You have a three-thousand-word page about your main service, well written, detailed — but your brand name appears only once, at the very bottom. To the AI, that page isn't about you: it's about the topic, and you're just an extra. You're producing content that educates the market without building authority for yourself. Being the recognizable protagonist of every page you publish is the bare minimum to go from invisible to recommendable — and it can be measured and fixed in just a few minutes.
It doesn’t matter how many times your brand appears in the text. What matters is where it appears and what sits around it. You can have a three-thousand-word page about auto body repair, mention your name once in the footer, and to the AI that page isn’t about you. It’s about the topic, and you’re an incidental detail.
This mechanism has a technical name, entity salience, and it decides whether you get associated with the page’s topic when an AI engine reads it. Let me explain how it works and what to do concretely so that ChatGPT, Perplexity and Gemini’s answers treat you as central and not peripheral within your own content.
What an entity’s salience really measures
In the world of Natural Language Processing research, entity salience is not “how many times you’re mentioned.” It’s a measure of centrality: how much your brand is the protagonist of the document, not an extra.
“Thus, entity salience in a text is defined as a binary or ordinal rating to quantify the extent to which a target entity is central to a given piece of text (Gamon et al., 2013; Dunietz and Gillick, 2014).”
Translated for you: salience is a rating — binary (central yes / central no) or on an ordinal scale — that the system assigns to every entity present in the text. Raw frequency doesn’t count. What counts is whether that entity is what the document is really about.
The practical consequence is simple. If you run an auto body shop in Turin and on the “bumper repair” page your name appears only once, hidden at the bottom of the page, to the AI that page isn’t “about” Esposito Auto Body. It’s “about” bumper repair, with an incidental entity that happens to be called Esposito Auto Body. The moment a user asks Perplexity “where can I repair a bumper in Turin”, your brand doesn’t get retrieved as an answer associated with the topic.
Why salience comes before embeddings and the knowledge graph
In the previous articles on embeddings I explained how the AI turns every word into a point in semantic space. Salience is the next logical step: when the AI builds a representation of the document, it doesn’t treat all entities with the same weight. It picks some as “the ones the text is about”, and demotes the others to background.
“Understanding the aboutness of a document is one of the long-standing goals of research in both Information Retrieval and Natural Language Processing (Gamon et al., 2013).”
The key concept here is aboutness: what a document is “about”. Information retrieval and NLP have been working on it for years because it’s at the heart of how an engine decides which sources to pull up for a question.
For you, this means that salience intertwines with author entity recognition and with the pyramidal structure of content: if your brand isn’t salient on the right page, it doesn’t matter how authoritative the author is or how well written the intro is. The AI simply won’t associate that page with your name when it builds the answer.
Repeating “Turin auto body shop” fifteen times doesn’t create salience for your brand.
The five-minute test with displaCy ENT
You don’t need a lab. Open displaCy ENT, paste in the text of your most important page — the homepage or the landing page for your main service — and look at which entities the system recognizes and with which label.
What you need to check is binary:
- Is your brand recognized as an organization-type entity (ORG)? Yes / no.
- Does it appear in the first 200 words of the text? Yes / no.
- Does it appear at least once in a heading? Yes / no.
If all three answers are “yes”, you’re in a decent position. If even one is “no”, you have a salience problem on that page. Keep in mind this is an entry-level check: real salience is measured with more sophisticated language models, and serious analysis requires professional tools. But to figure out whether you’re off track, these three binary checks are enough.
Open displaCy ENT and verify that the brand is recognized as ORG on the homepage and on the first three service pages.
Over the last 6 months across 40+ SMB sites: the pattern I’ve seen
Let me tell you what I observed by systematically looking at Italian SMB websites over the last six months, from Turin auto body shops to professional firms across the north and south: the pattern is so consistent it’s worth naming.
On a service page like “auto glass repair”, the body shop’s brand appears on average 1-2 times, often relegated to the footer or the “contact us” block. The topic, on the other hand (glass, bumpers, paintwork), appears 30-50 times. A brand/topic ratio below 1%. To the AI that page isn’t a commercial identity, it’s an anonymous technical manual.
What’s interesting is that when I ask ChatGPT or Perplexity “who repairs auto glass in Turin, Mirafiori area”, the sites that get cited almost always share three traits: brand in the H1 or the first paragraph, brand repeated 3-5 times in the body in different positions, brand + geographic area mentioned together at least once. It’s not a controlled experiment, it’s a recurring observation on a sample that’s not small but isn’t an academic study either. Still, the pattern holds.
The mistakes I see most often
Four patterns that recur on almost every SMB service page I review:
- Brand only in the logo and footer. The AI doesn’t read the logo as text. If your name isn’t written out clearly in the content, as far as the engine is concerned it doesn’t exist.
- Service page without the brand in the H1. “Auto body repair in Turin” instead of “Auto body repair in Turin | Esposito Auto Body”. You waste the first salience opportunity.
- Keyword stuffing instead of strategic presence. Repeating “Turin auto body shop” fifteen times doesn’t create salience for your brand. It creates salience for the topic. They’re two different things.
- Generic conclusion with no callback to the brand. The last paragraph is the one that carries weight during summarization. Closing it with “rely on experienced professionals” instead of “at Esposito Auto Body in Turin…” throws away the last prominent position.
What to do concretely
An operational audit, three steps, binary thresholds:
- Open displaCy ENT and verify that the brand is recognized as ORG on the homepage and on the first three service pages. If it isn’t recognized, the problem is upstream: it’s probably never written out in full, or it’s always written in lowercase, or it’s split across separate visual elements.
- Count: does your brand appear 3-5 times per service page in prominent positions (H1, first paragraph, a mid-page H2, conclusion)? If you count fewer than 3, rewrite. If you count more than 7, you’re stuffing and hurting readability.
- Compare with the 3-5 competitors the AI cites when you run the query for your industry in your city (“best auto body shop in area X Turin”). Look at how they handle their brand on the same service pages. Almost always you’ll discover that whoever gets cited follows the 3-5 mentions in prominent positions pattern.
None of this is a magic factor. Salience alone isn’t enough to get you into AI answers: it combines with authority, knowledge graph consistency, content structure. But without salience, everything else spins in a vacuum, because the AI isn’t even associating the page with your name.
The thread: from “they read you” to “they recognized you”
The goal of this series on the knowledge graph is to help you understand what happens inside an AI engine when it encounters your brand. Salience is the stage where the machine decides whether you’re the protagonist or the extra of your own site. In the next articles in the series we’ll look at how to move from “I’m recognized” to “I’m connected”: entity linking to hook your brand to the right node of the knowledge graph, Google Knowledge Graph entry to exist as an official entity, and named entity recognition to understand how the models identify you in the first place.
Three steps: being seen, being recognized, being connected. If you skip the first, the other two never happen.
