What AEO and GEO actually mean
AEO stands for Answer Engine Optimisation. GEO stands for Generative Engine Optimisation. AIO and LLM SEO mean roughly the same thing. The acronyms multiplied because the underlying behaviour changed faster than the vocabulary settled. The market has not converged on one term, but the practitioners worth listening to have. Google's Danny Sullivan said in early 2026 that good SEO is good GEO, AEO, AI SEO or any other label you want to use. Lily Ray at MozCon 2025 made the same point: AI search rewards SEO fundamentals applied with more precision. Aleyda Solis maintains the most respected open resource on the topic and treats it as a sharpening of existing practice, not a new discipline.
If you need a working distinction: AEO is about being the discrete answer to a specific question, the kind that fits in a featured snippet, a voice response or an AI Overview summary. GEO is about being one of the sources an LLM cites when it synthesises a multi-source answer. Both rely on the same upstream signals, which is why the rest of this guide treats them as one job and calls it AI search optimisation.
How AI search engines actually retrieve and cite
The four systems your customers use behave differently under the hood, and the practical implications matter.
Google AI Overviews and AI Mode run on a custom Gemini model and use what Google calls query fan-out: a single user question is decomposed into eight to twelve parallel sub-queries, then synthesised. iPullRank's analysis in late 2025 found AI Mode runs hundreds of fan-out queries on complex requests. Ahrefs and Profound research shows roughly 99% of AI Mode citations come from the top 20 organic results, and 38% to 76% from the top 10. Classical Google ranking still drives almost everything you see in an AI Overview.
ChatGPT uses three different crawlers. GPTBot is for training data. OAI-SearchBot is for ChatGPT's search index. ChatGPT-User is the one that fetches your page in real time when a user asks ChatGPT a live question, and it is the one that drives in-conversation citations. Search Engine Journal measured ChatGPT-User making 3.6 times more requests than Googlebot across a 24 million request sample in early 2026. ChatGPT's web grounding has historically used Bing as its data provider, which is why a clean Bing Merchant Center feed and Bing-friendly SEO matter for ChatGPT visibility. Wikipedia accounts for around 48% of ChatGPT's most-cited sources according to Profound's analysis.
Perplexity uses a three-layer retrieval pipeline with embedding search, cross-encoder reranking and an ML reranker that weights entity, authority and freshness signals. Reddit accounts for around 47% of its top citations. Freshness matters more on Perplexity than on any other system: a page from this week routinely beats a six-month-old page on the same topic.
Claude uses Brave Search as its web search provider. Web search is an explicit tool call rather than a default, so Claude only retrieves when reasoning calls for it. Profound measured 86.7% citation overlap between Claude and Brave's top results in 2025. Anthropic has publicly committed Claude will not run paid advertising, so citations remain purely earned.
The single most underrated technical action: server rendering
The biggest practical gap between most ecommerce sites and AI search visibility is JavaScript rendering. GPTBot, ChatGPT-User, ClaudeBot and PerplexityBot do not reliably execute JavaScript. Googlebot does, but Google AI Overviews and AI Mode pull from the indexed HTML view rather than the rendered DOM in many cases. If your product specifications, prices, availability, descriptions or review snippets only render client-side, AI bots see a near-empty page.
For merchants on Shopify, the default Liquid storefront is server-rendered and this is rarely a problem. For headless builds on Norce or composable Shopify, server-side rendering or pre-rendering for AI bots is now table stakes. Shopware renders server-side by default, but custom storefronts may not. Adobe Commerce with Hyvä ships fast server-rendered pages and is well-positioned. The rule is simple: open any product page with JavaScript disabled. If the price, specs and reviews are missing, an AI bot probably does not see them either.
What the evidence says about content patterns
The most-cited piece of academic work on this topic is the Princeton GEO study, published at KDD 2024 by Aggarwal, Murahari and others. They tested specific content interventions across multiple generative engines and measured citation uplift. Three interventions stood out: adding statistics with attribution gave roughly 30 to 40% more citations, adding quotations from authoritative sources gave 30 to 41%, and adding inline citations with source attribution gave around 30%. Keyword stuffing, the legacy SEO tactic, often had a negative effect.
The behavioural pattern matches what Aleyda Solis calls chunked retrieval: AI systems retrieve at the passage level, not the page level. Each section of your content should stand alone semantically. The heading should contain the question or topic. The first 40 to 80 words of the section should answer it directly, with at least one cited statistic or named source. The body expands. This structure satisfies both classical featured snippet logic and modern AI passage retrieval.
What about llms.txt
Skip it for now. Jeremy Howard proposed llms.txt in September 2024 as a way to give AI systems a curated map of your site. Adoption among AI labs is essentially zero. John Mueller at Google said publicly in June 2025 that no AI system currently uses llms.txt and compared it to the deprecated meta keywords tag. Flavio Longato audited 30 days of CDN logs across roughly 1 000 enterprise domains and found zero llms.txt fetches by GPTBot, ClaudeBot or PerplexityBot. Search Engine Land tested it on its own site for seven months and found no correlation with AI traffic.
The cost of implementing llms.txt is small. The expected return today is also small. Implement it only after the work in this guide is done, and only as a hedge in case adoption materialises. Anyone selling AI search optimisation primarily through llms.txt or AI submission services is selling something that does not exist.
Schema, structured data and what actually moves the needle
Schema markup is useful but not magic. LLMs do not parse JSON-LD at inference time. Schema influences AI search through indirect channels: Google's Knowledge Graph uses it, Bing parses it for Copilot and ChatGPT grounding, and entity-level schema helps clarify who you are across the web. The December 2024 study from Quoleady and Search Atlas found no direct correlation between schema coverage and AI citation rates. Google has been explicit that no special schema is required for AI Overviews.
For ecommerce, the priority list is concrete. Implement Organization schema with sameAs links to Wikidata, LinkedIn and Trustpilot. Implement Product schema with full Offer, AggregateRating, Review, GTIN or MPN, shipping and return data. Implement BreadcrumbList. On guides, implement Article with Person author markup. Skip FAQPage outside genuine FAQ pages — Google restricted the rich result in 2023 and stuffing FAQ markup onto product pages now risks looking like spam.
Brand mentions and entity signals
The pattern that runs through almost every credible analysis is the same. AI systems learn brands through co-occurrence in the training corpus and through citation in retrieval-augmented flows. When your brand appears repeatedly near your category in news articles, Reddit threads, YouTube descriptions and LinkedIn posts, models develop a high probability of associating you with that category. That is why Wikipedia, Reddit, YouTube, LinkedIn, Forbes, G2 and Yelp dominate AI citations across nearly every study.
For Swedish ecommerce merchants, this means brand-building investment compounds in two ways. A clean Wikipedia or Wikidata entity with sameAs links anchors every model to a single canonical version of you. Coverage in Resumé, Breakit, Ehandel.se, Computer Sweden and Dagens Industri Digital provides Swedish-language co-occurrence that is otherwise scarce in training data. Active Reddit, YouTube and LinkedIn presence shows up directly in retrieval. None of this is new PR work — it is the same investment with a measurable second return.
The Sweden-specific picture in 2026
Google AI Overviews launched in Sweden on 9 May 2025. AI Mode in Swedish followed on 8 October 2025 as part of Google's expansion to fifty new countries. The initial Swedish version of AI Mode is, per AdRelevance, missing direct Shopping Graph and Google Business Profile integration that the US version has, so commercial intent triggers AI responses less aggressively in Sweden than in the US.
The Swedish SEO community has converged on the same advice as the international one: do not abandon classical SEO, layer AI search structure on top. There is no Swedish equivalent of the SISTRIX 100 million keyword German dataset, so we extrapolate directionally from German numbers, which show position-one click-through rates dropping from 27% to 11% when AI Overviews appear and roughly 265 million lost organic clicks per month in Germany alone. Swedish merchants should expect the same direction with smaller absolute volumes.
What to do this quarter for your ecommerce store
The honest priority list, in order, for a merchant on Shopify, Shopware, Adobe Commerce with Hyvä or Norce:
Fix server rendering on product and category pages. Confirm specs, prices, availability and review snippets are in the initial HTML. This single change has more impact than every other tactic in this guide combined.
Clean your product feeds in Google Merchant Center and Bing Merchant Center. Google AI Mode and AI Overviews use the Shopping Graph. ChatGPT historically falls back to Google Shopping when its index is thin. Bing Merchant Center feeds Copilot and ChatGPT directly. GTIN, MPN, shipping, returns, sale-price-effective dates — all required.
Allow the right bots in robots.txt. Allow GPTBot, OAI-SearchBot, ChatGPT-User, Google-Extended, PerplexityBot, ClaudeBot and Applebot-Extended by default. Block them only on gated, account or internal-search paths. The correct default for almost every product retailer is allow.
Implement the priority schema set. Organization with sameAs, full Product schema, BreadcrumbList, Article with Person author on guides. Skip the rest.
Rewrite your top 10 informational pages with answer-first structure. Question in the heading, direct answer with one named source in the first 40 to 80 words, body expands. Add cited statistics and authoritative quotes where they fit naturally. The Princeton GEO findings are the closest thing to a recipe in this space.
Plan brand mentions in Swedish third-party sources. Wikipedia or Wikidata entity, Resumé, Breakit, Ehandel.se, Reddit and LinkedIn long-form. This compounds slowly but is the single highest-leverage long-term investment.
What to skip and what to watch
Skip llms.txt for now. Skip vendor pitches that promise guaranteed citations in ChatGPT or Perplexity — no such guarantee exists. Skip schema markup as a primary investment if your product pages are not server-rendered. Skip blanket-blocking AI crawlers unless you have a specific reason: more crawling means more visibility, not less.
Watch agentic commerce closely. Shopify and OpenAI launched Instant Checkout in September 2025 and pivoted to Agentic Storefronts in March 2026. Adobe Commerce supports the Universal Commerce Protocol. The day this reaches Sweden, the priority order in this guide changes. Read our agentic commerce guide, our AI in ecommerce guide and the MCP for ecommerce page for what comes after this.
If your AI-referred traffic in GA4 crosses 5% of total sessions, promote AI citation monitoring to a primary KPI. Until then, treat it as a quarterly review item rather than a daily metric.
How we work with AI search optimisation
Nordic Web Team treats AI search optimisation as part of normal ecommerce delivery, not as a separate workstream. We audit how your store renders to AI bots, fix server rendering and feed quality, implement the schema set above, rewrite the pages where it matters, and set up measurement so you can see what is actually being cited. We work platform-neutrally across Shopify, Shopware, Adobe Commerce with Hyvä and Norce. The right starting point is usually a short audit that maps your current state against the priority list above. From there it becomes a normal delivery.