So, what’s the best coffee shop in New York?

(And why “best X” content is quietly rewiring AI search.)

Dec 05, 2025

The Future of Search newsletter is the #1 newsletter on SEO.

Or so, me and my mom think.

If you’ve ever seen Elf, you know the vibe here. Buddy walks through Manhattan, spots a sketchy little café with a glowing “WORLD’S BEST CUP OF COFFEE” sign, and explodes with joy.

That sign is the web in one frame.

Everywhere you look it’s:

Best CRM software
Best AI tools for marketers
Best coffee shop in New York

And now, thanks to LLMs, those “world’s best…” signs aren’t just fooling tourists; they’re training the systems that tell the rest of us what’s “true.”

Recently, Ahrefs published a study on how “best X” lists show up in ChatGPT’s answers, looking at tens of thousands of URLs and the AI platforms that cite them. That’s the backdrop for this newsletter.

Let’s unpack what that means for “truth,” taste, and industry leadership.

The age of “best X” everything

Across the sites Ahrefs tracks, ChatGPT is now sending 8–9x more referral traffic than the next AI-first platform, Perplexity. Sam Altman says more than 800 million people use ChatGPT each week.

So where do all those answers come from?

In that study, when ChatGPT answered top-of-funnel queries like “best CRM for enterprise” or “best home treadmill,” “best X” blog lists were the single most common page type cited — 43.8% of all sources (Ahrefs).

Nearly half the truth-layer for these questions is… listicles.

They’re not static, either. Out of 1,100 “best” lists with clear dates, 79.1% had been updated in 2025, and about a quarter had been updated in just the last two months.These aren’t blog posts; they’re living documents, constantly tweaked to stay top-of-mind for both search engines and AI.

And yet, a big chunk of this apparent “truth” is on pretty shaky ground. When Ahrefs looked at domain authority for 3,000 cited “best” lists, 35% were on low-authority domains—often sites that look like they exist more for citations and link-building than actual humans (Ahrefs).

So when you ask:

“What’s the best coffee shop in New York?”

You’re really asking:

“Which combination of fresh, structured, and frequently-referenced listicles do today’s AI systems currently believe the most?”

Not exactly the same thing.

Frequency = “truth” (for LLMs and for us)

I’ve written before about this: for LLMs (and honestly, for humans), the more something appears, the more “true” it feels.

Large language models, explained with a minimum of math and jargon — Whatever you feed in LLM, it will spit out.

LLMs don’t have beliefs. They don’t “know” the best coffee shop.

They:

Ingest billions of words
Notice patterns (“these names show up together a lot in ‘best X in NYC’ contexts”)
Predict the next likely token

If the web has:

Thousands of “best X in New York” lists
Many repeating the exact same 8–15 places
And those lists are well-structured and updated constantly

…then those places become gravity wells in the model’s internal universe.

Ahrefs found that brands ranking higher within these lists show up more often in ChatGPT’s recommendations.

Some of that is just math: more lists, more mentions, more weight. But the effect is the same:

The model doesn’t find “the best.” It averages what’s most repeatedly claimed to be the best.

That’s exactly how “WORLD’S BEST CUP OF COFFEE” ends up in the training data.

LLMs are not very intelligent… and that’s the point

We keep talking about LLMs like co-workers:

“It prefers the top of the page.”
“It knows which tools are better for small teams.”
“It’s biased toward certain brands.”

But what is an LLM, really?

A giant probability engine
A lossy compression of everything we’ve published
A pattern-matcher that’s terrifyingly good at sounding intelligent

When you ask:

“What’s the best coffee shop in New York?”

It’s not walking the streets of Brooklyn with a tasting journal. It’s:

Pulling from training data full of “best NYC coffee shop” posts, reviews, and lists
Mixing that with whatever it can retrieve right now
Spitting out a fluent summary that sounds confident enough to believe

This is why truth media vs. fake media matters so much.

Truth media vs. fake everything

We’re surrounded by generated everything:

UGC farming engagement
Thin “best tools” roundups covered in affiliate links
AI-spun forum posts that mimic real discussion
News and reviews written for bots first, humans second

From the model’s perspective, all of this is just “text with patterns.”

So, I like the idea of truth media:

Content that:

Makes its incentives obvious
- “This is our list. We’re on it. Here’s why and who it’s for.”
Shows its work
- Methodology, criteria, trade-offs
Comes in a structure that’s easy for both humans and models to parse

Interestingly, Ahrefs found that when ChatGPT recommends a brand, it often cites internal landing pages more than homepages or generic blog content, especially for software and agencies.Those pages tend to be tightly focused, clear about what’s on offer, and aligned with a specific intent.

So the AI ecosystem isn’t just powered by listicles. It’s powered by:

Comparison content (who’s on the list, who ranks where)
Category explanations and landing pages (what this thing is, who it’s for)

That’s exactly where industry leadership comes in.

There is no “best coffee shop in New York”

First, let’s admit something obvious: there is no objective best coffee shop in NYC.

Even if you had:

Blind cuppings
Extraction profiles
Barista Olympics

You’d still run into wildly different definitions of “best”:

The freelancer wants plugs, seats, and quiet
The commuter wants “close to my train and fast”
The coffee nerd wants single-origin pour-overs and zero laptops
The parent wants stroller space and snacks

All of that gets collapsed into one unqualified, head-term query:

“Best coffee shop in New York”

That’s how most AI search works today: it tries to answer a universal question that should be deeply personal.

What we actually mean is:

“What’s the best coffee shop in New York for me?”

Which leads to the real story:

Personalization is the real “best X”

Right now, most AI answers are still universalized:

One list
Some pros/cons
A generic “it depends on your preferences” paragraph

But the obvious next step is personalization:

“Best coffee shops in New York if you need Wi-Fi, lots of seating, and you’re near Brooklyn”
“Best coffee shops in New York if you care more about beans than vibes”
“Best coffee shops in New York if you’ve got 20 minutes before a train at Penn Station”

LLMs are perfectly capable of doing this if they have:

Enough structured, specific content
A clear sense of who the user is and what they care about
Categories that map cleanly onto real-world trade-offs

This is where your content strategy has to move beyond “we’re on the list” into “we define the category.”

Specificity is the new king

If generic “best X” lists are the new link farms, then specificity is the new moat.

The Ahrefs data suggests a couple of things at once:

Being in “best X” lists (especially near the top) correlates with more AI recommendations.
AI assistants clearly favor fresh, clearly formatted content.
A lot of current sources are low-quality — which means there’s room to raise the bar.

So instead of playing the same volume game as everyone else, the smarter move is:

Hyper-specific lists
- “Best CRMs for solo consultants under $100/month”
- “Best ecommerce agencies for $2–10M brands that care about content-first growth”
Opinionated rankings
- Not 10 equal options, but 2–4 that you strongly argue for, with honest drawbacks
Structured explanations
- Clear sections, comparison tables, scenario-based recommendations — the stuff both humans and LLMs can easily digest

This doesn’t just make you useful. It sets you up for the next level of industry leadership.

Industry leadership: own your category and justify it

Here’s the shift I’d make in how we talk about being an “industry leader” in an AI-search world:

It’s not just showing up on “best X” lists.
It’s defining a category, dominating it, and making a compelling case for why that category is one of the best paths in the entire market.

Two steps:

1. Own your category

You don’t want to be just another name in “best project management tools.”

You want to be:

“The best project management tool for agencies that bill by the hour”
“The best email platform for privacy-first B2B teams”
“The best coffee shop for remote workers in Williamsburg who hate loud music”

That means:

Creating pages that clarify who you’re for and who you’re not for
Using structure that maps neatly to user needs (industry, team size, budget, workflow)
Being consistent across your site so LLMs see a clear, strong signal:
“This brand = this category = these problems.”

When AI assistants need to slot you somewhere in their answer, you’ve already done the work.

2. Argue why your category is one of the best options overall

Owning a niche is great, but in AI answers you’re often compared alongside totally different approaches:

“Do I hire an agency or buy a tool?”
“Do I use an all-in-one suite or a specialized stack?”
“Do I pick the cheap option or the flexible one?”

Industry leadership means you’re not just saying:

“We’re the best X.”

You’re saying:

“Here’s why this whole category (that we happen to lead) is one of the best ways to solve your problem compared to the other categories you’re considering.”

That looks like:

Explaining the trade-offs between categories, not just vendors
- Agencies vs in-house vs platforms
- All-in-one vs best-of-breed
- Physical vs digital solutions
Producing content that helps users decide:
- “When this category is right for you”
- “When you should pick something else” (yes, even if that means not choosing you)
Giving LLMs clean, quotable arguments:
- “This approach works best when…”
- “The main downside of this category is…”

You become the source of record AI systems can lean on when they need to explain why your approach is good… not just that your brand exists.

That’s a much stronger position than being one more name in a 27-item listicle.

So what do you actually do?

If you care about being visible in AI search when someone types “best X,” here’s the distilled playbook:

Have at least a few “best X” assets — but make them real.
- Clear audience, clear use case, clear trade-offs.
- If you include yourself and rank yourself highly, explain why like a grown-up, not a window sign.
Specialize your lists and pages.
- Move from “best tools for marketers” to specific segments, budgets, and situations.
- Align with how real users and buyers actually think about their choices.
Treat your category pages and landing pages as truth media.
- State who you’re for.
- Explain the category, not just the product.
- Compare categories honestly so LLMs have something nuanced to pull from.
Update relentlessly.
- Build a maintenance habit for your key comparison and category pages.
- Freshness is clearly rewarded in both AI and traditional search.
Aim for leadership, not just mentions.
- You want to be the brand that defines how your category is understood,
  and the one with the best argument for why that category deserves a seat at the “best overall solution” table.

Because in the end, there is no single “best coffee shop in New York.”

What we actually want and what AI search is slowly evolving toward is:

“Given who I am, where I am, and what I care about…
What’s the best option for me, and why this kind of option at all?”

Until then, we’re living in the Buddy-the-Elf era of AI search: lots of glowing signs, not a lot of context.

Our job as SEOs, marketers, founders is to build the kind of truth media that helps models (and people) tell the difference.

-Aaron

Discussion about this post

Ready for more?