Seven GEO Myths Shopify Founders Still Believe

Quick answer: Most of what Shopify founders hear about GEO (generative engine optimisation) is either unnecessary or unproven. Google’s own guidance says you do not need an llms.txt file, special schema markup, chopped-up pages or bought mentions to appear in AI Overviews and AI Mode; the page needs to be indexed, eligible for a snippet and genuinely useful. Adding sourced facts, quotes and statistics has research support, but only from a narrow test. Nobody can guarantee a citation, and a single “rank” in ChatGPT is not a stable number.

If you run a Shopify store, you have probably had a message this year telling you your store is “invisible to AI” and that the fix is a monthly fee. Some of that advice is sound. A lot of it is not, and when your budget is tight the difference matters.

A quick note on the term. GEO here means generative engine optimisation: the work of showing up in AI-written answers such as Google’s AI Overviews and AI Mode, ChatGPT and Perplexity. It has nothing to do with geotargeting or geo-redirect apps, which share the abbreviation.

This post takes seven claims I hear from founders and grades each one against sources I have opened myself: Google’s documentation, the llms.txt proposal, OpenAI’s crawler documentation, the 2023 research paper that introduced the term GEO, and a published study on how consistent AI recommendations are. I have left client results out on purpose. The grades rest on documents you can open and read yourself, and everything below is as at 30 October 2026.

How the grades work

Each myth gets one of three grades. The grade describes how well supported the correction is, not how popular the myth is.

• Documented: a platform’s own documentation addresses the claim directly.
• Plausible: a named study or a sound mechanism supports the correction, with limits I will state.
• Unproven: no good evidence either way. The claims come from vendors, opinion or small samples.

One limit up front. Google publishes detailed guidance on how its AI features work. ChatGPT, Perplexity and the others publish far less. So several “documented” grades below are documented for Google only, and I say so each time. Google is the sensible place to anchor, because it is the one that tells you what it does, but it is not the whole picture.

Myth 1: “You need an llms.txt file to get cited by AI”

Grade: Documented for Google. Unproven for everything else.

Where it comes from. The llms.txt idea is a proposal by Jeremy Howard, first published on 3 September 2024. The proposal page, which was last modified on 10 August 2026, describes a markdown file at the root of a site: a title, a short summary, and lists of links to the pages that matter most. It says thousands of sites publish one, that documentation platforms generate them, and that agents are expected to read the file and then follow its links. It also says the file is used most heavily for software documentation.

What the evidence says. Google’s guide to optimising for generative AI features says you do not need to create AI text files, special markup or markdown to appear in Google Search, because Google Search itself does not use them. It adds that creating one for other services will neither help nor harm your visibility in Google Search. That is as direct as documentation gets.

For other assistants the picture is thinner. The proposal page notes that OpenAI, Anthropic and Gemini publish llms.txt files for their own developer documentation. That is a lab publishing a file about its own product. It is not the same as an assistant reading your store’s file when it decides which skincare brand to name. The proposal describes what agents are expected to do; it does not point to a shopping assistant that confirms it uses a store’s file to choose products. I did not find one in the primary sources I read for this post.

What to do on a Shopify store.

1. Fix the things Google does document first: indexed pages, readable text, accurate policies.
2. If you still want an llms.txt file, treat it as a small, low-stakes task, not a project. For transparency, prism3.agency publishes an llms.txt file for that reason: it is cheap to keep accurate, and we make no claim that it earns citations. Before you pay anyone for one, open yourdomain.com/llms.txt and see what, if anything, is already there. List your brand summary and links to shipping, returns, sizing and your main collections. Keep it accurate, because a stale file is worse than none.
3. Do not pay a monthly fee for “llms.txt optimisation”.
4. If an agency reports an llms.txt file as a win, ask one question: which system fetched it, and when? A credible report would name the evidence.

Myth 2: “AI search needs special schema markup”

Grade: Documented for Google. Unproven for other assistants.

Where it comes from. Structured data (schema.org markup) is real, useful and easy to sell. Somewhere along the way “useful for rich results” became “required for AI answers”.

What the evidence says. Google’s guide says structured data is not required for generative AI search and that there is no special schema.org markup to add. It still recommends structured data as part of overall SEO, because it helps with rich result eligibility. Google’s AI features page (last updated 10 December 2025) adds the condition that matters most to a store: structured data should match the visible text on the page.

That last point is where Shopify stores actually get hurt. A price in the markup that differs from the price on the page, or an “in stock” flag on a sold-out product, is a mismatch. Google’s guide also says that Merchant Center feeds and Business Profiles help products and services be visible in AI responses, so the product data you feed Google is worth more attention than extra markup types.

What to do on a Shopify store.

1. Pick three product pages. Compare the price, availability and review count in your structured data with what a shopper sees on the page. They should match exactly.
2. If your theme and an app both output product data, check whether they conflict. Fix the conflict before adding anything.
3. Check your Merchant Center product data for the same three products.
4. Skip any quote that sells “AI schema” as a separate line item. Our Shopify product page SEO checklist covers the product data fields worth keeping right.

Myth 3: “Chop every page into small chunks and rewrite it for AI”

Grade: Documented for Google. Contested everywhere else.

Where it comes from. The idea is that language models read pages in pieces, so you should write in short, self-contained blocks, one question per block, and rewrite your existing pages in an “AI-friendly” style.

What the evidence says. Google’s guide says there is no requirement to break content into tiny pieces, that its systems can understand a page that covers several topics and show the relevant piece to the searcher, and that there is no ideal page length. It also says you do not need to write in a particular way for generative AI search, because the systems understand synonyms and meaning, so you do not need to chase every long-tail variation of a query.

The guide goes further on one tactic. Creating separate pages for every possible variation of a question, including the related searches the AI runs behind the scenes (Google calls this query fan-out), primarily to manipulate results violates Google’s scaled content abuse policy and is not an effective long-term strategy. If you are tempted by a plan that produces forty near-identical pages, read our piece on the programmatic SEO trap on Shopify first.

The opposing view exists. Digiday’s March 2026 article on GEO hype quotes one practitioner listing content chunking among tactics specific to AI search. It is that person’s view. It is not a test result, and it is not Google’s guidance, so I grade it as contested and leave it there.

What to do on a Shopify store. Ignore the format and keep the useful part: put the answer a shopper needs where they can find it fast. Google’s own guidance asks for content organised with clear headings and important information available as text.

• On a product page, put the facts people ask about (size, fit, ingredients, materials, delivery, returns) in plain text, not only inside an image or a tab that needs a click.
• Use the decision rule: if a change helps a shopper find an answer faster, make it. If it exists only to please a machine, skip it.
• Do not rewrite pages that already answer the question. Rewrites create risk and no documented benefit.

Myth 4: “GEO replaces SEO, so your old SEO no longer matters”

Grade: Documented for Google.

Where it comes from. AI answers sit above the links, so it is easy to conclude that ranking no longer matters.

What the evidence says. Google’s guide says generative AI features on Search are rooted in its core ranking and quality systems. They use retrieval-augmented generation (the model is given pages retrieved from Google’s index) and query fan-out. The guide says that from Google’s perspective, optimising for generative AI search is optimising for the search experience, and so still SEO. Its technical requirement is plain: to be shown as a supporting link, a page must be indexed and eligible to be shown in Search with a snippet, and there are no additional technical requirements. Google’s AI features page adds that AI Overviews are only shown when Google’s systems judge them additive to classic search, so they often do not trigger.

So where is GEO genuinely different? Two places, and neither replaces SEO.

• Other assistants have their own crawlers. OpenAI’s documentation says OAI-SearchBot surfaces sites in ChatGPT’s search features, that sites opted out of it will not be shown in ChatGPT search answers, and that GPTBot is a separate crawler for training data. So blocking the wrong one has a different effect.
• Output varies. I cover this under Myth 7.

What to do on a Shopify store. Work through this order, because each step depends on the last:

1. In Search Console, inspect your top five collection and product URLs. Confirm they are indexed.
2. Check those pages carry no nosnippet, data-nosnippet or restrictive max-snippet setting, which Google lists as the controls that limit what is shown.
3. Check that your robots.txt, and any security app or CDN in front of the store, is not blocking Googlebot or, if you want ChatGPT visibility, OAI-SearchBot. Our Shopify robots.txt guide shows how to read and edit the file.
4. Only then spend time on AI-specific work. Our guides to Google AI Overviews and AI Mode for ecommerce and how to rank in ChatGPT cover what that looks like.

Myth 5: “More backlinks and brand mentions will make AI recommend you”

Grade: Unproven (contested).

Where it comes from. Classic SEO taught us that links and mentions build authority. It is a short step to “AI recommends brands that are mentioned everywhere, so buy mentions”.

What the evidence says. This is the murkiest myth, so I will separate what is documented from what is opinion.

Google’s guide says its generative AI features can show what is being said about products across the web, including blogs, videos and forums. But it also says that seeking inauthentic mentions is not as helpful as it might seem, because its core ranking systems focus on high-quality content and other systems block spam. So real discussion is part of the picture, and manufactured discussion is not a plan.

On the other side, Digiday’s March 2026 article quotes one practitioner arguing that brands need to manage their reputation across the platforms AI systems draw from. That is a practitioner’s view, not a test result.

The paper behind Myth 6 argues that because generative engines answer from page content, backlink building should not disadvantage small creators. That is the authors’ reading of a benchmark; they did not test backlinks or social mentions directly. I did not find a primary study that isolates either as a cause of AI citations, which is why the grade is unproven and not “false”.

What to do on a Shopify store.

1. Never buy mentions or links to “train” an AI. Google’s guide says inauthentic mentions are not as helpful as they might seem for its own features, and for other assistants I found no evidence that bought ones work.
2. Spend on mentions you would want if AI did not exist: customer reviews, press, creators who actually use the product, and real answers in the forums your buyers read.
3. To see what matters in your category, run ten buyer questions in the AI tools your customers use and write down which sources get named: review sites, Reddit, publications, retailers. If the same few sources show up repeatedly and you have no presence there, that is a concrete gap. A generic “get more mentions” brief is not.

Myth 6: “Add statistics, quotes and citations and your AI visibility rises 40%”

Grade: Plausible.

Where it comes from. This one is not made up. It comes from “GEO: Generative Engine Optimization”, a paper by researchers at Princeton, IIT Delhi and others, accepted to the KDD 2024 conference (arXiv 2311.09735, latest version 28 June 2024). The paper introduced the term GEO. Its abstract says the methods it tested can boost visibility by up to 40% in generative engine responses. That number has been repeated in a great many sales pages.

What the evidence says. The paper is worth reading properly, because the details are more useful than the headline.

• The setup. The authors built a benchmark of 10,000 queries across 25 domains. For each query they fetched the top five Google results and had an AI model (gpt-3.5-turbo) write an answer from them. They then rewrote one source using a language model, applying one method at a time, and measured how much more of the answer was attributed to that source.
• What visibility meant. “Visibility” was a position-adjusted share of the answer’s wording attributed to your source, plus a score from a model-based judge. It was not clicks, sessions or sales.
• What worked. Adding statistics, adding quotations from credible sources and citing sources performed best: gains of 30 to 40% on the main measure and 15 to 30% on the subjective one. Better fluency and readability also helped. A second test on Perplexity.ai showed improvements of up to 37% on one of its two measures.
• What did not. Keyword stuffing gave little to no improvement and was about 10% worse than the baseline on Perplexity. A more authoritative tone produced no significant improvement.
• Who benefited. When every source on a results page was optimised, lower-ranked pages gained most. The paper reports that Cite Sources raised visibility for fifth-ranked pages by 115.1% while the top-ranked page’s visibility fell by 30.3% on average. So the effect depends on where you start.
• Limits the authors state. They did not evaluate how these methods affect search rankings, engines evolve, and query patterns change.

Put together, “up to 40%” is a best case from a controlled test built around one answering model, with a second check on Perplexity.ai, run in 2023 and 2024 on rewritten content and measured as share of an answer. It is the best published support for the idea that sourced, specific content gets used more. It is not evidence that Google’s AI Overviews, which the paper did not test, will behave the same way.

What to do on a Shopify store. Use the finding for what it supports: specific, sourced facts are more useful to a reader and, in this test, to an answer engine.

• Replace vague claims with facts you can back up. This is an illustrative example. Before: “Quiet, powerful and built to last.” After: “Rated at 68 dB in our own kitchen test (22 October), 250 W motor, two-year warranty.” Only publish figures you measured or can source, and say where they came from.
• Link to the source where you cite one: a supplier specification sheet, a standards body, your own test notes.
• Do it on five buyer-question pages first, such as your top collection pages and best sellers, and note the date. Check back in Search Console’s generative AI report (Myth 7). Treat any change as an observation, not proof.
• Never invent a statistic to fit the pattern. The paper tested real content; fabricated numbers harm trust and Google’s guide asks for first-hand, non-commodity content.

Myth 7: “A tool can show you where you rank in ChatGPT”

Grade: Plausible.

Where it comes from. Rank tracking is how SEO has long been measured, so a dashboard that says “you are #3 in ChatGPT for best vitamin C serum” feels natural.

What the evidence says. There is no stable ranking to report. SparkToro published a study on 28 January 2026, written with a co-researcher from the AI-tracking startup Gumshoe.ai, in which about 600 volunteers ran 12 prompts across ChatGPT, Claude and Google’s AI results, producing 2,961 responses in total. They found that the chance of getting the same list of brands in any two of 100 runs was under 1 in 100 for ChatGPT and Google’s AI, and that it took roughly 1 in 1,000 runs before two lists came back in the same order. Brand names did recur, though. In one example, a hospital appeared in 69 of 71 ChatGPT responses (97%) but was the first-named in only 25 of them. The authors concluded that visibility percentage, measured across many prompts run many times, is a reasonable metric, while a single “ranking position” in AI is not.

Take that with its stated limits. The authors say they are not professional researchers, the work was not peer reviewed, participants used default settings (so location, device and history were not controlled), newer models may behave differently, and one researcher works at Gumshoe.ai, which they flag as a conflict of interest. It is a small, honest study, not a law of nature. That is why the grade is plausible, not documented.

Google’s guide points the same way: it says no third-party tool has access to its internal ranking or AI systems, so be wary of tools that promise ranking success or claim to use internal Google metrics.

What to do on a Shopify store. Measure what you can verify.

1. Search Console. Google launched its Search Generative AI performance reports on 3 June 2026 and says it had rolled them out to all websites worldwide as of 31 August 2026. The launch post lists impressions, pages, countries, devices and dates. It does not mention clicks or queries, and Google says it is working on additional metrics over time. AI feature impressions are also still counted in the overall performance report. Open it, note your baseline and check it monthly.
2. GA4. Look at referral sessions from AI assistants as a separate group, and compare orders with Shopify, because the two count differently (see GA4 vs Shopify Analytics). Expect small numbers and judge them as a trend.
3. A prompt log of your own. This is our adaptation of SparkToro’s suggestion, not a standard method. Write 15 to 20 buyer questions in your category and country. Run each one five times in two tools. Record whether your brand appears (yes or no), not where. Calculate the share, and repeat monthly. Note the date, the tool, your location and whether you were signed in, so runs are comparable.
4. If a report gives you a single “rank” for an AI prompt, ask how many runs it is based on and what the prompts were.

What to do instead: a one-week plan

Strip the seven myths away and what is left is short and mostly dull, which is a good sign.

1. Day 1: Inspect your top five collection and product URLs in Search Console. Confirm they are indexed and not restricted by snippet controls.
2. Day 2: Read your robots.txt and any bot-blocking app settings. Decide deliberately which crawlers you allow.
3. Day 3: Match structured data and Merchant Center data to the page for three products.
4. Day 4: Put shoppers’ answers into plain text on your top pages, with first-hand facts and named sources.
5. Day 5: Record baselines: the Search Console generative AI report, AI referral sessions in GA4 and a first prompt log.
6. Days 6 and 7: Stop. Buy nothing until you have two weeks of baseline.

Questions to ask before you pay anyone for GEO

These work on any agency, freelancer or app, ours included.

1. Which of Google’s documented requirements does this change address? If the answer is “none, it’s for other AI systems”, ask what evidence shows those systems use it.
2. How will we measure it, and what is the baseline?
3. How many runs and prompts sit behind any AI “ranking” you show me?
4. What happens if nothing changes in 90 days? Anyone who guarantees an AI citation is guessing.
5. Would fixing a page shoppers already visit do more for the money?

Key takeaways

• Google’s guidance says AI Overviews and AI Mode are rooted in its core ranking systems. The technical requirement is an indexed page that is eligible for a snippet.
• For Google, you can skip llms.txt, special schema, chunking, rewriting for AI and bought mentions. Keep structured data accurate for rich results.
• For other assistants, documentation is thin. Treat claims about them as unproven until someone shows evidence.
• The Princeton paper supports adding specific, sourced facts, but “up to 40%” is a best case from one test that did not cover Google’s AI features.
• AI answers vary run to run. Measure the share of prompts where you appear, not a single rank.
• Use Search Console’s generative AI report, GA4 and a simple prompt log to build a baseline before spending.
• No one can guarantee a citation in any AI answer.

FAQs

What does GEO stand for, and is it the same as SEO?

GEO stands for generative engine optimisation: work aimed at appearing in AI-written answers. Google says that from its perspective, optimising for its generative AI search is optimising for the search experience, and so still SEO. Outside Google there are real differences, such as separate crawlers and answers that vary between runs.

Do I need an llms.txt file on my Shopify store?

For Google Search, no. Google says it does not use these files and that having one neither helps nor harms your visibility there. For other AI systems I found no primary source showing that an assistant uses a store’s file to choose products. If you want one, keep it short and accurate, and do not pay a monthly fee for it.

Does schema markup help me appear in AI Overviews?

Google says structured data is not required for its generative AI features and no special schema is needed. It still recommends it for rich results, and it should match the visible page. The practical risk is wrong data, such as a price or stock status that differs from what shoppers see.

Should I block AI crawlers in robots.txt?

It depends on what you want. Google’s crawler documentation says Google-Extended does not affect a site’s inclusion in Google Search or work as a ranking signal. It manages whether content is used for training Google’s Gemini models and for grounding in some other Google products. OpenAI says OAI-SearchBot controls whether you are shown in ChatGPT search answers, and GPTBot is a separate training crawler. If you want to appear in ChatGPT search, allow OAI-SearchBot. Check your robots.txt first, and any CDN or hosting layer in front of the store, because Google lists those as places crawling can be blocked.

Can anyone guarantee my store will be cited by ChatGPT or AI Overviews?

No. Google says that meeting its requirements does not guarantee crawling, indexing or serving, and AI Overviews often do not trigger at all. The SparkToro study found that the list of brands in an AI answer rarely repeats between runs. Be wary of anyone who promises a citation.

How do I measure AI search visibility on Shopify?

Use three sources. Search Console’s generative AI performance report shows impressions, pages, countries, devices and dates for Google’s AI features. GA4 shows referral sessions from AI assistants. A prompt log of your own buyer questions, run several times, shows how often your brand appears. Record a baseline before you change anything.

Want a second opinion before you spend on GEO?

If you are weighing up a GEO quote and want a straight view of what your store needs and what it does not, our GEO service is the place to start. It is sold alongside SEO, because both use the same pages and data, and we do not guarantee citations by any AI assistant. A good first step is the free call: book a free 30-minute diagnostic or tell us about your store. We look at your store, and at what the AI assistants say about your category, before we recommend anything. No pitch.