What we compared, and why it matters for GEO
We compared five AI engines on how they cite UK brands and sources: mention rate, citations per answer, source diversity and favoured domains, using one identical prompt set. This matters because generative engine optimisation is not one game. An engine that cites many sources rewards different tactics from one that names few.
Citation behaviour, in this context, means the pattern of which brands an engine names and which sources it points to when it does. Two engines can answer the same question and reach for entirely different evidence, which is exactly what makes a side-by-side comparison useful rather than academic. This piece sits alongside our generative engine optimisation pillar guide, which covers the discipline in full.
ChatGPT vs Perplexity vs Gemini: the citation comparison
The table below compares each engine on brand mention rate, average citations per answer, source diversity and how often it links out, for the same UK prompts. It shows which engine surfaces the most brands and the most sources.
| Engine | Brand mention rate (% of answers naming ≥1 UK brand) | Avg. citations per answer | Source diversity (unique domains) | Links out? |
|---|---|---|---|---|
| ChatGPT | <!-- DATA: % --> |
<!-- DATA --> |
<!-- DATA --> |
<!-- DATA --> |
| Perplexity | <!-- DATA: % --> |
<!-- DATA --> |
<!-- DATA --> |
<!-- DATA --> |
| Google Gemini | <!-- DATA: % --> |
<!-- DATA --> |
<!-- DATA --> |
<!-- DATA --> |
| Claude | <!-- DATA: % --> |
<!-- DATA --> |
<!-- DATA --> |
<!-- DATA --> |
| Google AI Overviews | <!-- DATA: % --> |
<!-- DATA --> |
<!-- DATA --> |
<!-- DATA --> |
Caption: <!-- CONFIRM: sample size + date range -->. Measures citation behaviour, not answer quality.
Which domains do AI engines cite most?
AI engines lean on a repeated set of trusted sources, and the mix differs by engine. This section reports the most-cited domain types, and named domains where notable, across the study.
| Cited-domain type | Share of citations | Engines that favour it |
|---|---|---|
<!-- DATA: e.g. review platforms --> |
<!-- DATA: % --> |
<!-- DATA --> |
<!-- DATA: e.g. reference/encyclopaedic --> |
<!-- DATA: % --> |
<!-- DATA --> |
<!-- DATA: e.g. news/press --> |
<!-- DATA: % --> |
<!-- DATA --> |
<!-- DATA: e.g. directories --> |
<!-- DATA: % --> |
<!-- DATA --> |
<!-- DATA: e.g. brand-owned --> |
<!-- DATA: % --> |
<!-- DATA --> |
<!-- CONFIRM: cited-domain breakdown after study runs -->. See our companion post on where AI answers come from for the mechanics behind this.
Methodology
We ran one identical set of UK prompts through five AI engines from a UK location and logged, per answer, the brands named, the citations shown and the domains linked.
- Engines (5): ChatGPT, Perplexity, Claude, Google Gemini, Google AI Overviews.
- Prompts: UK buying/research prompts held constant across all engines for a fair comparison.
<!-- CONFIRM: publish full prompt list --> - Location: United Kingdom.
<!-- CONFIRM: locale/settings --> - Dates: . Each prompt run times per engine to reduce variance.
- Measured per engine: brand mention rate, average citations per answer, unique cited domains (source diversity), whether links are shown, and most-cited domains.
<!-- CONFIRM: exact definitions --> - Tooling: AI-answer capture via NeuralGen’s SE Ranking account plus manual verification.
<!-- CONFIRM -->
Why do the engines cite so differently?
The engines differ by design. Some are built as answer engines that show sources by default, others summarise without linking, and each draws on a different index and set of trusted domains. So citation behaviour reflects architecture, not just content quality.
An engine that is architecturally built around retrieval, fetching and citing live sources at answer time, will naturally show more citations than one that relies more heavily on what it learned during training and summarises without linking out. Neither approach is inherently better for a user; they are simply different, and they demand different tactics from anyone trying to be cited.
What this means for your GEO strategy
The practical takeaway is to match tactics to engine behaviour: earn citations in the domains a given engine trusts, and structure content so retrieval-based engines can quote you cleanly. One approach does not fit all five.
Concretely, that means auditing which engines your buyers actually use, checking which domain types each one favours from the table above, and prioritising the citation sources that matter most for the engines your audience is on, rather than treating “AI visibility” as a single undifferentiated target.
Our AI SEO agency service applies this engine-by-engine approach for clients.
Limits and how often we repeat this
Engines update frequently, so citation behaviour is a moving target. Results vary by prompt phrasing, date and personalisation, and our prompt set is a sample, not the whole web.
We re-run this study periodically and republish with a fresh date <!-- CONFIRM -->. Check the “last updated” date on this page before citing a specific figure from it.