Norrhavet measurement · Q3 2026 · CC BY 4.0
AI Visibility Q3 2026 — Swedish B2B, measured on a frozen question panel.
During Q3 2026, 30 frozen Swedish B2B questions were run against several OpenAI and Google models. Norrhavet was mentioned in 32% of 200 recorded answers, and 12.5% of the answers linked to a norrhavet.com URL. Every figure on this page comes straight from the measurement database and is available as raw data under CC BY 4.0.
Headline figures
Computed 2026-08-17 from the panel. Raw data: https://norrhavet.com/rapporter/data/q3-2026.json
Results per engine
| Engine | Answers | Share of citation | URL hit rate |
|---|---|---|---|
| ChatGPT (GPT-5 mini) | 100 | 32% | 0% |
| Gemini (3 Flash) | 100 | 32% | 25% |
Only engines with recorded answers are reported. Perplexity and Claude are not part of Q3 2026 because we did not run them during the period — they are omitted rather than estimated.
Quick answers
What does Norrhavet's AI visibility report measure?
Two things per quarter: share of citation, meaning the share of AI answers that mention Norrhavet, and URL hit rate, meaning the share of answers that link to a norrhavet.com URL. The basis is a frozen panel of Swedish B2B questions run unchanged through the whole quarter against several models.
Which models are included in the measurement?
Models from OpenAI and Google in two size classes each, executed through the Lovable AI Gateway: GPT-5 mini, GPT-5, Gemini 3 Flash and Gemini 2.5 Pro. Perplexity and Claude are only included in periods where active keys exist for them; otherwise they are reported as not measured rather than estimated.
Why freeze the prompt panel?
If the question list changes between quarters you are measuring changes in the list, not in visibility. The panel is therefore frozen once per measurement period and stamped in the database. Changing it creates a new panel under a new key instead of rewriting the old one.
Is the raw data open and auditable?
Yes. Per-question outcomes are published as JSON at https://norrhavet.com/rapporter/data/q3-2026.json and generated from the same tables as the text on this page, so table and raw data cannot drift apart. The licence is CC BY 4.0 with the attribution "Norrhavet".
How these numbers should not be read
- The measurement covers Norrhavet's own visibility on a Swedish B2B panel — it is not a market share and not an industry median.
- Model answers vary by account, region and model version. Your own replication may land a few percentage points off.
- The panel is small in statistical terms. Quarterly movements of a few percentage points should be read as noise.