most reliable levers to move an entity out of Ignored.

Lever 1: Make your category unavoidable (in the exact words people ask)

Models don’t just retrieve “companies.” They retrieve companies in categories.
If your public footprint doesn’t loudly connect you to the category language users type, you’ll stay Ignored even if the model recognizes your name.
What to do:

Why it works:
When the model sees the same category associations across many sources, it becomes easier to retrieve you when the question is asked.

Lever 2: Build “adjacency” to well-known entities (comparisons, alternatives, integrations)

LLMs frequently answer by anchoring on a few “default” names. If you are not connected to those names in public text, you don’t get pulled into the answer.
What to do:

Why it works:
Adjacency is how models learn who belongs in the same shortlist.

Lever 3: Increase third‑party validation (models trust outside voices more than you)

If your footprint is mostly self-written, the model often treats it like marketing especially in recommendation mode.
The strongest “Named” signals are third-party:

What to do:

Why it works:
Third-party sources reduce the model’s uncertainty. Lower uncertainty = more willingness to name you.

Lever 4: Fix naming and disambiguation (models hate messy names)

The dataset includes “unknown but named” cases entities that show up in answers even when the entity test says models don’t recognize them. That’s a warning sign: models sometimes “name-drop” based on text patterns.
You want the opposite:

What to do:

Why it works:
If the model can’t reliably map mentions to one entity, it will often avoid mentioning you at all.

Lever 5: Give the model “safe-to-recommend” proof (reduce perceived risk)

In recommendation answers, models try to be safe. They lean toward:

If you don’t look “safe,” you don’t get named.

What to do:

Why it works:
The model is more likely to name entities that look established, verifiable, and low-risk.

Lever 6: Show up where models expect to “find vendors”

Many Ignored entities live on the edges of the public knowledge graph.
Beyond Wikipedia, there are “vendor surfaces” models often learn from, such as:

What to do:

Why it works:
Models learn patterns from repeated, consistent placement. Being scattered across low-quality pages doesn’t help much.

What NOT to do (because it backfires)

If your goal is long-term AI visibility, avoid tactics that pollute trust:

The point is to become more verifiable, not louder.

A practical executive playbook (30–60–90 days)

Days 0–30: Diagnose and clean identity

Days 31–60: Build category and adjacency

Days 61–90: Add third-party confidence

Ongoing: Measure mention-rate, not just awareness

Recognition is nice. Mentions are the KPI.

The simple takeaway

Your Wikipedia article showed something important: Wikipedia multiplies recognition, even inside the Ignored bucket.

This article is the next step:

In an AI-first discovery world, the winners won’t just be the companies that exist online.

They’ll be the companies that AI can confidently say out loud.

Leave a Reply

Your email address will not be published. Required fields are marked *