Norrhavet · Method
How Swedish companies actually appear in AI search.
AI search isn't one engine — it's four: Google AI Overviews, ChatGPT, Perplexity and Claude. Each retrieves and cites differently. This is the mechanism, the surfaces that matter for Swedish-language queries, and how Norrhavet measures who actually gets cited.
- Mechanism explained, not 'tips and tricks'
- All four AI surfaces measured separately
- Methodology open — audits are reproducible
Definition
What is AI search?
AI search is any query interface where the answer is generated by a large language model that cites web sources — Google AI Overviews (Gemini), ChatGPT (browse + memory), Perplexity (live RAG) and Claude (tool use). It is not the same as classic SEO: the engine picks one or more sentences to quote, then links the source. Visibility means being the cited sentence, not ranking #1 on a list.
Mechanism
How an AI answer is built — and where you fit.
Step 1: the user asks a question. Step 2: the model retrieves relevant pages (live web search for Perplexity/ChatGPT-browse/Google AI Overviews; tool use for Claude; training corpus for plain ChatGPT). Step 3: the model summarises and cites the strongest 2–5 sources. To be the cited source, your page must (a) be retrievable for that exact query, (b) have a clean question→answer structure the model can copy verbatim, (c) carry schema.org markup so the answer engine recognises the format, and (d) be authoritative — named entity + source links + freshness.
Surfaces
The four surfaces that matter in Sweden 2026.
Google AI Overviews — appears above classic results for question-style queries; cites 3–6 sources; favours pages with FAQPage / QAPage schema. ChatGPT — browse mode retrieves live web; memory mode pulls from training corpus (lag of 6–18 months). Perplexity — pure live RAG; cites 3–8 sources per answer; favours short answer paragraphs. Claude — with tool use enabled, cites verbatim and attaches anchors. Norrhavet runs the same Swedish prompt across all four every week and measures who gets cited.
Measurement
How visibility in AI search is measured.
Define a prompt panel of 50–200 real customer queries. Run each prompt against all four surfaces on a fixed cadence (weekly). Record (1) was the brand mentioned, (2) was it cited with a link, (3) what was the position in the citation list, (4) which competitor was cited instead. The metric is Share of Citation, not rank position. Norrhavet's open methodology is documented at /vad-ar/mata-ai-synlighet.
Next
Start with AEO, then measurement.
AEO — Answer Engine Optimization — is the practice that makes a page citable by AI answer engines. Read the pillar on /aeo for the full method, then the measurement guide on /vad-ar/mata-ai-synlighet to set up your own prompt panel.
Frequently asked
Appearing in AI search — FAQ.
What does 'appearing in AI search' actually mean?+
Being the source that an AI answer engine — Google AI Overviews, ChatGPT, Perplexity or Claude — cites when generating its answer. Unlike classic SEO it is not about list rank: the engine quotes one or a few sentences and links the source page. Visibility = being the quoted sentence.
How do you appear in AI search as a Swedish company in 2026?+
Be retrievable (live-crawlable + present in the engine's index or training data) AND citable (clean question→answer structure, schema.org markup such as Article/QAPage/FAQPage, named entities, source links, recent lastReviewed date). Then measure Share of Citation across Google AI Overviews, ChatGPT, Perplexity and Claude weekly to know what is working.
Which AI surfaces matter for Swedish-language queries?+
Four: Google AI Overviews (the biggest volume), ChatGPT (browse and memory), Perplexity (pure live retrieval) and Claude (with tool use). Each has different retrieval logic, so a page can be cited in one and absent in another. They must be measured separately.
How do you measure visibility in AI search?+
Build a prompt panel of 50–200 real customer queries. Run them weekly against all four surfaces. Record whether the brand was mentioned, whether it was cited with a link, citation position, and which competitor was cited instead. The single metric is Share of Citation — not rank position. Norrhavet publishes the methodology openly at /vad-ar/mata-ai-synlighet.
How long until a new page appears in AI answers?+
For live-RAG engines (Perplexity, ChatGPT browse, Google AI Overviews): hours to days after Googlebot has indexed the page. For training-corpus engines (plain ChatGPT, Claude without tool use): the next training cut, typically 6–18 months. Therefore live-retrieval citation is the fastest signal that a page is structured correctly.
How is AI search different from classic SEO?+
Classic SEO ranks documents in a list — the user clicks. AI search generates an answer and cites 2–8 sources — the user reads the answer in place. Optimisation shifts from keyword density and backlinks to question-answer structure, schema.org markup, named entities and authoritative source links. Backlinks still matter for authority; structure matters more for being the quoted sentence.