Shopify’s defaults are a good starting point, but customizing llms.txt and agents.md can give AI agents clearer, more accurate context about your store.
For most merchants, the defensible answer is: leave them, and spend the energy where agents actually look.
That is not a dismissal of the files. It is what the evidence supports. Google's May 2026 guidance says llms.txt is not used for AI Overviews, AI Mode or other generative Search features. Crawler-log research covering 500 million bot events shows the major AI crawlers rarely fetch the file at all. They prefer your HTML.
Meanwhile, the systems that drive AI sales on Shopify (ChatGPT shopping, Copilot, AI Mode) consume structured product data through Catalog, which Shopify keeps synchronized for you. And since Shopify made agents.md canonical and pointed the llms files at it, the default state is coherent, current and maintained by the platform.
A default you never touch cannot drift out of date. That property is worth more than it sounds.
The case for customizing exists, but it is narrower than the GEO industry implies, and it comes with a maintenance contract.
The files matter most in the agentic layer, what some practitioners call business-to-agent surfaces. This is where autonomous agents fetch context before acting for a user. If an agent is checking your return window before recommending a purchase, or locating your size guide, warranty terms or support channel, a curated file that routes it to the canonical page beats letting it guess among lookalike URLs.
Merchants with real reasons to steer get genuine value here: regulated products, complex policies, wholesale versus retail surfaces, documentation-heavy goods. Practitioner guides converge on the same shape for a good custom file. Canonical policy and category URLs. Plain factual claims. Even an explicit section on what you do not claim, because models echo their sources, and an overclaiming file produces overclaiming AI answers about you.
The risk that decides most cases is staleness. Yotpo's guidance is blunt: an outdated file listing wrong prices is worse than no file at all. Log-based analyses add that stale files feed agents an outdated map of the site and erode the trust agents place in the source over time.
Shopify's defaults regenerate with the platform. Your hand-edited version regenerates when you remember.
So the honest decision test is a maintenance question, not an SEO question. If no one owns a quarterly review of the file, do not fork it.
If you do customize, follow the structure Shopify's three Liquid templates enable. Put capabilities and policies in agents.md. Let the llms files mirror or slim it. Never hardcode prices or inventory. Link canonical pages instead of duplicating their content into the file.
One more reframe keeps the effort proportionate. These files are free to keep and cheap to maintain, which matches their expected value: a modest, insurance-like payoff in agent readability, not a rankings lever. The channels with measured revenue behind them, meaning the AI orders now landing in Shopify admin with attribution, run on product data quality, eligibility and reviews.
The insight: customize the files when you have something specific to tell an agent and someone accountable for keeping it true. Otherwise, Shopify's defaults are the rare SEO artifact best left alone, and your optimization hours belong in the catalog the agents actually read.