GEO vs SEO for ecommerce: what actually changed

"GEO" showed up in marketing plans like a new line item: a separate discipline, a separate budget, maybe a separate agency to hire. That framing causes more confusion than it resolves. Most of what makes a product page easy for an AI system to recommend is the same thing that already made it easy for Google to rank: a real HTML page a crawler can read, a Product and Offer schema that matches what's on screen, a sitemap that's current, and content that actually answers the question a shopper is typing. That is SEO. It did not stop mattering the week ChatGPT started answering shopping questions directly, and a store does not need to choose between funding SEO and funding GEO.

What changed is smaller than the acronym suggests. A handful of crawlers to let in, a short file that works as a map rather than a dump, and one more thing worth measuring alongside rank. This article separates the part of GEO that is SEO a store should already be doing from the part that is genuinely new, so the next month of work goes toward the right few changes instead of a rebuild.

GEO is mostly SEO with a different scoreboard

Search engines and AI answer engines both start from the same requirement: a page has to be reachable and readable before anything else about it matters. A category page that returns errors, sits behind a login wall, or renders its content only after JavaScript runs is invisible to a classic crawler and to a model's retrieval step for the same reason - nobody indexes what they could not fetch.

The difference shows up further down the pipeline, not at the front. A search engine turns what it indexed into a ranked list, ten or more links on a page, and a shopper picks one. An AI system turns similar material into a short written answer that names two to six stores, with no page two. Ranking rewards being findable among many options; a shopping answer rewards being the option a model is confident enough to name out loud. Mostly the same inputs, a different output shape, and a different way of keeping score - rank position on one side, being named on the other.

That is also why GEO is not a rival discipline competing with SEO for the same hours. It is closer to a short list of additions on top of a program that, if it is already solid, does most of the work required for both scoreboards. How ChatGPT actually picks which stores to name goes deeper into the mechanics behind a single answer; this article stays on what that means for planning the work.

What stayed the same

None of the following is new, and none of it should be re-labeled as a GEO task on a roadmap. It is SEO, and skipping it is the single biggest reason a store is missing from AI answers.

• Crawlable pages. Product and category pages that return a clean 200, aren't blocked by an overzealous robots.txt rule, and don't require a login or a cookie banner dismissal to see the content.
• Honest Product and Offer schema. Structured data generated from the actual product record - price, availability, SKU - not a static block copied across every page that never updates when the product does.
• Real HTML. Price, stock status, and specifications present in the page a crawler fetches, not only inside client-side JavaScript that renders after the fact for a human browser and never for a bot.
• Current sitemaps. A sitemap that reflects what is actually for sale this week, not a snapshot from a redesign two years ago with discontinued products still listed.
• Content that answers the question. A product page that states the fact a shopper is asking about - material, sizing, compatibility - in plain text, instead of making them infer it from a photo or a PDF spec sheet.

A store with these five in place has already done most of what a generative engine needs to read it with confidence. The checklist that walks through each of them in more detail, including the exact schema fields and robots.txt lines to check, is in the GEO checklist for online stores.

The few extras that are actually new

Three things do not have a direct SEO equivalent from before. They are short, and none of them requires touching the rest of the program.

Let the AI crawlers in

A classic robots.txt written years ago, often from a generic template, sometimes disallows crawlers by name without anyone noticing: GPTBot, Google-Extended, ClaudeBot. Nothing a model has not fetched can be cited, no matter how good the page is. Checking robots.txt for a blanket rule against these names, and scoping any block down to admin, cart, and account routes instead of the whole catalog, is a five-minute fix with an outsized effect.

A short llms.txt, not a sitemap dump

An llms.txt file at the site root is a plain-text map for a model: what the store sells, a handful of category links, and the policy pages that matter for trust. It works because it is short - a one-page orientation, not the whole file cabinet. Pasting every product URL into it defeats the purpose and buries the useful part under noise a model has to wade through.

Being named, not just ranked

The last extra isn't a technical fix, it's a new question worth asking every week: did a shopping question about your category name your store this week, and how. That question doesn't replace a rank check, it sits next to it.

How you measure it: rankings vs mentions

A rank tracker still answers a real question: where a page sits in Google's results for a given keyword. That number moves slowly, mostly with backlinks, site authority, and content depth, and it's still worth watching for the traffic it drives.

It doesn't answer a second question that now matters just as much: when a shopper asks a shopping question inside ChatGPT or Gemini, or reads a Google AI Overview, is your store one of the two to six named in the answer. That's a mention, and tracking it needs cold prompts - shopping questions that never mention a brand or domain - run repeatedly against the same models, since a single chat is a snapshot and a signed-in account's memory can quietly steer what it names.

The free Aigely Scan runs ten open shopping prompts against ChatGPT for one domain, once a week, no card required, and returns a scorecard showing whether a store was named. It's a fast way to get a first answer to "were we named this week" without changing anything about how rank is tracked elsewhere. A free account keeps that scan running weekly and stores the history; the full text behind each answer opens once a project is set up, which is also free.

Once a few weeks of mentions have built up, a second number becomes visible: share of voice, which competitors get named on the same set of prompts, and how often. That's not a metric rank tracking gives directly, since a Google results page already lists every competitor at once. In a shopping answer with room for two to six names, being one of them at all is the first threshold, and how often over several weeks is the second.

What to do this month

A realistic month of work, in order:

1. Run the free scan first. Point the Aigely Scan at the store's domain and read the scorecard. It's the baseline: named or not, on ten cold prompts, before anything else changes.
2. Fix the SEO gaps the scan or checklist surfaces. Crawlable pages, a current sitemap, real HTML for price and stock, Product and Offer schema that matches the page. Use the checklist to work through them in order.
3. Check robots.txt for the three crawler names, and publish a short llms.txt. Both are small, one-time fixes, not ongoing work.
4. Decide if weekly tracking across more than one engine is worth it. A 7-day trial, card required, unlocks fifty prompts across ChatGPT, Gemini, and Google AI Overviews together (not Claude) - enough to see mention rate move over a week rather than guess from one scan.
5. Pick a plan that matches which engines matter. Starter tracks ChatGPT and Google AI Overviews. Pro adds Gemini. Business adds Claude on top of the rest. Agency adds white-label reporting for stores managing this for clients. Compare the plans against which engines your shoppers actually use.
6. Run a full audit once the basics are fixed. A site audit checks llms.txt, crawler access, and Product/Offer schema across the sitemap, or a scoped list of URLs, and points at exactly which pages are still the weak link.

What not to do

• Don't cancel or shrink the SEO roadmap to fund a separate "GEO program." The two share the same foundation; most of the budget question disappears once that's clear.
• Don't add a blanket crawler block without checking it by name. A "block all bots" line copied from an old template can silently include GPTBot, Google-Extended, or ClaudeBot.
• Don't paste the full sitemap into llms.txt. A file meant to be a short map stops working the moment it's as long as the site itself.
• Don't track only rank once mentions start mattering. A store can rank well and still be missing from every AI answer in its category; the two numbers move independently, and both are worth watching.
• Don't judge visibility from one signed-in chat. Memory and prior conversations can steer what a model names, for reasons that have nothing to do with what a new shopper would see.

The fastest way to see where a store actually stands is the free Aigely Scan: run it against cold shopping prompts on ChatGPT and read the scorecard. For weekly tracking across more engines, and the schema and crawler fixes an audit surfaces, compare the plans to find the right coverage.