llms.txt for Shopify: Test Whether AI Bots Fetch It and Send Traffic

Probably not much. Google's guidance says Google Search does not use llms.txt and lists it among tactics you can ignore, and on Shopify the address now redirects to /agents.md unless your theme overrides it. Nothing I found shows it sends traffic. Test it on one store for four weeks, and spend your effort on product data and policy pages.

That is a position, not a verdict, and you can check it on your own store instead of taking my word. Three things decide it: what Shopify actually serves at those addresses, what a store owner can and cannot see, and whether referral traffic from AI assistants moves when the file changes. Product data and specific policy pages most plausibly matter more, and they help every channel anyway.

Follow me on Instagram@sagnikteaches

What Shopify serves at /llms.txt today

A Shopify staff reply in the developer community on 20 May 2026 said the redirect from /llms.txt to /agents.md is expected, and that Shopify now treats /agents.md as the main discovery file for agentic and AI experiences. A developer write-up dated 14 May 2026 reported that the generated files appeared across stores in the first week of May with no changelog post, and community posts on 27 May reported that Shopify again serves /llms.txt and /llms-full.txt from theme templates where those exist, falling back to /agents.md where they do not.

Connect on LinkedInSagnik Bhattacharya

Three practical consequences follow.

Subscribe on YouTube@sagnikteaches

• You may not need to create anything. Shopify serves a generated file for your store at /agents.md.
• If you create a file called llms.txt by uploading it and pointing a URL redirect at it, expect it not to work. Community reports say redirect and app-proxy workarounds stopped being honoured once Shopify began serving the files itself.
• To change the content you edit a theme template. The staff answer points to templates/agents.md.liquid for /agents.md. Community members report that templates/llms.txt.liquid and templates/llms-full.txt.liquid control the other two addresses, though that is community testing, not Shopify documentation.

Overrides live in the theme. Switching themes loses them, and they do not work for headless stores, as developers in the same thread pointed out. Shopify's staff reply also pointed to Catalog Mapping as the way to influence how product data appears in AI shopping channels, which is a clue to where Shopify itself thinks the value lies.

What Google, and the evidence, actually say

Google's AI optimisation guide (last updated 10 July 2026) says you do not need to create new machine-readable files, AI text files, markup or Markdown to appear in Google Search, including its generative AI features, because Google Search itself does not use them. It adds that Google may crawl many kinds of files without treating them in a special way, and it lists creating unnecessary AI text files such as llms.txt among tactics you can ignore. It answers its own question, whether SEO is still relevant for generative AI search, with a yes, and says a page must be indexed and eligible to show with a snippet. Its measurement advice is to use the Generative AI performance report in Search Console.

That covers Google. Other assistants publish less. Some tools and audits do read the file: for example, Chrome's Lighthouse auditing tool added an agentic browsing category with an llms.txt check in version 13.3.0, released 7 May 2026, which is evidence that some software looks for it. Whether the chat assistants that send shoppers to your store use it is the unproven part, which is why the test below measures the outcome, not the file.

What you can and cannot observe on Shopify

The clean way to test whether a bot fetches a file is a server log. Shopify merchants do not get raw server logs. A community thread asking how to access them had no confirmed answer, and the same thread's participants asked whether a content delivery layer could expose logs without any working method being confirmed. Some Shopify apps advertise crawler counting without logs, but their claims come from their own marketing pages, so I would not use one as evidence.

That leaves three things you can observe:

1. What the address returns. Open yourstore.com/llms.txt and yourstore.com/agents.md in a browser, or check the redirect from a command line, and note exactly what comes back and on what date.
2. Whether AI assistants describe your store correctly. Ask them, and record the answers.
3. Whether visitors arrive from AI assistants. Count referral sessions in your analytics before and after any change.

Number 3 is the one that matters commercially, and it is the one the test centres on.

A canary link: the nearest thing to a fetch log

You cannot see bot requests, but you can see visits to a page. If you override the template, add one link that carries a tag used nowhere else, for example /pages/store-facts?utm_source=agentsmd. The page itself may sit in your sitemap, but the tagged address exists only in the file, so any session that arrives carrying that tag came from something that read the file and followed the link: a person, an AI tool or a crawler that runs scripts. Most crawlers do not run analytics scripts, so zero proves nothing, while a handful of tagged sessions proves the file is being read somewhere. Add the canary count as an extra column in the tracking sheet below.

The four-week test, designed for a small store

The design is a simple before-and-after with a long enough baseline. It cannot prove cause in the strict sense, but it can show whether something dramatic happened.

Weeks 0 to 4: baseline

Change nothing about llms.txt. Record each Monday: sessions from AI sources, total sessions, orders from those sessions, and the status of the two addresses. Where to find AI sessions depends on your analytics; tracking ChatGPT traffic to your store walks through finding AI-referred orders, and AI for ecommerce analytics compares the tools. In general, look for referral sources named after the assistants, and for UTM tags some assistants add to outgoing links. Check what your report really shows instead of assuming.

Week 4: the one change

Make one change only, so you know what caused any difference. If you want to test the file, edit the theme template so /agents.md carries a clean, accurate summary of your store (see the sample below). Record the date and keep a copy of the file.

Weeks 5 to 8: measure

Keep recording the same numbers. Do not change prices, ads or the product range if you can help it.

Eighteen sessions to twenty-one: reading the noise

Take an illustrative print shop on Shopify with 3,000 sessions a month. In the baseline month, AI-assistant referrals are 18 sessions, producing 1 order. After the change, the next month shows 21 sessions and 1 order. That is a 17 per cent rise in a number so small that one extra person could cause it. The honest reading is "no detectable effect".

What would count as a signal? A rough rule: if the baseline sits at 18 and a later month shows 40 or more with no other cause, look closer. Even then you need to check that nothing else moved. A newsletter mention, a press link or a seasonal peak can produce the same jump. Write the list of other events next to the data, because it prevents you claiming a win you did not earn.

A tracking sheet to copy

The figures are illustrative. The columns matter more than the numbers: the last column is what stops you misreading a spike.

What to put in the file if you do edit it

If you override the template, write what a shopper or assistant would need and nothing you would not say on the site. Keep it short, accurate and consistent with your policy pages. An illustrative version for a print shop:

# Example Print Shop

> Personalised notebooks, mugs and gift tags, printed to order in our workshop.

## What we sell
- Notebooks: A5 and A6, hardback, personalised cover
- Mugs: ceramic, 11 oz, hand wash only
- Gift tags: packs of 12 and 24

## Policies
- Delivery times and costs: /policies/shipping-policy
- Returns: personalised items are final sale; /policies/refund-policy

## Contact
- /pages/contact

## Store facts
- /pages/store-facts?utm_source=agentsmd

Check three things before you save it. Every price and delivery claim must match your live policy pages. Nothing in it should contradict a product page. And you should be comfortable with a competitor reading it, because the file is public.

The fixes that matter more than the file

If the file is the minor part, what is the major part? The short answer, from Google's guidance and from how AI shopping works, is that assistants rely on the same things search does: pages they can reach, clear product facts, and trustworthy signals. Four fixes cost little and help in every channel.

1. Product data that answers the question being asked

An assistant asked for "a navy A5 notebook with a name on the cover" looks for those attributes. If your product title says "Notebook - Style 4" and the details sit in an image, it has nothing to match. Put material, size, colour, personalisation limits and what is included in the text of the product page. Shopify points to Catalog Mapping for how product data reaches AI shopping channels, so check that your category and attributes are filled in there.

A before and after: the product title "Style 4 Notebook" becomes "Personalised A5 Hardback Notebook, Navy Cloth, 120 Ruled Pages", with the description listing the printing limit of 20 characters. The assistant now has something to quote.

2. Policy pages that are specific

Shipping times, return conditions and personalisation rules are what shoppers ask assistants about. A vague "delivery in 3 to 10 days" page gives a vague answer. A page that says "printed in 2 working days, then 3 to 5 working days with the courier" gets quoted correctly. If you use the Inbox agent, the same pages feed it, which is another reason to fix them once; Shopify's built-in AI by plan shows what else reads them.

3. Not blocking what you want to be found by

Check that your robots rules are not shutting out the crawlers that assistants use. Shopify generates a robots file you can edit with a robots.txt.liquid template, and a past tweak, an app or a security setting may have blocked more than you intended. Read the file at yourstore.com/robots.txt and look for broad disallow lines that cover product pages. Shopify's help page calls editing the template an unsupported customisation and warns that incorrect use can lose you all your traffic, so change it only when you know what each line does. Do not copy a list of bot names from a web page without checking what each does, since some are for training and some for live answers.

4. Reviews and real-world mentions

Assistants weigh reviews and mentions from other sites. A product with 40 genuine reviews on the page gives something concrete to cite. Ask buyers for reviews at the right moment, and do not invent them.

An answer-check you can run in 20 minutes

Whatever you do to the file, test what assistants say. Do this before and after, with the same questions each time.

Questions to ask (use the same wording each time):
1. What does [store name] sell?
2. Does [store name] ship personalised notebooks, and how long does delivery take?
3. What is the return policy at [store name]?
4. Which stores sell a personalised A5 hardback notebook?

Record for each: assistant used, date, whether the store is named, whether the facts are right, which page was cited.

An illustrative result: the assistant names the store in answer 1 but gives a wrong delivery time in answer 2, citing an old announcement page. That points you to the real fix, which is to correct or remove the old page. It tells you more than any file check would.

One assistant answer is not a result

One failure to avoid: you ask a single assistant once, it does not mention your store, and you conclude the file failed. Assistants vary between runs and between users, and they often rely on live search. One answer is an anecdote. Ask each question three times on different days and record the pattern.

Checking the redirect from a command line

A browser hides redirects, so use a command that shows them. On a Mac or Linux terminal, or in Windows PowerShell with curl.exe, run this against your own domain.

curl -sI https://yourstore.com/llms.txt

HTTP/2 301
location: https://yourstore.com/agents.md

The two lines under the command are an illustrative result. A 301 with a location of /agents.md means Shopify is redirecting as described in the staff reply. A 200 means a file is being served directly, which happens if a theme template is overriding it. A 404 means something is wrong with the template or the theme, and you should check where you put the file. Repeat the check after any theme change, since an update can silently remove an override.

The uploaded-file redirect that quietly stopped working

A store owner follows an older guide, uploads llms.txt to the Files area, and adds a URL redirect from /llms.txt to the file. It worked in March. In June a customer asks why the store's AI file shows the generic Shopify version. The cause: since the platform began serving the file natively, redirects stopped being honoured. The owner never re-checked. The lesson is to put a quarterly reminder to run the command above in your calendar, because the platform side can change without notice, and developers in the community thread said the May change arrived with no changelog entry.

Reading your eight weeks of numbers

When the test ends, use a decision table rather than a feeling.

The point of writing the table before you start is that you decide what counts in advance, so you cannot move the goalposts after seeing the data.

A second store, with different traffic

Consider an illustrative wedding planner selling stationery and table signs from a store that gets 900 sessions a month. At that size, AI referrals may be 3 or 4 sessions a month, and a single click moves the figure by 25 per cent. The test cannot measure anything here. So for this store the better use of an afternoon is the answer-check: ask four assistants the four questions, find that two give a wrong lead time, and rewrite the shipping page to state the exact working days. The cost is two hours, the outcome is a correct answer wherever the page is quoted, and no test of the file is needed.

The same logic applies to a three-person IT support firm that sells hardware bundles online: with a small number of high-value orders, one correct assistant answer can matter more than a month of referral numbers, so check the answers by hand.

What the test costs

The whole exercise is cheap in cash, which is why it is worth doing once. Illustratively: an hour to set up the sheet and record the baseline, ten minutes a week for eight weeks, an hour for the theme template if you decide to edit it, and an hour for the answer-checks at start and end. That is about 5 hours. At $25 an hour that is $125 of time, with no software cost. Compare it with a quote for "AI visibility optimisation" and decide which return you can actually measure.

Where this sits next to Shopify's other AI features

Shopify's own changes in 2026 show where the platform is heading. Shopify's Agentic Storefronts page says the feature is enabled automatically for eligible stores and puts products in ChatGPT, Gemini and other AI channels, and its pricing page lists a $0 Agentic plan for adding products to Shopify's Catalog and selling in AI channels. I could not confirm from official pages whether any particular assistant reads the llms files, so I would not build a plan on it. For the feature-by-feature picture of what the platform includes, see what Shopify includes on each plan, and for how recommendations work on the store itself, built-in against paid recommendation apps.

Should an agency sell you this?

If someone quotes you to "optimise llms.txt for AI visibility", ask what outcome they will measure and how. The honest version of their pitch is a small task worth an hour, not a retainer. For the view from the agency side, does llms.txt work, and what SEO agencies should tell clients covers how to talk about it without overclaiming.

Which stores should bother testing the file

• If you have under 5,000 sessions a month, skip the file test and fix product data and policy pages. The numbers are too small to measure a file effect.
• If you have more traffic and a developer on hand, run the four-week test and keep the result.
• If you use a headless build, remember theme overrides do not apply, so your options differ.
• If AI referrals already produce orders, find out which pages they land on and improve those first.

If you would like help setting up the baseline and choosing the fixes, that is the kind of job my AI implementation consultation covers, and the contact page is the way in.

llms.txt questions from store owners

Does Google use llms.txt for AI Overviews?

No. Google's AI optimisation guide says you do not need machine-readable files, AI text files or special markup to appear in Search, including its AI features, because Google Search does not use them, and it lists creating files like llms.txt among tactics you can ignore. Its advice is to follow ordinary SEO practice and measure with the Generative AI performance report in Search Console.

What happens if I visit /llms.txt on my Shopify store?

Since May 2026 Shopify has treated /agents.md as its main AI discovery file, and a Shopify staff member said on 20 May that the redirect from /llms.txt to it is expected. Community posts later that month reported that a theme template for llms.txt or llms-full.txt is served where one exists. Run the curl check on your own address, because themes and rollouts differ.

Can I see which AI bots visit my store?

Not through Shopify itself as far as I could find: merchants do not get raw server logs, and a community thread on the topic had no confirmed workaround. Third-party apps claim to count crawler hits, but treat their claims as unverified. Referral visits in your analytics are the measure you can rely on.

Is it worth editing the theme to override the file?

Only if it costs you under an hour and you accept it is a bet with no proof of payoff. Overrides sit in the theme, so they vanish if you change themes and do not work for headless stores. Spend the same hour on product data and policy pages first.

Further reads

• How Much Does It Cost to Add AI to a Shopify Store? — Price the wider AI changes you might make to a store.
• Best AI Chatbots for Small Shopify Stores (2026) — Compare chatbots once your store pages are accurate.
• Shopify Inbox AI Agent: Test It on Your Policy Questions First — Test your policy pages the way an AI agent reads them.
• How to Write Shopify Product Descriptions With AI That Convert — Improve the product copy AI systems actually read.
• Best AI Assistant for Shopify: Sidekick, ChatGPT or Claude by Job — Choose an assistant for store tasks by job.
• AI for a Dropshipping Business: Where It Helps and Where It Fails — Understand where AI helps and fails in a dropshipping store.
• ChatGPT Shopping Agent: Can It Find a Small Online Shop's Products? — A merchant checklist for product data ChatGPT can read: routes in, feed fields, price and stock accuracy, returns text, and a three-product test to run.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: Google Search Central's AI optimisation guide (last updated 10 July 2026), Shopify's developer community thread on llms.txt and agents.md (20 May 2026 staff reply and later posts), a developer write-up on Shopify's generated agent files (14 May 2026), a Shopify community discussion of server logs, Shopify's help page on editing robots.txt, its Agentic Storefronts and pricing pages, and the Lighthouse 13.3.0 release notes.