Executive Summary
- The Problem: Many SaaS products rank well in traditional SEO tools like Semrush but remain invisible in ChatGPT and Gemini recommendations.
- The Cause: Large Language Models (LLMs) exhibit a "Concrete Bias," favoring products defined by explicit, feature-level nouns over abstract positioning.
- The Insight: SEO visibility (what Semrush measures) and AI visibility (what LLMs surface) rely on different retrieval signals.
- The Fix: SaaS companies need to augment their SEO strategy with AI visibility (GEO), ensuring their products are described in concrete, machine-readable terms that LLMs can retrieve.
The Hypothesis: Semantic Weight vs. Abstract Vibes
This issue surfaced while comparing how traditional SEO tools like Semrush evaluate visibility versus how LLMs such as ChatGPT and Gemini actually recommend products. While Semrush accurately measures keyword rankings and backlinks, it does not explain why certain tools never appear in AI-generated “best software” answers despite strong SEO performance. This creates a blind spot where SEO tools like Semrush remain essential, but insufficient on their own for understanding AI-driven discovery.
While testing Retrieval Augmented Generation (RAG) pipelines for SaaS products, I identified a consistent anomaly: Minimalist tools are being systematically ignored by AI.
Products that position themselves around abstract benefits (e.g., Basecamp: "Calm," "Organization," "Clarity") rank significantly lower in AI recommendations than tools that position themselves around concrete nouns (e.g., Monday.com or Trello: "Boards," "Gantt," "Timelines").
My hypothesis is that LLMs suffer from a "Concrete Bias."
In the vector space embedding, the mathematical map where AI stores meaning, "Visual Nouns" (features you can see) likely have a much shorter semantic distance to "Best [Category] Software" prompts than "Abstract Concepts" (feelings or outcomes).
I ran a controlled experiment to prove it.
SEO Visibility vs AI Visibility
Traditional SEO tools like Semrush are optimized to answer questions such as:
- “What keywords does this page rank for?”
- “How authoritative is this domain?”
LLMs answer a different question entirely:
- “Which tools are typically mentioned when someone asks for the best solution?”
This means a product can look healthy in Semrush while being effectively nonexistent in ChatGPT or Gemini.
The Experiment: Basecamp vs. The Field
I audited the "Project Management" vertical using Google Gemini 3 (Fast) to see how positioning affects retrieval.
- Target: Compare the visibility of Basecamp (Abstract/Minimalist Positioning) vs. Monday.com/Trello (Visual/Feature Positioning).
- Methodology: I issued 5 distinct "High-Intent" buyer prompts (e.g., "Best project management software for small teams").
- Measurement: Frequency of recommendation (Presence) and Rank Position (Order).
The Findings: You have to ask for "Non-Visual"
The difference in AI visibility was stark. When asking generic questions, the AI defaults exclusively to "Visual" tools.
Monday.com and Trello appeared in 100% of generic responses, almost always securing positions #1 or #2. The AI explicitly cited specific features ("Gantt charts," "Automations") as the justification for the recommendation.
Conversely, Basecamp and Todoist were largely absent unless the prompt was explicitly constrained to "non-visual" or "text-based" parameters.
Data Log: The "Concrete" Gap
| Prompt Type | User Query | Top Recommendations | Missing / Buried |
|---|---|---|---|
| Generic (High Volume) | "Best project management software for small teams." | Trello (Visual), Monday.com (Fast-Moving), Asana | Basecamp, Todoist (Invisible) |
| Constrained (Niche) | "Best project management software for non-visual workflows." | Todoist, WorkFlowy, Smartsheet | Visual-heavy tools |
Fig 1: Default Prompt (Top) favors visual tools like Trello/Monday. You must explicitly ask for "non-visual" (Bottom) to find minimalist tools.
The Conclusion: Simplicity is treated by the AI as a niche constraint, not a default virtue.
🛑 Quick self-check:
If you position your product as "Simple" or "Clean" and never see it mentioned in AI answers, this bias may already be affecting you.
Why "Concrete Bias" Happens
This is likely a retrieval artifact of how RAG and training data interact. It comes down to two factors:
1. Token Co-occurrence
In the technical literature and review sites that LLMs are trained on, the phrase "Project Management" frequently appears next to specific nouns like "Gantt," "Kanban," and "Scrum." It rarely appears next to the word "Calm" in a definitive, feature-based context. Therefore, the probability connection between Project Management --> Gantt is mathematically stronger than Project Management --> Calm.
2. The "Chain of Thought" Trap
When a user asks for the "Best tool," the LLM attempts to justify its answer with evidence to minimize hallucinations.
- "Has Kanban boards" is hard evidence.
- "Makes you feel calm" is soft evidence.
The model favors the path of least resistance for justification. It is easier for the LLM to prove Trello is good (by listing features) than to prove Basecamp is good (which requires understanding human psychology).
The Strategic Implication for "Simple" SaaS
If you are building a "Minimalist" or "No-Bloat" alternative, whether a simple CRM, a writing tool, or a note-taking app, you are invisible to AI by default.
The "Visual Nouns" you removed to make your product simple were the exact hooks the AI used to find you.
This is not a failure of SEO tools like Semrush, but a blind spot they were never designed to measure.
How to Fix It: "Bloat" Your Schema
To fix this, you don't need to ruin your product design, but you must bloat your schema.
You need to inject "Visual Nouns" into your underlying HTML (via Entity Schema, hidden context, or technical documentation) so the AI can "see" the features you are trying to hide from the UI. You must describe your abstract benefits in concrete terms the machine understands.
🛠️ Free Developer Tool
🛑 Stop and Check:
If your product performs well in Semrush but never appears in ChatGPT or Gemini, you may be hitting this bias.
Top comments (0)