New Brand Invisible in AI Search? Three First Steps for GEO

AI search is changing how brands get discovered. If you've just founded a brand and excitedly ask ChatGPT, Perplexity, or Gemini "Do you know my company?", you'll most likely be met with silence. This doesn't mean AI has something against you—it's simply how the system works: a model can only reference what it has seen.

The GenScore team illustrated this phenomenon with a vivid example in their blog—when you ask an AI what small-team collaboration tools exist in a particular niche, the AI will fluently list one or two established products while saying nothing about a new brand that launched just two months ago. No negative review, no misinformation—purely "nonexistent."

This "zero mention" state is exactly the real starting point for a new brand entering the field of GEO (Generative Engine Optimization).

Why AI Doesn't Mention You—It's Not Punishment

A generative model doesn't "decide" to ignore you. Its answers come from content it has seen in training data, or from web pages retrieved in real time when answering. A brand that's only two months old, with almost no external content footprint and no third-party mentions, is simply "no rice to cook with" for the model.

GenScore points out that the key question isn't "Why is AI ignoring me?" but rather "What content about my brand on today's internet can the model actually find?" For most new brands, the honest answer is: almost nothing.

This is fundamentally different from another common dilemma. If you already appear in AI responses but the information is outdated or wrong, that's a data correction problem. New brands face an earlier stage—there's nothing to correct, everything needs to be built from scratch.

Where does a generative model's "knowledge" come from?

Understanding the "zero mention" phenomenon requires first understanding the two mechanisms by which large language models acquire information. The first is training data memory: during training, the model reads massive amounts of web pages, books, and conversations, which get compressed into model parameters to form what it "inherently knows." The second is real-time retrieval augmentation (RAG): when a user asks a question, the model first searches the web for relevant pages, then generates an answer based on the retrieval results.

For brands, this means: either your brand information appears extensively in training data (usually requiring years of accumulation and sufficient online presence), or your content can be retrieved in real time and ranks highly. New brands fail on both fronts—the training data cutoff predates your founding, and your pages have extremely low authority in real-time retrieval. This is the complete technical meaning of "a model can only reference what it has seen."

Three Core Directions for GEO Getting Started

GenScore proposes three focal points for new brands building AI visibility, ranked by controllability and speed of results as follows.

First, Make Your Website Clearly State "Who You Are"

Before pursuing external mentions, the most important thing is to ensure any visitor—whether human or AI crawler—can find basic answers about your brand on your official website: what you are, what you sell, whom you serve, how long you've been around.

A complete "About Us" page, consistent brand name spelling across all pages, plus accurate contact and address information—these are the foundation for all subsequent work. GenScore's advice is straightforward: check your homepage and about page. Can someone who knows nothing about you figure out what you do and whom you serve within 30 seconds?

This is entirely within your control, so it should be your first priority.

Publish an llms.txt File

llms.txt is a plain text file placed in your domain's root directory, designed to help AI engines visiting your site more clearly understand the site's content structure—essentially a "content navigation map" for AI.

GenScore specifically emphasizes that this file itself won't bring visibility, nor can it replace genuine content accumulation. It's a structural aid that makes existing content easier to read, not a magic shortcut. Fortunately, the technical barrier is very low: a plain text file, no complex development needed, can go live this week.

What is llms.txt and why does it exist?

llms.txt is a community convention proposed in 2024, initiated by Jeremy Howard of Answer.AI. Its design motivation is: traditional website HTML pages are full of noise—navigation bars, ads, scripts—making AI crawler parsing costly and prone to misinterpretation. llms.txt uses a concise Markdown format to list a site's core pages, content summaries, and structural relationships—essentially providing AI with a "close-reading edition table of contents."

Its difference from robots.txt is: robots.txt tells crawlers "what not to scrape," while llms.txt tells AI "what's most worth reading." Currently, major players like OpenAI and Anthropic haven't committed to officially supporting this standard, but adoption is gradually expanding. For brands, the value of llms.txt isn't in immediate results, but in establishing structured signals ahead of time—so when AI engines start reading your site more seriously, you're already prepared.

Get Your First Batch of Verifiable External Mentions

This is the slowest but most important of the three directions. AI models place far more trust in third-party descriptions of you than in self-introductions on your own website.

But for new brands, this doesn't mean expensive PR campaigns. GenScore's definition is: establishing presence in real places within your industry—a listing in an industry directory, a genuine user review, a mention in a vertical media outlet or forum, a helpful answer in a community you actually participate in.

The key qualifiers are "genuine" and "verifiable." GenScore explicitly advises against buying mentions or reviews to accelerate this process: fake mentions won't build the authority that generative models value, and they're easily detected.

Why It Can't Be Rushed, and Why There Are No Shortcuts

Building a new brand's "entity presence" in AI is a completely different operation from correcting a piece of wrong data. There's no single source you can fix and then wait for results. You need to accumulate enough owned content, enough external mentions, and enough consistency over time before the model has something real to reference.

GenScore declines to give a specific timeline, reasoning that it depends on the industry, publishing frequency, and each engine's knowledge base update cadence—giving fabricated numbers would be worse than giving none.

What is a "brand entity" in AI's eyes?

In the context of generative engines, "entity" is a knowledge graph concept. Search engines and AI models create an entity node for every identifiable object—person, company, product, place—with attributes (founding date, business scope, founders) and relationships (which organizations, people, events it's connected to).

For a brand to "be known by AI" essentially means growing your node in the model's knowledge graph. Node establishment requires cross-verification from multiple sources: your official website, Wikipedia entries, industry media coverage, social media accounts, third-party reviews—all corroborating the same entity. The more sources, the more authoritative, the more consistent—the more solid the node. This also explains why purchased mentions don't work—isolated, low-quality paid comments won't enter the model's knowledge graph construction pipeline, and even if they did, they couldn't form cross-validation with authoritative sources.

Treat Your First Scan as a Baseline, Not a Verdict

If you scan your brand's AI mentions now and get 0 out of 15 queries, that's not a bad result—it's exactly the normal performance of a new brand without historical accumulation, and it's the most useful information right now: it precisely tells you where your starting point is.

GenScore recommends this cadence:

• Do a first scan even if you expect low results—this is the measurement baseline for all future progress
• Re-scan every few weeks, not every day—building entity presence is a slow process, and frequent scanning just measures noise
• Order the three directions by controllability: website entity information first, because it's most controllable
• Save every result—going from 0/15 to 2/15 within two months, while still small numbers, is already a real signal of progress

Mistakes to avoid: don't buy mentions or reviews; don't expect changes within days; don't confuse "zero mentions" with "negative sentiment"—the latter is a different problem.

When This Problem Becomes a Different Challenge

If you start appearing in AI responses but AI always recommends competitors first, that's no longer a "zero mention" problem. GenScore categorizes this as another stage of GEO: not "whether you appear" but "where and in what order you appear"—this requires a different set of strategies and measurement systems.

For localized businesses and professional service firms, GEO's starting logic has its own unique dimensions, including geographic signals and weighting differences in local directories, which deserve separate planning.

From "Whether Mentioned" to "Ranking and Recommendation Priority"

Once a brand crosses the zero-mention stage, GEO's core metrics shift. The zero-mention stage focuses on recall—whether AI knows you exist; the next stage focuses on ranking weight—where you rank in AI-generated recommendation lists and how you're described.

Factors affecting ranking include: the depth of your brand's specialized content in a vertical domain (AI tends to recommend brands that "seem to really understand this field"), the recommendation density from authoritative third-party sources, the sentiment of user reviews, and consistency of brand information across platforms. A common trap is: a brand finally gets mentioned by AI, but the description stays at a superficial level ("also has a similar product"), while competitors are described as "industry-leading." This requires targeted content strategy to change AI's "impression" of your brand positioning, rather than simply increasing mention frequency.

Start from the Baseline, Measure Real Change

The GenScore team states that their content is based on real scan data from their own tools in operation, and they clearly flag claims lacking evidence. For new brands, the first step isn't pursuing a perfect AI mention rate, but honestly measuring your starting point, then building progressively according to controllability priorities. There are no shortcuts in the 0-to-1 process, but there is a clear path.

One-line Summary

A new brand's zero mentions in AI search is a normal starting point—build GEO visibility in three steps ranked by controllability.

Paragraph Summary

The article points out that when new brands ask AI like ChatGPT and get zero mentions, it's not AI hostility—it's that models can only reference what they've seen in training data or real-time retrieval. GEO (Generative Engine Optimization) should start with three directions ranked by controllability: first, refine your website's brand entity information so AI crawlers and visitors can quickly understand your positioning; second, publish an llms.txt file to provide AI with structured content navigation; finally, acquire genuine, verifiable external mentions—the slowest but most important step. The article emphasizes there are no shortcuts—buying mentions doesn't work, and zero mentions is different from negative reviews. A first scan showing 0 results should be treated as a measurement baseline, with re-scans every few weeks to measure real progress. When a brand starts being mentioned but ranks poorly, the problem shifts to a ranking optimization challenge.