The sources behind our GEO page, with the evidence level for each tactic.

GEO for Shopify: the evidence behind every claim we make.

GEO is the work of making your brand and products easy for AI assistants to find, read and describe correctly. We do it on Shopify, we measure it with dated repeat runs, and we say where the evidence ends.

Talk to us about this Book a 15 minute call

Under 1 in 100times ChatGPT or Google’s AI gave the same list of brands in a 2,961-run test, with Claude slightly higher (SparkToro and Gumshoe, Nov to Dec 2025)
0.664correlation of branded web mentions with AI Overview brand mentions, against 0.218 for backlinks (Ahrefs, 75,000 brands)
31 Aug 2026the date by which Search Console’s Generative AI report had reached all sites worldwide
On by defaultShopify’s agentic storefronts for eligible stores, so there is no listing to buy

A page card linked by a thread to a mechanical reader whose beam lights a different bottle while the page's own bottle stays in shade.

What GEO is

GEO in plain words, and what it is not

GEO stands for generative engine optimization. It is the work of making a brand and its products easy for AI assistants to find, read and describe correctly when a shopper asks them for options. The assistants are ChatGPT, Google AI Overviews and AI Mode, Gemini, Perplexity and Microsoft Copilot. You will also see AEO, AI SEO and LLM SEO. The names matter less than the job.

It is not a new kind of ranking. Nobody gets position one in an AI answer, because the answer changes from run to run and from shopper to shopper. Google also says that, from its side, this work is still SEO. So we treat GEO as sound SEO plus four extra jobs: get the product data that assistants read right, let their bots reach it, help true outside sources say true things about you, and measure honestly.

It is also wider than our AI shopping work. That page asks one narrower question: which product an assistant picks for a buying question, and why a rival got the slot. GEO is the field around it: brand mentions, bot access, store data, outside sources and measurement.

Watch for any tool or agency that blurs two measures. Brand-level tracking asks whether your brand was named. Product-level tracking asks whether one product was named, at what price, beside which rival. Tools differ in which they offer and on which engines. A figure is only useful when it says which kind it is and which engine it covers.

No guarantees. Nobody can guarantee a citation, a recommendation, a rank or a sale from an AI assistant. Below we show why, and we list what we will not do.

How it differs

Classic SEO, GEO and our AI shopping work side by side

Comparison of classic SEO, GEO and AI shopping work
Question	Classic SEO	GEO (this page)	AI shopping work
What it tries to settle	Where your pages sit in a list of results.	Whether assistants find, read and describe your brand and products correctly.	Which product wins the slot when an assistant answers a buying question.
What a result looks like	A position that holds fairly steady between checks.	A share of repeat runs that named you. It moves between runs.	A named product, a price and the rival product beside it, per engine.
What the work touches	Pages, links, crawling and speed.	Store data, bot access, pages, outside coverage and measurement.	Product records, variants, barcodes and the facts a comparison ran on.
How well the evidence supports it	Long track record.	Thin. The next table rates each tactic.	Thin. Shopify says Catalog inclusion does not guarantee a place in any answer, and OpenAI says feed eligibility does not guarantee display.

Google says its AI features are rooted in its core Search ranking and quality systems, and that no special optimization is needed to appear in them. Source: Google: optimizing for generative AI features on Search

Known and folklore

What the evidence supports, and what is folklore

We rate each common tactic by the best evidence we found. Most large studies come from firms that sell SEO or AI tracking software, and none is a randomised test of a tactic.

Common GEO tactics rated by the evidence behind them
Tactic or claim	What we found	Kind of evidence	What we do with it
Correct, complete product data in Shopify Catalog and the Google feed	Shopify and OpenAI say ChatGPT gets Shopify product data through Shopify Catalog, and Google says to keep Merchant Center data current for its AI shopping results. None says better data earns a higher place. Shopify says Catalog inclusion does not promise a position in any answer.	Platform documentation	The core of the work. We check it and repair it.
Let AI search bots reach your pages	OpenAI says to allow OAI-SearchBot if you want to appear in ChatGPT search. GPTBot is its separate training crawler.	Platform documentation	A condition to meet, not a lever to pull.
Earn mentions on independent sites	Across 75,000 brands, branded web mentions correlated 0.664 with brand mentions in Google AI Overviews. Backlinks scored 0.218. The data covers AI Overviews only, and only brands above an authority threshold.	Correlation (Ahrefs, 2025)	We plan real coverage and never buy it.
Rank well in classic search	Across 15,000 prompts, about 12% of links cited by the AI assistants Ahrefs tested were in Google's top 10 for the same prompt, on average. Perplexity was the outlier at nearly 1 in 3. Each of the others was under 9%.	Correlation (Ahrefs, Aug 2025)	Keep the basics. Do not expect one result to carry across engines.
Add schema markup to get cited	Ahrefs compared 1,885 pages that added JSON-LD with 4,000 matched pages. AI Overviews citations on those pages ended 4.6% below the matched pages. Both groups were already falling before the markup was added, and Ahrefs says it cannot tell if markup helped a little or did nothing. AI Mode (+2.4%) and ChatGPT (+2.2%) moved by amounts that could be chance. Google says no special markup is needed for its AI features.	Matched test (Ahrefs, 11 May 2026), the best we found	We use markup for Merchant Center and rich result eligibility, not as an AI lever.
Publish an llms.txt file	Google says Google Search ignores such files. SE Ranking modelled about 300,000 domains and found no effect. Shopify already serves one on every store.	Platform statements and a correlation study	We do not sell it.
Cut pages into small chunks, or rewrite them for AI	Google says neither is needed. A July 2026 review of 45 GEO studies says rewrites aimed at being cited can hurt retrieval.	Platform statement and a literature review	We edit for the shopper.
Add statistics, quotes and sources to the text	The paper that named GEO reported relative gains of 30 to 40% on one visibility measure in a lab benchmark. The 2026 review says that does not show lasting traffic effects. We did not find a replication on live engines.	Lab benchmark	Good editing. No promised uplift.
Add FAQ markup	Google stopped showing the FAQ rich result on 7 May 2026. Ahrefs' matched test saw no reliable AI effect from markup in general.	Platform statement and a matched test	Keep real questions on the page for shoppers. Skip the markup claim.
Refresh dates often	Cited pages averaged 25.7% younger than ordinary results in an Ahrefs study. Ahrefs notes Google's John Mueller warns against changing a date without changing the page.	Correlation (Ahrefs, Jul 2025)	Update facts that changed. Never touch dates alone.
Rank your own brand first on a 'best of' list you wrote	'Best of' lists made up 43.8% of source page types in an Ahrefs sample of 750 ChatGPT prompts, and brands placed higher on several outside lists were recommended more. Ahrefs found no downside to such lists in its data but notes many read as self-serving, and says it will not always put itself first.	Correlation (Ahrefs, Dec 2025)	We do not do it. A list that ranks its own writer first is a trust risk with readers, whatever the data says.

A July 2026 review of 45 GEO studies found no technique with a stable, long-run causal effect across platforms. Read each row as the best available evidence, not as proof. Source: arXiv 2607.14035: critical survey of GEO studies (preprint, 15 July 2026)

What Shopify already ships

What Shopify does for you, so we do not sell it back

Shopify defaults that matter for AI assistants, and whether we sell them
What Shopify does	What it means for you	Do we sell it?
Shopify Catalog takes in eligible products automatically	You cannot opt out of Catalog itself. Inclusion does not promise a place or exact wording in any answer. Each channel decides final ranking and wording.	No. We check the data inside it.
Agentic storefronts are active by default for eligible stores	ChatGPT, Google AI Mode and Gemini, Copilot and Meta can reach your products. You accept the terms. You keep the customer, and orders show their channel.	No. We check the conditions are met.
/agents.md, /llms.txt and /llms-full.txt are served on every store	Shopify says you need no third-party app to make them. They do not replace Catalog.	No.
Themes that call the structured_data filter print JSON-LD for product and article pages	Shopify’s own guide says this covers the basics. Breadcrumbs, FAQs and return policies usually need adding by hand or by app.	Repair and gaps only. See our structured data page.
A free Knowledge Base app holds FAQs that AI shopping agents use	You can edit the answers it generates for your store.	We review them with you.
Sales channels, Agentic, shows performance insights, and orders show a channel	It is the first place to look for AI-sourced orders.	We read it with you.

Shopify's ChatGPT channel lists selling to US customers as a requirement. If you do not sell to the US, check whether it applies to you before planning around it. The markup row comes from Shopify's blog guide to ecommerce schema. Source: Shopify Help: agentic storefronts

Where stores lose the answer

Four ways a store loses the answer without knowing

None of these needs a tool to spot. Each needs someone to look.

Cited but not recommended
An assistant can link your page as a source and still name a rival's product in the answer. We call this being cited but not recommended. It is easy to count: in each run, did your domain appear in the sources, and was your product named? A wide gap can have several causes, so we read the raw answers before we say why. OpenAI says product picks depend on the shopper's query and context, such as Memory or custom instructions, and on structured data from providers. It does not say that being cited in the sources sets the pick.
Source: OpenAI: shopping with ChatGPT search
The page and the feed disagree
Google settles clashes between sources in a fixed order. For return and shipping terms, a product feed in Merchant Center outranks markup on the page. OpenAI notes that price and shipping updates can lag. If your theme prints one price and your feed carries another, shoppers and assistants can see two answers, and we cannot say in advance which one an assistant will use. We compare page, markup and feed for the same products and list each mismatch.
Source: Google: merchant listing structured data
A bot rule nobody remembers
OpenAI keeps its search crawler, OAI-SearchBot, apart from its training crawler, GPTBot. A store can block the second on purpose and the first by accident, through robots.txt.liquid, a CDN rule or a firewall. Shopify adds that blocking AI crawlers in robots.txt or at the network only affects open-web discovery. It does not stop Catalog sending data to the storefronts you switched on. So we test what each bot can fetch, then read what the rule is doing.
Source: OpenAI: crawlers and bots
A score that moves for no reason
In a test of 2,961 runs, ChatGPT and Google’s AI gave the same list of brands fewer than 1 time in 100 (Claude was slightly higher), and the same list in the same order about 1 time in 1,000. A score will also move if you change the prompt list, so we keep the list fixed and note any change. Models and source mixes can change too, which is why we repeat runs and date them. We report the share of runs that named you, with the run count, the date and the engine. We never call it a rank.
Source: SparkToro and Gumshoe: AI recommendation consistency

Check it yourself

Test your own store in ten minutes, no tool needed

Answers differ by person and by run. OpenAI says ChatGPT uses Memory and custom instructions, and Perplexity says its Memory can draw on your search history, so use a clean profile if you can. Five runs is a spot check, not a measurement. The test behind our first fact, SparkToro’s, ran each of 12 prompts 60 to 100 times on each platform.

Write three shopper questions
Write them the way a customer would: long and specific. One names no brand. One compares your product type with a close alternative. One names you.
Run each one five times
Start a new chat every time and stay on one engine. Save the answer text and every link it shows.
Count three things
How often your brand is named. How often one of your products is named. How often your domain appears among the sources.
Look for the gap
Your domain in the sources but your product not named is being cited but not recommended. A product named with an old price or a wrong detail goes on a list of wrong facts. Do not read any of it as a rank.

The work

Seven jobs, with what each delivers and what it does not

We scope the work after we look at your store. You may not need all seven.

Measure
Baseline and question panel
We deliver a written list of shopper questions agreed with you, at least 20 runs of each question on each engine that matters in your category, with the raw answers saved and dated, and a first-party baseline from Search Console, Google Analytics 4, Merchant Center and Shopify. We do not deliver one score, a rank, or product-level figures on engines where tools cannot see products.
Source: Google Search Console Help: Generative AI performance report
Shopify
Supply path check
We check that your products can flow to the assistants: Catalog eligibility, agentic terms accepted, the Google and YouTube channel connected with product sync on, policies complete, and legal disclosures inside the first 6,000 characters of the description, which Shopify asks for. We do not get you listed, because Shopify lists eligible products itself. We do not set up checkout inside ChatGPT.
Source: Shopify Help: ChatGPT agentic storefront
Data
Product data repair
We read the fields Shopify says AI platforms consider: title, description, images, type, vendor, collections, tags, barcode, and variant and option names. We fix gaps and clashes, use Catalog Mapping where your real data sits in metafields or tags, and compare page, markup and feed. We never invent barcodes or attributes. We do not claim better data buys a higher place.
Source: Shopify Help: optimizing your products for AI platforms
Access
Bot access and rendering
We test what the search bots of OpenAI, Google and Perplexity can fetch, read robots.txt.liquid and any CDN or firewall rule, and confirm product data sits in the first HTML. Our technical SEO page has the detail. We do not build an llms.txt file, because Shopify serves one. We do not open your store to training crawlers unless you ask.
Source: OpenAI: crawlers and bots
Pages
Product and collection pages
Shopify advises detailed specifications, comparison with similar products, sizing, material and care details, and complete policies. We write and structure those from your real facts and photographs, with the decisive facts first. Our answer-first content page has the method. We do not rewrite pages only for machines, and we do not make a page for every way to phrase a question.
Source: Shopify Help: optimizing your store for AI
Reputation
Outside evidence plan
The strongest signal in the large studies sits off your site: branded mentions on independent sites. We read the sources cited in the saved answers for your category, then write a plan for earning true coverage there, such as honest reviews on platforms shoppers use, press, creators and community answers you can stand behind. We do not buy, fake or incentivise any of it.
Source: Ahrefs: 75,000 brands and AI Overview mentions
Repair
Wrong facts log
When an assistant states an old price, a dropped ingredient or a wrong policy, we save the answer with its date and prompt, trace the likely source in your data or on the web, fix what is yours, and check again. Shopify says each channel controls final ranking and wording, so we cannot force a correction.
Source: Shopify Help: Shopify Catalog

Our limits

What we will not do

A published list of work we decline, with the reason for each
We do not	Why
Promise citations, a rank, traffic or revenue	Brand lists changed on almost every run in a 2,961-run test, and OpenAI says product picks depend on the shopper's query and context. There is no position to promise.
Sell an llms.txt or agents.md file as a growth lever	Shopify serves both on every store, and Google says it ignores such files.
Buy or fake reviews, mentions or forum posts, or pay for a review that has to be positive	The FTC final rule, announced in August 2024, bans fake and purchased reviews, including AI-generated ones. Google says seeking inauthentic mentions is not as helpful as it might seem.
Publish a 'best of' list that ranks our client first	Ahrefs saw no downside in such lists in its data, but its sample also showed brands placed high on independent lists were recommended more. We would rather earn that second kind of place. Many shoppers also see self-ranked lists as thin.
Show one AI visibility score with no method behind it	A figure should name the engine, the questions, the run count, the date and the location. Google says no outside tool can see its internal ranking or AI systems.
Claim product-level tracking on every engine	Tools differ in what they can see by product and by engine. We state the engine for every figure.
Sell checkout inside ChatGPT	Shopify says ChatGPT is a discovery channel and the purchase finishes on your own checkout.
Charge on 'AI-influenced revenue'	An assistant can influence a buyer and send no click, which analytics cannot show. You could not check the bill.
Show a GEO result for a client	We publish none here. Until we can show one with a named store, a period, a method and permission, we will not.

Process

A first quarter, week by week

This is a template, not a promise. Dates move with your approvals and your theme.

Weeks 1 and 2
Look, and change nothing
We run the supply path check, the bot access test and a product data read. We agree the question panel with you and take the first runs and the first-party baselines. You get a written findings list. Each finding shows its evidence and the kind of evidence it is.
Week 3
Agree what to fix
We rate each fix as documented mechanism, plausible but unmeasured, or folklore to skip. You choose. Nothing ships without your say.
Weeks 4 to 8
Fix the store
Theme and page changes go onto a duplicate theme first, so you read them before they go live. Data and feed changes are listed for your approval before they go live. Bot rules change only with your approval.
Weeks 4 to 12, alongside
Plan outside coverage
We read the sources in the saved answers and write the plan for earning real coverage. You or we carry it out, in your name and without any paid placement.
Week 10
Run the same panel again
Same questions, same engines, same method. We report the change with the run counts. Many differences will be noise, and we say which ones.
Week 12, then monthly
Report, decide, repeat
A monthly report with the fields shown below, a list of what changed on the store, and a decision on what to do next. Some months the right answer is to do nothing.

The engines

The engines are not one thing

They pick products differently, take data from different places and show different shoppers different answers. Study numbers disagree too. Reddit is about 12% of ChatGPT citations in Similarweb's US data for January and February 2026, and 1.8% in Profound's data for August 2024 to June 2025. The two studies cover different periods, prompts and counting rules, so we never quote one share as the number.

Platform docs
ChatGPT
OpenAI says ChatGPT chooses product results on its own, by relevance to the shopper's query and context, including Memory and custom instructions, and that they are not ads. Labels such as 'Most popular' are generated, not verified. In one vendor study of 3,312 shopping prompts run on six engines, ChatGPT showed a product card on 87% of prompts. The vendor warns its prompt mix probably overstates that.
Source: OpenAI: shopping with ChatGPT search
Platform docs
Google AI Overviews, AI Mode and Gemini
Google says there are no extra requirements to appear in AI Overviews or AI Mode, and that Merchant Center feeds and Business Profile can help products show. Shopify says AI Mode and Gemini reach eligible stores through the Google and YouTube channel or Merchant Center, and that these surfaces, and its direct checkout for US buyers, are rolling out, so they may not be on for you yet. One vendor study (BrightEdge, holiday shopping season 2025, tens of thousands of shopping prompts) found retailers were about 4% of citations in Google AI Overviews and about 36% in ChatGPT.
Source: Google: AI features and your website
Platform docs
Perplexity
Perplexity says it ranks product listings by authority and relevance, personalises by past searches and saved preferences, and is more likely to recommend merchants with deeper product details such as availability, reviews, pricing and specifications. Its links overlap Google’s top 10 more than the other assistants in the Ahrefs data. We could not open its current merchant programme terms, so we do not describe them.
Source: Perplexity Help: what is Instant Buy
Platform docs
Microsoft Copilot
Microsoft says the merchant stays merchant of record in Copilot Checkout, and that stores share a feed through approved paths such as Microsoft Merchant Center or partners like PayPal and Stripe. Microsoft says Shopify merchants are enrolled automatically, with controls in Shopify admin, and Shopify limits direct checkout to US customers. Ahrefs found Copilot links overlapped Bing’s top 10 more than ChatGPT or Gemini did, though the average across engines was only about 10%.
Source: Microsoft Advertising: agentic commerce

Measurement

Why we report shares of runs, never ranks

In the SparkToro and Gumshoe test, hundreds of volunteers ran 12 prompts 60 to 100 times each on ChatGPT, Claude and Google's AI. Almost every answer differed in which brands it named, in what order and how many. For ChatGPT and Google’s AI the same list of brands came back fewer than 1 time in 100. Claude was slightly higher. The same list in the same order came back about 1 time in 1,000.

The authors say the share of runs in which a brand appears is probably a reasonable measure. They also warn that real shoppers word prompts in many different ways, so no panel covers everything. A brand that shows up in most of many runs is visible. A brand that shows up in few is not. That is the number we report. We never report a share from fewer than 20 runs of a question on one engine, and we always print the run count, the engine, the date and the location. It is a poll of the answers we sampled. It is not a ranking, not your share of real shopper conversations, not traffic and not revenue.

Tools differ in how they collect. Some call a model through its API and some drive the chat window. One independent methodology guide notes that API runs can lack web search and personalisation, so they may not match what shoppers see. We ask every vendor, and ourselves, which method produced a number. Google adds that no outside tool can see its internal ranking or AI systems.

Scores also move for reasons that have nothing to do with your work. Semrush saw ChatGPT's citations of Reddit and Wikipedia fall sharply in September 2025, and its own analyst said the popular explanation was not the whole story.

Where we look first. Before any third-party score, we read what the platforms report: Search Console, Google Analytics 4, Merchant Center and Shopify. The next table says what each can and cannot show.

First-party numbers

What each measure can show, and what it cannot

Measurement sources for AI visibility with their limits
Source	What it shows	What it cannot show
Search Console, Generative AI performance report	Impressions in AI Overviews and AI Mode by page, device and country. By 31 August 2026 it had rolled out to all sites worldwide.	What the answer said. Anything outside Google.
Google Analytics 4, AI Assistant channel	Visits from sources such as ChatGPT, Gemini, Deepseek, Copilot and Grok. It excludes AI Overviews and AI Mode. Shopify says Google Analytics and custom pixels do not fire inside direct checkout in Google AI Mode, Gemini and Copilot, so those orders will not show here.	Answers that sent no click. Treat it as a floor, and check your own property, because older setups and custom groups may differ.
Merchant Center, AI performance report	Your share of voice in AI Mode and AI Overviews against the competitors Merchant Center lists for your account, for English-language queries in Australia, Canada, India, New Zealand and the United States, with a few days of lag.	ChatGPT or any non-Google engine. It reads 100% if Merchant Center has no competitors for your account.
Shopify admin, Agentic insights and order channels	Orders with a channel or referrer, and performance insights for agentic storefronts. The Knowledge Base app shows how often agents request your store information. Shopify says agentic performance reports are not yet available for headless stores.	Shoppers who research in an assistant and later search your name. They land as organic or direct.
A repeat-run question panel	The share of runs that named your brand or product, by engine and date, with the sources cited.	A rank, your share of real conversations, traffic or revenue.
A third-party visibility score	Whatever the vendor sampled, if it states its method.	Anything Google keeps internal. Google says no outside tool has access to its ranking or AI systems.

The Google Analytics help page for the AI Assistant channel carries no fixed date, so we confirm the channel in your own property before relying on it. Source: Google Search Console Help: Generative AI performance report

Reporting

A monthly report: the fields, with no numbers filled in

This is the layout of what you receive. The cells are empty on purpose. We will not show invented figures as a sample.

Layout of the monthly GEO report, with field names only
Part of the report	Fields we print	What it lets you check
Panel	Question text, engine, location and language, run date, runs per question.	That you can repeat any figure yourself.
Brand level	Share of runs naming your brand, share naming each rival, change since last month with both run counts.	Whether a change is bigger than run-to-run spread.
Product level, where a tool can see it	Share of runs naming each product, price mentioned, rival product named beside it, engine covered.	Which engines a figure covers. Blank where we cannot see.
Sources	Domains cited, whether your domain was cited, count of runs cited but not recommended.	Where answers in your category get their facts.
Wrong facts	Claim as stated, raw answer, date, likely source, fix made, date of re-check.	Which errors are yours to fix and which are not.
First-party numbers	Search Console Generative AI impressions by page, Google Analytics 4 AI Assistant sessions, Merchant Center AI performance, Shopify agentic orders.	Numbers that come from the platforms, not from us.
Changes shipped	Page or setting changed, date, who approved it, link to the difference.	What changed on your store between two runs.
Limits	What this month’s data cannot show.	That we are not claiming more than we know.

Who it is for

Who this suits, and who it does not

It suits you if

You sell products people compare before buying, in a category where a shopper might ask an assistant for options.
You want a list of fixes with the evidence for each, not a score to show a board.
Your product data, policies or pages are the weak part, and you want someone to check and repair them.
You are happy to be measured by repeat runs and to read results that may be flat.

It does not suit you if

You want a promise of citations, a rank, traffic or revenue.
You expect us to buy reviews, placements or forum posts.
You need a result by a fixed date. We could not find a study that gives a reliable time to results.
You want a single visibility figure with no method behind it.

Platform

Shopify next to other platforms, stated as facts

How products reach AI shopping surfaces on Shopify and on other platforms
Surface	Shopify store	Store on another platform
ChatGPT shopping	Product data is already integrated through Shopify Catalog. OpenAI says no additional work is required.	Can apply for a direct product feed. OpenAI's feed spec lists nine required fields: item_id, title, description, url, brand, seller_name, image_url, availability and price. It says eligibility does not guarantee display.
Microsoft Copilot	Merchants are enrolled automatically, with controls in Shopify admin. US customers only.	Share a feed through an approved path, such as Microsoft Merchant Center or a partner like PayPal or Stripe. Microsoft asks for a UCP-ready feed.
Google AI Mode and Gemini	Products arrive through the Google and YouTube channel or Merchant Center. Shopify lists conditions: a US-based store selling to US customers, a valid Merchant Center account, product sync on, and policies that match Merchant Center. Shopify also says these surfaces and its direct checkout are rolling out, so they may not be on for your store yet.	A Merchant Center feed. Google says feeds can help products show in AI responses.

This page covers Shopify. On other platforms the supply path is a feed you build and keep current, which is different work from what we describe here. The ideas about accurate data, bot access, independent evidence and honest measurement carry over. Source: Shopify Help: agentic storefronts for ChatGPT, Google and Microsoft

Engagement fit

Start now, or fix the basics first

A short test before you spend anything.

Start now if

You can approve changes on a duplicate theme within days.
Someone on your side can answer product questions: ingredients, sizes, compatibility.
You sell to US customers, or you know which surfaces apply to you if you do not.
You accept that the first honest result may be a short list of fixes and a flat chart.

Do something else first if

Your product pages are thin or copied from a supplier. Start with our answer-first content work.
Your site is slow or breaks on phones. Start with Core Web Vitals.
You do not sell to the US and your plan depends on Shopify’s ChatGPT channel. Shopify lists US customers as a requirement.
You are looking for a way to buy visibility rather than earn it.

Questions

Questions, answered

What is GEO, and is it different from SEO?

GEO is the work of making your brand and products easy for AI assistants to find, read and describe correctly. Google says that from its side the work is still SEO, and that its AI features rest on its core ranking systems. The difference is in the result. An assistant gives a list that changes between runs, so we measure shares of runs instead of positions.

Do I have to do anything to appear in ChatGPT shopping if I am on Shopify?

OpenAI says Shopify product data is already integrated through Shopify Catalog and no additional work is required. Shopify lists conditions: you sell to US customers, your products are eligible, you accept the agentic terms, and your Terms, Privacy and Return policies are complete. Meeting them does not guarantee a place in any answer.

Can you guarantee citations, a rank, traffic or sales?

No, and nobody honestly can. Brand lists changed on almost every run in a 2,961-run test. OpenAI says product picks depend on the shopper's query and context. Shopify says Catalog inclusion does not guarantee a place in any answer. What we promise is a documented method, dated records and reports that state their limits.

Should I build an llms.txt or agents.md file?

No. Shopify serves /agents.md, /llms.txt and /llms-full.txt on every store and says no app is needed. Google says it ignores such files, and SE Ranking found no effect in a study of about 300,000 domains.

Does schema or FAQ markup help assistants cite me?

The best test we found did not show a reliable lift. Ahrefs compared 1,885 pages that added JSON-LD with 4,000 matched pages. AI Overviews citations fell 4.6%, and AI Mode and ChatGPT moved by amounts that could be chance. Google ended the FAQ rich result on 7 May 2026. We still use product markup for Merchant Center and rich results, which is a separate job.

How will I know if it is working?

Start with what the platforms report: Search Console's Generative AI report, the AI Assistant channel in Google Analytics 4, the AI performance report in Merchant Center where your country is covered, and Shopify's agentic insights and order channels. Then read our repeat-run panel with its run counts. None of these shows everything, and answers that send no click do not appear in analytics at all.

How much traffic does AI send, and does it convert?

On its Q2 2026 call, Shopify said AI-driven traffic and orders to its stores tripled year over year, and that the volume is still small beside its total sales. An academic study of 973 sites from August 2024 to July 2025, by researchers at the University of Hamburg and the Frankfurt School and reported by Digital Commerce 360, found ChatGPT referral traffic converted below every traditional channel except paid social. On the same call, Shopify said new buyer orders from AI channels came in at nearly twice the rate of other channels, but the report does not say how that was measured. Treat AI traffic as growing from a small base, with conversion evidence that conflicts.

How long does it take?

We could not find a study that gives a reliable time to results, so we do not quote one. Some delays are documented. Google says changes from a connected store platform, such as Shopify, can take 24 to 48 hours to show in Merchant Center, and Shopify says removing Catalog access can take up to 7 days to take effect. How long assistants take to show a change after that is not documented. We plan our first re-run for week 10 and report what we see.

Will buying reviews or posting on Reddit help?

We do not do it and we advise against it. The FTC final rule bans fake and purchased reviews, including AI-generated ones, with civil penalties, and Google says seeking inauthentic mentions is not as helpful as it might seem. Real coverage on independent sites is the best-supported lever in the large studies, but those studies show correlation, not cause.

Can shoppers buy inside ChatGPT?

Shopify says ChatGPT works as a discovery channel and the purchase finishes on your own checkout. We plan on that basis and do not sell checkout inside ChatGPT. Shopify says direct checkout is rolling out, and can appear for eligible US buyers in Google AI Mode, Gemini and Copilot, and that you can turn Copilot’s off. OpenAI’s own help page still mentions an Instant Checkout option for some eligible merchants, so check your store’s settings rather than assuming.

Keep reading

Related pages.

Sources (42 pages, all opened on 2026-10-20)

Want an honest read on how assistants see your store?

Send us your store and three questions your customers would ask. We scope the work after looking at it, and we tell you what we would change first, what the evidence supports and what it does not.

Talk to us about GEO Book a 15 minute call

GEO for Shopify: the evidence behind every claim we make.

Cited but not recommended

The page and the feed disagree

A bot rule nobody remembers

A score that moves for no reason

Write three shopper questions

Run each one five times

Count three things

Look for the gap

Baseline and question panel

Supply path check

Product data repair

Bot access and rendering

Product and collection pages

Outside evidence plan

Wrong facts log

Look, and change nothing

Agree what to fix

Fix the store

Plan outside coverage

Run the same panel again

Report, decide, repeat

ChatGPT

Google AI Overviews, AI Mode and Gemini

Perplexity

Microsoft Copilot