The Agent Economy Is Here. Your Brand Is Probably Invisible In It

AI agents are starting to spend money. They discover services, evaluate them, and buy, all in milliseconds. I tracked 2,014 companies across ChatGPT, Claude, and Gemini. 85.7% score near zero on AI visibility. If agents can't find you today, they won't buy from you tomorrow.

Want to see where your brand stands? Run a free report at loamly.ai/check.

---

Garry Tan Just Asked the Right Question

Last week, YC president Garry Tan published a piece asking: "How do I make sure agents know about my product and service and choose it?" (Garry Tan on LinkedIn)

His answer: discovery has to be programmatic. Agents don't browse. They query. Your marketing site is invisible to an agent at runtime.

He's right. But here's the part most people are missing: the agent discovery layer isn't some future thing to build. It already exists. It's the same AI recommendation layer that ChatGPT, Claude, and Gemini use right now when a human asks "what's the best tool for X?"

I have data on how 2,014 companies perform in it.

85.7% of Companies Are Already Invisible

Across 2,014 companies in Loamly's dataset, 85.7% (1,726 companies) score between 0 and 20 on AI visibility. The top 5.2% score 80+. (Full benchmark)

AI Visibility Tier	% of Companies	Avg Score
Emerging (0-20)	85.5%	1.6
Growing (20-50)	5.8%	33.2
Established (50-80)	3.5%	62.7
Leaders (80+)	5.2%	93.4

There's almost nothing in between. You're either visible or you're not. The bar is on the floor right now, which means the window to establish presence is wide open.

This matters for the agent economy because AI agents use the same underlying models (GPT-4, Claude, Gemini) to discover and evaluate services. If ChatGPT doesn't recommend your brand to a human today, an agent built on the same model won't find you tomorrow.

---

The Three Platforms Don't Agree

I asked ChatGPT, Claude, and Gemini the same questions about the same 2,014 brands. Their citation rates are surprisingly different.

Platform	Avg Citation Rate
ChatGPT	70.4%
Claude	63.9%
Gemini	57.6%

Same queries. Different recommendations. Claude has the highest variance too, meaning it's the most unpredictable in who it recommends.

For agent commerce, this means your service needs to be visible across all three model ecosystems, not just one. An agent built on Claude will discover different services than one built on GPT-4.

---

Brand Authority Beats Technical SEO by 5x

Everyone in the GEO space is talking about optimizing for AI search. Schema markup, structured data, keyword placement.

The data tells a different story.

Factor	Correlation with AI Visibility
Brand Authority	0.386
GEO Score (technical)	0.080

Brand authority correlates 5x more with AI visibility than technical optimization does.

Concrete example from the dataset: one company has a GEO score of 39 with zero schema markup and achieves 100% ChatGPT visibility. Another has a GEO score of 88 with perfect schema and sits at baseline. Technical optimization doesn't create visibility. Brand authority does.

For agents making purchasing decisions, this tracks with what Garry describes. Agents optimize for reliability and confidence. Brand authority is how AI models currently measure both.

---

The Discovery Gap: Recognized But Not Recommended

The most consistent finding across the entire dataset:

Metric	Average Score
Brand Recognition	0.615
Category Recommendation	0.092
Discovery Gap	0.515

97% of companies (1,378 of 1,423 with category data) score lower on category recommendation than brand recognition. (Full discovery gap analysis)

AI platforms know who you are. They just won't recommend you when someone asks "what's the best tool for X?"

In agent commerce, this is the difference between existing and being purchased. An agent doesn't care if it recognizes your brand. It cares if you show up when it queries for a solution.

---

Your AI Traffic Is Already Here. You Just Can't See It

The agent economy isn't purely future tense. AI-referred traffic is already hitting your site. You're almost certainly not tracking it correctly.

When someone asks ChatGPT for a recommendation and clicks through, GA4 labels it as "direct." ChatGPT strips the referrer header. GA4 sees the visit but can't attribute the source.

Across the sites I've tracked: roughly 2.4x more AI visits hide in "direct" than GA4 correctly labels. The numbers: 14,413 dark AI visits vs 6,015 visible ones.

Traffic Type	Visits	Conversion Rate
Dark AI (hidden in Direct)	14,413	10.21%
Visible AI (correctly labeled)	6,015	10.21%
Non-AI traffic	—	2.46%

AI-referred traffic converts at 4.15x the rate of regular traffic. The visitor asked an AI a question, got your brand as the answer, and clicked through. Warm lead by definition.

Now scale this to agents. When an agent evaluates your service, it's the same pattern: query the model, get a recommendation, act on it. The difference is speed: milliseconds instead of 30 seconds.

The same AI recommendation layer that drives human traffic today will drive agent purchasing decisions tomorrow. If you can't measure AI traffic now, you're blind to the channel that agents will use to find you.

---

Bot Traffic Is About to Become Important

AI bot traffic grew 6,900% year-over-year in 2025, according to HUMAN Security. 48.3% of all AI bot traffic comes from crawlers that bundle search indexing with AI training.

Right now most companies either ignore bot traffic or try to block it. That's about to flip.

In the agent economy:

• AI crawlers indexing your content = agents being able to discover you
• Blocking GPTBot = making yourself invisible to OpenAI's agent ecosystem
• Not tracking which bots visit = not knowing which agent platforms can find you

I built Loamly specifically to track this. Six detection methods for identifying AI-referred traffic that GA4 misses. Dark AI detection that separates genuine direct visits from AI-stripped referrals.

The companies that start measuring agent and AI traffic now will have months of baseline data when agent commerce scales. The ones that don't will be guessing.

---

What Garry's Checklist Looks Like In Data

Garry's piece ends with a checklist for making your service agent-native. Here's how each item maps to what I see in the data:

Garry's Requirement	What the Data Shows
Machine-readable capabilities	85.7% of companies have near-zero AI visibility. Most aren't discoverable in the current recommendation layer
Provable reliability	Brand authority (0.386 correlation) is the strongest predictor. AI already uses reliability signals
Faster and cheaper than self-computation	AI traffic converts 4x higher. The model already "trusts" certain brands. That trust transfers to agent decisions

The gap between "agents need to discover you programmatically" and "you're invisible to the models agents use" is the opportunity.

---

What to Do About It

If you're thinking about the agent economy, here's what the data says matters:

1. Check your current AI visibility. Before building agent-native APIs, find out whether AI models recommend you at all. If you score below 20 (like 85.7% of companies), that's the first problem to fix. Check at loamly.ai/check. Free, no signup.

2. Track your AI traffic. If you don't know how much AI-referred traffic you're getting, you can't measure whether efforts are working. GA4 won't catch it. You need detection that goes beyond referrer headers.

3. Build brand authority, not just technical SEO. The correlation is 5x stronger. Get mentioned in the places AI models trust: Reddit (50+ mentions correlates with 22.5 avg visibility vs 1.4 for zero), Wikipedia (24.0 vs 6.9 avg visibility), and established publications.

4. Don't block AI crawlers. Every blocked bot is a closed door to an agent ecosystem. Track which bots visit, which pages they hit, and how often. That's your future distribution channel.

5. Monitor all three model ecosystems. ChatGPT, Claude, and Gemini disagree on who to recommend. Optimize for one and you might be invisible on the others.

---

The Window Is Now

The agent economy infrastructure is production-ready. Stripe's x402 protocol has processed 20 million+ agent payments. MCP was donated to the Linux Foundation with backing from OpenAI, Google, Microsoft, and Amazon. Ali Ghodsi at Databricks says 80% of new databases are created by agents, not humans (Databricks Data + AI Summit).

But the discovery layer, the part where agents figure out which service to buy, is still being built on the same AI recommendation models that power ChatGPT and Claude today.

85.7% of companies are invisible in that layer right now. The window to establish presence before it calcifies is open, but it won't stay open.

I'm tracking this at Loamly. AI visibility monitoring, dark AI traffic detection, bot analytics. Real data about where you stand in the layer that agents will use to find you.

Want to see where your brand stands? Check at loamly.ai/check. No signup required. Real data. No marketing spin.

---

Methodology: Analysis covers 2,014 companies tracked across ChatGPT, Claude, and Gemini (January 2026). AI visibility scores range 0-100 based on citation frequency, recommendation likelihood, and response positioning. Traffic data from Loamly's dark AI detection across active workspaces. Brand authority scores combine domain authority, backlink profiles, social mentions, and platform presence signals. Limitation: Perplexity data excluded due to collection issues in this dataset.