Welcome to the Winner-Take-All Reality of “AEO”
Let me tell you exactly what happened.
Last week I needed email inboxes for our AI agents. QBee, 10K, and the rest of the SaaStr agent stack all need to send and receive email, and Gmail is not built for agents.
I asked Claude. Claude gave me a short list. The #1 recommendation was AgentMail. YC S25. $6M seed from General Catalyst, Paul Graham, and Dharmesh Shah. Python SDK, TypeScript SDK, webhooks, SPF/DKIM/DMARC handled.
I signed up. I’m a paying customer. I never even looked at #2. I never looked at #3 either.
That right there is the entire game. If you’re still thinking about AEO the way you thought about SEO, you’re going to miss it.
SEO Was a Casino. AEO Is a Lottery With One Winner.
For 20 years, B2B buyers started on Google. Google gave you 10 blue links. Users scanned 3 to 5 of them before clicking. If you ranked #3, you still got real traffic. If you ranked #7, you still showed up on the first page. The top position captured the most clicks, but everyone in the top 10 got something.
That’s not how AI search works. Not even close.
When a buyer asks ChatGPT or Claude or Perplexity a recommendation question, they get 1 to 3 answers. Not 10. And most of them try the first one, period. The funnel isn’t “scan and compare.” It’s “take the recommendation.”
The data is now backing this up hard:
- AI search traffic converts at 14.2% compared to Google organic’s 2.8%, with Claude users converting at 16.8%, ChatGPT at 14.2%, and Perplexity at 12.4%
- Claude users show the highest conversion because Claude cites fewer sources, making each click-through more deliberate
- 73% of B2B buyers now incorporate AI tools into their research process
- 37.5% of ChatGPT usage is “generative intent.” Users aren’t searching. They’re asking AI to draft vendor comparisons, build shortlists, create evaluation frameworks
- Vercel reported that ChatGPT now refers approximately 10% of new user signups, up from 1% six months earlier
Read that conversion number one more time. AI search visitors convert at roughly 5x the rate of Google organic. And Claude specifically, which returns the shortest, most decisive recommendation lists, converts the highest of all.
The reason is exactly what happened to me with AgentMail. The AI already did the comparison. It already filtered. It already ranked. By the time the buyer lands on your site, they’re not shopping anymore. They’re buying.
Position Bias Is Real. And It’s Brutal.
Most founders building for “AEO” don’t realize this yet: LLMs have a documented, measurable bias toward their own first recommendation.
Published academic research shows that LLM-based recommendation models suffer from position bias, where the order of candidate items in a prompt can disproportionately influence output. Items that appear earlier are more likely to be favored. And when the same model is asked follow-up questions, it reinforces its original ranking.
In plain English: the model itself is wired to double down on whatever it already decided is #1. If you get surfaced first, the model’s own reasoning reinforces that choice when the user asks follow-up questions. If you’re #3 in the first answer, you get demoted to “also mentioned” in the second answer, and invisible by the third.
Combine this with human behavior (the user stops at #1 if #1 looks good enough) and you get something qualitatively different from SEO. You get winner-take-all dynamics on a per-query basis.
What Made AgentMail the #1 Answer
Let me break down why Claude put AgentMail at the top of the list for “email API for AI agents.” … per Claude:
1. They own the category phrase. “Email inboxes for AI agents, like Gmail does for humans.” One sentence. That exact phrase matches the query intent. No competitor in adjacent categories (email APIs, transactional email, inbox providers) is using agent-native language that precisely.
2. They have YC S25 plus the right investors. YC companies generate massive corpus presence across Hacker News, YC’s own writeups, founder interviews, and alumni lists. Add Paul Graham, Dharmesh Shah, and General Catalyst and you have thousands of cross-referenced mentions across the tech web. LLMs treat this as authority signal.
3. Their pricing is public and their docs are open. $20/month dev plan, $100/month starter plan, clear API surface with Python and TypeScript SDKs. Compare to half the “enterprise only, contact sales” stacks that LLMs literally cannot evaluate because there’s nothing for them to read. If you lock your product behind a login, you’ve removed yourself from every AI recommendation in your category.
4. They nailed freshness. The funding round was recent. The company was in Claude’s retrieval window. The signals were fresh, loud, and clustered. That matters a lot more than most people realize for new categories.
The Winner-Take-All Math for Founders
Here’s the part that should scare every B2B founder reading this.
In the old model, you could be the 4th best project management tool, rank on page 1 of Google for “project management software for small teams,” and still build a $50M ARR business on the scraps. You were a participant in an attention market.
In the new model, being the 4th best in your category is closer to being invisible. The buyer asks Claude. Claude says “use X.” The buyer tries X. If X is good enough, the buyer never runs the second query.
A few uncomfortable implications:
- Category position is now worth 10x what it was. Being #1 in “email infrastructure for AI agents” is worth vastly more than being #4 in “email APIs.” Pick narrower categories. Own them outright.
- Brand salience compounds. The more you’re cited in fresh, authoritative sources, the more you become the default answer. The more you become the default answer, the more you get used, reviewed, and cited again. It’s a flywheel that rewards early wins and punishes late entries.
- Copycats are in a worse spot than ever. In a Google world, a “cheaper alternative to X” could rank on comparison keywords and siphon 10% of X’s demand. In an LLM world, the model just recommends X. The alternative doesn’t get surfaced because the user never asked for one.
- Platform fragmentation is real. Only 11% of domains are cited by both ChatGPT and Perplexity, per a 680 million citation study. Only 12% of cited sources match across ChatGPT, Perplexity, and Google AI. You can be the #1 answer in Claude and invisible in ChatGPT. Win every relevant platform, not just one.
If Your AEO Tool Just Gives You a Score, It’s Useless.
There’s a whole category of AEO tools launching right now that will sell you a “visibility score” or a “brand presence index” or an “AXO rating.” They’ll tell you your AI visibility went from 34 to 47 this month. Congratulations.
Ignore all of this.
A generic visibility score is a vanity metric. It averages across queries that don’t matter and drowns out the ones that do. Your “AI visibility” going up 10% means nothing if the queries that got better are ones no buyer actually types, and the 20 queries that drive 90% of your pipeline are still returning your competitor as #1.
This is exactly where classic SEO lived for 15 years, just on steroids. In SEO you didn’t measure “search visibility.” You measured your ranking on the specific keywords that drove revenue. “Email API pricing” mattered. “What is email” did not. The discipline was picking the right 50 queries and obsessing over your position on each one.
AEO is identical. Except now the queries are longer, the universe is bigger (ChatGPT, Claude, Perplexity, Gemini, AI Overviews all behave differently), and the stakes on each query are higher because there’s only one winner per answer.
Here’s what actually matters:
- The 20 to 50 specific queries your buyer types when they’re in the buying moment. Not generic category terms. The actual prompts: “best email API for AI agents,” “alternatives to SendGrid for agent workflows,” “how do I give my AI agent an inbox.”
- Your #1 rate on those specific queries across all 4 major LLMs, measured weekly.
- The delta between you and whoever is currently #1 on queries where you’re not.
- The co-mention graph on queries where you show up but not as #1. Who keeps getting recommended alongside you or above you? That’s your real competitive set.
Everything else is noise. If an AEO product can’t tell you “you are #1 on 7 of your 20 target queries in Claude, 3 in ChatGPT, 0 in Perplexity, and here’s the exact content and source gap to close on each one,” it’s not an AEO product. It’s a dashboard.
What to Actually Do This Quarter
If you’re a B2B founder, stop optimizing for “AI visibility” and start optimizing for “AI #1.”
- Pick your 20 to 50 buyer queries. Not category terms. The specific prompts your buyer types when they’re ready to spend money. Write them down. Share with your team.
- Run the actual queries. Go ask ChatGPT, Claude, Gemini, and Perplexity each one. Write down the #1 answer. If it’s not you, that’s your real competitive set. Not who you think your competitors are.
- Narrow your category until you win it. If you’re not #1 in “project management software,” are you #1 in “project management for remote engineering teams on Linear”? Keep cutting until you find the exact query where you are, can be, or should be the default answer. Own that.
- Make your positioning statement a single quotable sentence. AgentMail did this with “email inboxes for AI agents, like Gmail does for humans.” If a model can’t repeat your positioning verbatim, it won’t.
- Open everything machine-readable. Public pricing. Open docs. Clear product surface. If your homepage says “contact us for pricing” and your docs are behind a login, you’ve voluntarily withdrawn from the AEO game.
- Seed the corpus with authority signals, not content mill slop. A single mention in a YC writeup, a Paul Graham tweet, a well-sourced TechCrunch piece, a detailed Reddit thread with real operator voices. Each of these is worth 100 SEO blog posts written for Google. LLMs weight source authority heavily.
- Track your #1 rate, not your visibility score. On those 20 to 50 specific queries, across all 4 major LLMs, weekly. That’s the only AEO number that matters. If you’re at 20% today, get to 50% by end of quarter. Then 80%. Then own the category.
This Will Only Accelerate
The question for the next decade of B2B isn’t “can buyers find you.” It’s “when the AI answers their question, are you the answer.”
Not in the answer. The answer.
AgentMail got me as a paying customer last week because Claude said use them, and I did, and I never bothered with the backup options. Every B2B founder should sit with how much of their pipeline is about to work exactly that way. Because mine already does.
Go ask Claude what the best company is in your category. Right now. And ask it 4-5 different ways.
If the answer isn’t you, that’s the work.
