llms.txt Won't Get Your Products Into AI Shopping Results. Your Feed Will.

Somewhere in the last year, "add an llms.txt file" became standard advice in almost every AI-shopping-visibility checklist. It rarely works. In a 12-week server-log study across 83 sites published in 2026, OpenAI's and Anthropic's crawlers requested llms.txt files a combined 16 times, while requesting robots.txt on the same sites more than 7,100 times. A second study, 14 days of logs on a single site, recorded zero llms.txt fetches from any identified AI crawler. This is for anyone managing a product feed who has been told llms.txt is the lever for AI shopping visibility. By the end, you will know which mechanism actually controls whether your products appear in Google AI Mode, ChatGPT, and Perplexity, and what to fix instead.

What llms.txt Actually Is (and Isn't)

llms.txt is a proposed convention, not a shopping or e-commerce standard. Developer Jeremy Howard published the original specification on September 3, 2024, and a version 2 update in August 2026 describes it as a markdown file at /llms.txt: an H1 with the site or project name (the only required element), an optional blockquote summary, free-form descriptive sections, and H2-delimited lists of links to other pages. The intent is to give a language model a short, curated map of a site instead of forcing it to parse full HTML pages.

Nothing in the specification mentions product data, pricing, availability, or shopping feeds. It is a documentation and content-discovery convention, closer to a curated sitemap than to a Google Merchant Center feed or a structured product catalog. That distinction matters, because most advice telling e-commerce teams to publish one treats it as if it were a feed format.

Google Has Said Directly That It Won't Use llms.txt

At Google's Search Central Deep Dive event in Asia-Pacific on July 23, 2025, Google's Gary Illyes stated that Google does not support llms.txt and has no plans to, telling attendees to keep investing in standard SEO to appear in AI Overviews rather than a separate file format. No public Google statement since then has walked that back. Some 2026 coverage has also described a "W3C proposal to standardize llms.txt," which overstates things: the relevant W3C Technical Strategy issue sits at the "Investigation" stage, explicitly introduces no new format, and its last substantive comment dates to May 2025. There is no W3C standard, and Google has been unambiguous about not using the file for ranking or AI Overviews.

What Two 2026 Server-Log Studies Found

Two independent measurements back up the "not being fetched" half of the story with actual traffic data, not just policy statements.

• EZY Research, April 27 to July 19, 2026, across 83 sites from a 556-domain panel: GPTBot requested llms.txt 7 times and ClaudeBot 9 times; PerplexityBot 0 times. The same crawlers requested robots.txt 3,990 times (OpenAI), 3,120 times (Anthropic), and 775 times (Perplexity).
• Saaslinks, one site's traffic from July 20 to August 3, 2026: 147,180 total parsed requests, 723 AI-crawler requests to robots.txt, 151 to actual pages (90 percent from OpenAI), and zero llms.txt requests from any identified AI system. The file's 13 total requests came from data brokers, profiling tools, and human visitors.

Both studies flag real limitations: modest sample sizes and self-reported user-agent strings. Neither claims to know what a model does internally with content it never fetched. Still, the direction is consistent across both panels and lines up with what Google has said publicly. If a widely cited "every 15 minutes" anecdote about OpenAI fetching llms.txt is circulating on your team, treat it as one anecdote, contradicted by two systematic measurements.

The Mechanism That Actually Powers Google's AI Shopping Surfaces

Google's own Shopping Help documentation explains where AI Mode, AI Overviews, and Google Shopping get their product information: "Brands, retailers, and other content providers send their product info directly to Google through tools like Merchant Center and Manufacturer Center." That data feeds the Shopping Graph, described as "a dynamic repository of product info that provides an up-to-date view of the products available," which "power[s] AI-driven experiences like review summaries, buying guidance, or product recommendations." The page does not describe crawling as a source of shopping product data.

This is the practical reason llms.txt does nothing for Google's AI shopping surfaces: those surfaces are fed, not crawled. The lever is feed data quality inside Merchant Center, not a file at your domain root. Our agentic commerce feed readiness guide covers what AI Mode, UCP, and ACP require at the protocol level; this article stays narrower, on what does and doesn't move the needle for AI shopping visibility.

What ChatGPT's Instant Checkout Actually Requires

ChatGPT's purchase flow, Instant Checkout, launched September 29, 2025, built on the Agentic Commerce Protocol OpenAI developed with Stripe. It currently supports single-item purchases, with US Etsy sellers and more than a million Shopify merchants as the initial partners. Merchants apply through a dedicated portal, and OpenAI states a merchant already processing payments with Stripe can enable agentic payments in "as little as one line of code."

The technical surface for this is OpenAI's published Product Feed Specification for the Agentic Commerce Protocol, not llms.txt. At a high level it defines:

• A feed header with feed_id, account_id, target_merchant, and target_country (a two-letter country code), all required.
• A product object with a required, stable id, optional title, description, url, and media, and a required variants array.
• Each variant with a required id and title, plus price, list_price, and unit_price as integer amounts in ISO 4217 minor units, structured barcodes (type and value, such as a GTIN pair), an availability object combining a boolean flag with a status enum (in_stock, backorder, preorder, out_of_stock, discontinued), condition values such as new or secondhand, categories, variant_options, and seller.

Separate two things OpenAI has said here. Instant Checkout runs on this structured feed submitted by applied merchants. General product mentions in an ordinary ChatGPT conversation are described by OpenAI as "organic and unsponsored, ranked purely on relevance," pointing back to normal content and SEO signals rather than a merchant-submitted feed. A feed gets you into the transactable Instant Checkout surface; it does not by itself guarantee a mention in open conversation. Our ChatGPT product recommendation playbook covers that open-conversation side in more depth.

What Gets You Into Perplexity Shopping

Perplexity's shopping assistant sources product data two ways, both feed-adjacent rather than crawl-based. The first is platform integration, primarily Shopify, which gives access to "the most recent and relevant information on products across Shopify-powered businesses." The second is the Perplexity Merchant Program, a free program through which larger retailers "share their product specs" directly, gaining payment integration for in-assistant checkout and free API access. Neither path runs through llms.txt or general crawling, consistent with PerplexityBot's zero recorded llms.txt fetches in the EZY Research study above.

Where On-Page Schema.org Markup Still Matters

None of this makes structured data on your product pages irrelevant, it just answers a different question. Google's Search Central documentation for merchant listing rich results lists required Product properties (name, image, a nested offers) and required Offer properties (price, priceCurrency), with recommended properties including gtin, mpn, sku, brand.name, availability, and itemCondition. This governs organic merchant listing rich results and Search Console's Merchant Listings report. It runs alongside your Merchant Center feed, not instead of it, and category and attribute accuracy in the feed itself, the kind our product category mapping handles, still does the heavier lifting for the Shopping Graph.

A Feed Row, Before and After

The gap between a typical Google Shopping feed row and an ACP-ready variant is small, but it hides a bug that trips up almost every first attempt: prices as integer minor units, not decimal strings.

Before, a standard Merchant Center-style row:

id: SKU-4821-BLK-M | title: Classic Wool Coat - Black - Medium | price: 129.00 USD | availability: in stock | gtin: 4006381333931

After, restructured toward what the ACP feed spec expects:

product.id: SKU-4821 (stable parent) | variant.id: SKU-4821-BLK-M | variant.price: { amount: 12900, currency: "USD" } | variant.availability: { available: true, status: "in_stock" } | variant.barcodes: [{ type: "gtin13", value: "4006381333931" }] | variant.condition: ["new"]

Two changes matter most: price moved from a decimal string to an integer in minor units (129.00 becomes 12900, not 129), and availability became a structured object instead of free text. Both are common failures the first time a team maps a Shopping feed onto the ACP structure.

A Practical Checklist

1. Stop treating llms.txt as a shopping-visibility task. Deprioritize it for product data work; any role it plays is in general documentation discovery, outside this guide's scope.
2. Audit your Merchant Center feed against Google's required and recommended Product and Offer attributes first; this is what actually feeds the Shopping Graph, and it's the fastest way to see gaps using Feedance's feed optimization checks.
3. Add or verify on-page schema.org Product and Offer markup separately from your feed, since it governs a different surface.
4. If you sell through Shopify or Etsy, confirm your Instant Checkout application status rather than assuming platform integration alone is enough.
5. If applying to the ACP feed directly, map prices to integer minor units and availability to the structured enum before your first submission, not after a rejection.
6. For Perplexity Shopping beyond default Shopify coverage, apply to the Perplexity Merchant Program directly.
7. Keep the feed current. Stale price or stock data reaching these surfaces late does more damage than a missing llms.txt file ever will; Feedance syncs daily on Essentials and hourly on Business, and the right cadence depends on how often your prices and stock move.

Where This Breaks Down

This is a feed-and-data-source guide, not a full technical SEO audit. On-page schema.org implementation is a template or CMS task, not something a feed management platform configures directly, and a dedicated technical SEO or structured-data validation tool will catch page-level markup errors a feed platform like Feedance was never built to see. The server-log studies cited here skew toward smaller sites and can't know what a model does internally with content it never fetched. The Agentic Commerce Protocol and Perplexity's Merchant Program are also young and will change; treat field names here as a snapshot and check the live specification before building against it. None of this covers regional AI shopping surfaces outside Google, OpenAI, and Perplexity, which move on their own timelines.

Frequently Asked Questions

Does llms.txt help my products show up in ChatGPT or Google AI shopping results?
No public evidence supports this. Google has stated directly it won't use llms.txt for AI Overviews, and two independent 2026 server-log studies found AI crawlers request it far less often than robots.txt, in one case zero times in fourteen days.

What's the difference between the Shopping Graph and a Merchant Center feed?
Your Merchant Center feed is the data you submit. The Shopping Graph is Google's internal, continuously updated repository built from that submitted data, which then powers Shopping results, AI Overviews, and AI Mode.

Do I need separate feeds for Google AI Mode, ChatGPT Instant Checkout, and Perplexity Shopping?
Effectively yes, in different formats: a Merchant Center feed for Google, an Agentic Commerce Protocol feed for ChatGPT Instant Checkout after merchant approval, and either Shopify platform data or a direct submission for Perplexity's Merchant Program.

Does on-page schema.org Product markup replace my Merchant Center feed?
No. It governs a separate surface, organic merchant listing rich results, and Google treats it as complementary to feed submission, not a substitute for it.

How do I get my products into ChatGPT's Instant Checkout?
Apply through OpenAI's merchant portal. Selling on Shopify or as a US Etsy seller puts you closer to eligible than a brand-new integration; merchants already on Stripe can reportedly enable agentic payments in one line of code.

How do I get my products into Perplexity Shopping?
Sell through Shopify for baseline coverage, or apply to the Perplexity Merchant Program directly to share product specs and enable in-assistant checkout.

Should I bother publishing an llms.txt file at all?
For product and shopping visibility specifically, no, it isn't the priority the SEO conversation has made it out to be. Any role it plays is in general documentation discovery, a different problem from getting products into AI shopping surfaces.

If you want a concrete read on your own feed's attribute coverage before you decide what to fix first, run it through Feedance's free product feed audit tool.