New brand invisible in AI answers: how to speed up recognition

You launched the brand six months ago, the site is doing well, you already have a few mentions in the press — but in AI answers you don't exist. It's not a quality problem: to the models you're still a stranger, a name that has been around for too short a time, with too few structured signals to deserve trust. Meanwhile, competitors with established brands keep collecting visibility that should also be going to you. Four specific interventions can speed up recognition from years to four or five months.

Let me guess: you launched a new brand six months ago, you get decent traffic, you have your first mentions in the trade press. But when someone asks ChatGPT for advice in your vertical, your name doesn’t come up. Normal? Yes, and I’ll explain how to speed up AI visibility even for a new brand.

The point is that AI engines aren’t ignoring you out of spite: they simply don’t know who you are yet. To them you’re an emerging entity. A string of text that has only recently appeared, with few structured signals around it, without a stable slot inside their representation of the world. And until you earn that slot, you’re outside the set of recommendable brands.

What it means to be an emerging entity for an AI model

In the world of NLP research, the problem of recognizing new entities was studied systematically at the WNUT workshop in 2017, with a dataset built specifically to measure how capable systems were of handling rare and newly appeared entities.

“The goal of this task is to provide a definition is to develop systems that are less sensitive to tion of emerging and of rare entities, and change, and can handle rare and emerging entity based on that, also datasets for detecting types with ease.”

Derczynski et al. (2017)

Translated into concrete terms: the systems that today sit behind ChatGPT, Claude, Gemini and Perplexity have inherited an old problem. Recognizing established entities (Apple, Ferrari, University of Bologna) is easy because the signal is enormous; recognizing new entities is hard because the signal is scarce and noisy.

The operational consequence for you is simple. If your brand is less than 2-3 years old and you’ve done nothing to make your entity identity explicit — a Wikidata page, a Wikipedia entry when possible, structured data on the site, mentions in authoritative sources — the model doesn’t recommend you not because you’re bad, but because to it your existence is statistically weak.

You can have content written in the inverted pyramid, you can have well-built E-E-A-T, you can have nurtured author entity recognition. If your brand name isn’t a recognized entity, everything else loses force. The model reads your content, but doesn’t know who to attribute it to in its internal graph.

The case: an eight-month-old B2B SaaS from Padua

Let me tell you a concrete, anonymized case.

A B2B SaaS from Padua in the HR tech segment, launched eight months ago. Solid product, 60 paying customers, stable organic traffic around 4-5k users a month, a few articles on HR trade publications and two podcasts. No AI citations: when the founder asked ChatGPT or Perplexity for “best platforms for HR onboarding in Italy”, the brand never came up. Zero times out of 20 queries tested across three different engines.

The intervention, over five months, was this:

• creation of a complete Wikidata entry (item with Italian and English label, short description, properties “instance of: software”, “country: Italy”, “date of founding”, link to the site, link to Crunchbase and company LinkedIn)
• addition of `Organization` + `SoftwareApplication` structured data on the homepage, with `sameAs` fields pointing to Wikidata, LinkedIn, Crunchbase
• publication of a well-filled-out Crunchbase profile (founding round, founders with links to profiles, consistent category)
• two press releases to strongly indexed digital publications (one general tech, one HR vertical) with the brand name always in an identical canonical form
• an “About us” page rewritten in the inverted pyramid, with brand name + what it does + who it’s for in the first 30 words

After five months: out of 20 test queries redone with the same protocol, the brand came up cited 6 times out of 20 (from zero to 6). The most interesting pattern was that Perplexity cited it before the others, probably because it aggregates fresh web sources more quickly.

Honest limitation of this case study: it’s a single client, a specific industry, a narrow query sample. It’s not a study. But I’ve seen the pattern repeat on other B2B tech startups I followed afterwards.

The test you can run in 15 minutes on your brand

Before you move, check where you stand today.

Open Wikidata and search for the exact name of your brand. If there’s no dedicated item, you’re officially invisible at the most structured level of the graph. If it exists but is empty (just a label, no properties), the signal is weak.

Open Google’s Rich Results Test, paste your homepage URL and look in the result for the `Organization` block. If it’s not there, the site isn’t sending the web the basic signal “I am this company, these are my profiles”. If it is there, check that the `sameAs` field is present with at least LinkedIn and Crunchbase.

Then do the direct test. Open ChatGPT, Claude, Perplexity and Gemini and ask 3-5 queries like “best [your category] in Italy” or “alternatives to [an established competitor in your industry]”. Count how many times your brand is cited. Binary threshold: if you’re at zero across all four, the problem is an entity one. If you’re cited 1-2 times by someone, you have a presence signal but still a fragile one.

These are entry-level checks. The full analysis — semantic distance from the cited competitors, coverage across engines, co-occurrence patterns — requires professional tools. But these three tests in 15 minutes already tell you where to start.

The mistakes I see most often with new brands

Non-canonical brand name. The site uses “Acme Solutions”, LinkedIn “Acme Srl”, Crunchbase “ACME”, the press releases “Acme Italia”. To the AI these are four weak entities, not one strong one.

Wikidata ignored. The founder thinks Wikidata is stuff for encyclopedists. In reality it’s the structured registry most read by AI engines to anchor entities. Not having an entry there when you’re new is one of the very few levers with zero cost and high return.

Structured data missing or buggy. The site has a WordPress theme that inserts a minimal `Organization` without `sameAs`. The engine reads an entity disconnected from the rest of the web.

Chasing Wikipedia too soon. A Wikipedia entry on an 8-month-old brand gets deleted for non-notability. First you build notability (press, mentions, Crunchbase, Wikidata), then the Wikipedia entry comes naturally when you’re ready.

What to do concretely in the next 60 days

• lock the canonical name of the brand into a single form and align it everywhere: site, footer, meta tags, LinkedIn, Crunchbase, press releases
• create the Wikidata entry using the items of similar, more established companies as a template, with completed properties and sameAs links
• add `Organization` structured data to the homepage, with `sameAs` pointing to Wikidata, LinkedIn, Crunchbase and all official social profiles
• fill out Crunchbase completely: founders with profile links, consistent category, round, headquarters
• plan 2-3 PR pieces over the next months on strongly indexed publications, with the brand name always in the canonical form

It’s not magic. It’s making the AI understand that you exist, under what name, in what category, who you’re connected to. From there, if the product is good and the content holds up, the citation in AI answers comes.

The thread of AI visibility

Working on emerging entity detection is the foundation for showing up in AI answers when your brand is young. Without this step, every investment in content yields less than it could, because the model doesn’t know who to attribute you to.

In the next articles we’ll look at the other nodes of the knowledge graph that work together with this one: how disambiguation is built when your brand name is ambiguous with other entities, how the relationship between entities and co-occurrence works in AI answers, and how you measure the strength of your entity over time. If you want to go deeper into how the models represent entities in vector space, you’ll find the dedicated piece on embeddings and vector space.