Amazon has more money than god. It has Prime, warehouses in every zip code, a delivery fleet that could probably invade a small country, and a grocery arm — Amazon Fresh — that's been aggressively expanding since 2007. Instacart is a marketplace that doesn't own a single truck.
Guess which one the AI recommends.
We ran a head-to-head brand fight on our /vs tool: Amazon Fresh vs Instacart. Five buyer-shaped prompts, three frontier models, one winner per model. The kind of question a real person types when they're deciding what to sign up for. "Best grocery delivery to buy right now." "What's the best service for weekly grocery orders." That kind of thing.
The result was cleaner than we expected.
The Verdict: INSTACART — DOMINATION
Instacart won 3-0. ChatGPT picked it. Gemini picked it. Perplexity picked it.
The raw mention counts were tied inside every single model — five for Amazon Fresh, five for Instacart across the prompt set on each. But mention count isn't the same as recommendation. When each model was asked to name the best option, or to lead its list, Instacart got the top slot every time. Amazon Fresh got the "also consider" slot.
| Model | Amazon Fresh mentions | Instacart mentions | Pick |
|---|---|---|---|
| ChatGPT | 5 | 5 | Instacart |
| Gemini | 5 | 5 | Instacart |
| Perplexity | 5 | 5 | Instacart |
For a category as commoditized as grocery delivery — where 90% of users will try the first name they hear and never switch — being second on the list is being nowhere.
Why Instacart Wins the Prompt
Before we tear into what Amazon Fresh should do, it's worth understanding why the models leaned the way they did. We can't crack open the training corpuses. But we can reason about the signal.
Instacart is a category-defining brand. Amazon Fresh is a feature of Amazon.When you ask a language model "what's the best grocery delivery service," you're triggering a retrieval pattern. The model reaches for the noun most associated with the category. In grocery delivery, that noun is Instacart. It's been the default reference point in tech press, VC coverage, and consumer reviews for a decade. Amazon Fresh, by contrast, is usually mentioned as one of many things Amazon does — sandwiched between Amazon Basics and Amazon Pharmacy in some listicle about Prime perks.
That framing bleeds into the model. Instacart is a grocery delivery company. Amazon Fresh is Amazon's grocery delivery feature. Guess which one wins a "best of category" prompt.
Review corpus depth. Instacart has been reviewed thousands of times as a standalone service. Wirecutter, NYT, Reddit threads, comparison articles — the review universe treats it as the reference product. Amazon Fresh reviews are thinner and often nested inside broader Amazon coverage. The reviews that do exist tend to be lukewarm ("fine if you already have Prime"), which is a nightmare for AI retrieval. Models don't reward brands that inspire shrugs. The AI isn't picking Instacart because it's better. It's picking Instacart because the internet talks about Instacart as if it were the answer. Breadth vs depth of retailer selection. Instacart's pitch is that it delivers from your existing local stores — Costco, Safeway, Wegmans, ALDI, Publix, whatever. That story is easy to compress into one AI-friendly sentence: "Shop from your favorite local stores." Amazon Fresh's pitch is that you buy from Amazon's own inventory. That's a narrower, less flexible story, and it doesn't play as well when someone asks a broad question like "best grocery delivery." Geographic coverage narrative. Instacart is available in ~14,000 US cities. Amazon Fresh has same-day delivery in a couple hundred metros. Models notice this asymmetry in coverage claims, and they weigh it heavily when the prompt is about "best" rather than "cheapest" or "fastest for Prime members." Recency of PR. This one matters more than most brands realize. Instacart IPO'd in September 2023 and generated a flood of coverage. Post-IPO earnings calls, financial press, executive interviews. Amazon Fresh has been quietly restructuring its physical stores and cycling through leadership. Which brand do you think generated more crawlable, positive-framed English text in the last 24 months?What Amazon Fresh Could Actually Do About It
Generic "invest in content" advice is useless here. Amazon has infinite content. What Amazon Fresh lacks is the right kind of content in the right shape for AI retrieval. Some concrete moves.
Split the brand out of Amazon's shadow. The number one problem is that Amazon Fresh reads as a subroutine of Amazon. In the AI's mental model, it's not a peer of Instacart. It's a sub-page. Every piece of Amazon Fresh marketing that leans on "part of Prime" reinforces this. The fix is to build a corpus where Amazon Fresh is discussed as a standalone service that competes with Instacart, DoorDash grocery, Walmart+, and Kroger Delivery. That means dedicated review outreach, dedicated affiliate programs, dedicated PR angles, dedicated URLs that don't nest under amazon.com/fresh. Seed comparison pages. When someone types "Amazon Fresh vs Instacart" into Google or an LLM, the top results shape how AI models learn to compare them. Right now, most of those comparison pages are written by SEO farms with no relationship to Amazon and no incentive to frame Amazon Fresh favorably. Amazon Fresh should be commissioning long-form, honest comparison content from credible sources — food writers, personal finance sites, mom blogs with real audiences. Not sponsored posts. Editorial reviews with clear disclosure. Fix the review sentiment problem. A lot of Amazon Fresh's public review sentiment is "it's fine, but Instacart has more selection." That framing has been baked into the crawlable web for years. The only way to shift it is to give reviewers something new to say — a new feature, a new coverage area, a new pricing model. Something that generates fresh, positively-framed articles that the next generation of AI training runs will ingest. Show up in "best of" listicles as the lead. Every "best grocery delivery service 2025" article ranks Instacart #1 by default. Not because the writer did deep research — because Instacart is the default. Amazon Fresh needs a coordinated PR push to change the ranking narrative in the outlets that get scraped: Wirecutter, Good Housekeeping, CNET, Forbes Vetted, NYT. These articles are the training data. If you're not in the top slot in those articles, you're not going to be in the top slot in the LLM's answer. Build a Q&A layer optimized for AI. LLMs love structured Q&A content. Amazon Fresh should have a public help center, a public blog, and a set of comparison pages that answer real buyer questions in short, quotable paragraphs. "Is Amazon Fresh cheaper than Instacart?" "Does Amazon Fresh deliver from Whole Foods?" "How does Amazon Fresh compare to Instacart for weekly orders?" Each of these needs a canonical, well-written answer that a model can quote verbatim. Amazon Fresh isn't losing because Amazon is bad at grocery. It's losing because Amazon is bad at being talked about as if it were a grocery company.The Category-Wide Lesson
Amazon Fresh and Instacart are two players. The category also includes Walmart+, DoorDash grocery, Uber Eats grocery, Kroger Delivery, Shipt, Gopuff, and a dozen regional services. If you're running any of them, this brand fight has three lessons you should internalize.
Lesson 1: Standalone brand identity beats parent-company halo. Walmart+ has the same problem as Amazon Fresh. When a model retrieves "best grocery delivery," Walmart+ gets treated as "the Walmart thing" rather than as a peer product. Shipt, technically owned by Target, has done a slightly better job of maintaining its own review corpus and press identity. That's not an accident. It's a positioning choice. Own it. Lesson 2: Mention parity is a trap. In our brand fight, Amazon Fresh and Instacart tied on raw mention count in every single model. If Amazon Fresh's brand team ran a simple share-of-voice analysis, they'd conclude they were doing fine. They're not. Mention count without position weighting is a vanity metric. What matters is whether you're in slot one when the model is asked to pick a winner. Everything else is noise. Lesson 3: The models agree with each other more than you think. ChatGPT, Gemini, and Perplexity are trained on overlapping corpuses. When they agree — as they did here, 3-0 — it means the underlying web consensus is unambiguous. That's actually good news for the brand that's losing. It means the problem is diagnosable and fixable at the corpus level. You don't need to game three different systems. You need to change what the web says about you, and all three models will follow.What Buyer-Shaped Prompts Actually Reveal
One more thing worth calling out. The prompts we tested weren't gotcha questions. They weren't "compare and contrast the operating leverage of asset-light versus vertically integrated grocery models." They were the questions a real person types when they're about to open their wallet.
"Best grocery delivery to buy right now."
"What service should I use for weekly grocery orders."
These are bottom-of-funnel prompts. The user has decided to buy. They're picking a brand. And in every single one of these decision-moment prompts, Instacart got the nod.
For a category leader, this is the whole ballgame. The user isn't going to open five tabs and comparison shop. They're going to sign up for what the AI recommends. If the AI recommends Instacart three times out of three, Instacart wins the customer three times out of three. Amazon Fresh gets to watch.This is why we built the /vs tool the way we did. Head-to-head buyer-intent prompts across all three major models, in a format that takes about five seconds to read. Not because a single brand fight tells you everything. But because it tells you the one thing that actually matters: when someone's ready to buy, which brand does the AI hand them?
Running Your Own Brand Fight
Every category has this dynamic. There's a brand the AI treats as the default answer, and there's a brand the AI treats as the honorable mention. Sometimes the mapping matches market share. Often it doesn't. That gap — between how big you are and how the AI describes you — is where the real opportunity lives.
You can run a brand fight against any competitor on Foxish. Head to foxish.ai/vs and plug in the matchup. You'll get the same three-model breakdown we ran here: raw mention counts, per-model verdicts, and an overall score. Takes about a minute. And if you want the full picture on where your brand shows up across the AI landscape — not just against one competitor, but across every prompt in your category — you can run a free scan at app.foxish.ai/signup. First scan's free, no card required.
Instacart won this round because the internet has been calling it the default for a decade. That's a lead. It's not a moat. The next round starts as soon as someone at Amazon Fresh decides to fight for it.
Start a free scan →