Research · Practitioner
State of GEO Report: H1 2027
The State of GEO report is MentionX's research on how brands appear in AI-generated answers: how often they're mentioned, recommended and cited, which sources engines rely on, and how this varies by category and engine. This page publishes the method in advance. Figures will be added only once the H1 2027 scans are complete and checked; nothing here is a result yet.
Why publish the method first?
Many statistics about AI search circulate without a method. We want readers to be able to judge ours: which prompts, which engines, how many runs, which definitions. Publishing the method before the results also stops us from adjusting it to fit a story.
What will the report measure?
| Metric | Definition |
|---|---|
| Mention rate | Share of answers naming a brand |
| Recommendation rate | Share of answers actively suggesting a brand for the need |
| Citation rate | Share of answers linking a brand's own domain |
| Share of voice | A brand's mentions ÷ all tracked brand mentions |
| Source mix | Share of citations by source type (own, review, media, community, reference) |
| Variation | How much answers change across repeated runs |
Definitions follow our glossary: AI share of voice, AI citation, visibility score.
Which categories, engines and markets?
Planned scope:
- Categories: a set of B2B SaaS and ecommerce categories.
- Engines: ChatGPT (search on), Google AI Overviews, Google AI Mode, Perplexity, Gemini. Google notes AI Overviews and AI Mode may give different responses (Google), so we report them separately.
- Markets: starting with English-language markets.
How will prompts be built?
From real buyer language (sales and support phrasing, community threads, question-shaped search queries) across learn, compare and buy intents. No branded prompts in headline figures. The prompt categories and sample prompts will be published with the report.
How will we handle variation?
Each prompt will be run multiple times per engine, in clean sessions with fixed location and language. We'll report ranges, not only averages, and note where differences are too small to be meaningful.
What won't the report claim?
- Causes we can't test (e.g. "engines prefer X") without supporting evidence.
- Traffic or revenue effects we didn't measure.
- Comparisons with other studies that use different methods.
How can you get it?
Once published, the report will be free to read here, with charts, the full method, and a downloadable data summary.
Sources
- Google Search Central, AI features and your website
- Aggarwal et al., GEO: Generative Engine Optimization, for the original framing of visibility in generative engines
Questions
When will the State of GEO report be published?
After the H1 2027 scans are complete and the data has been checked. Figures are published only from real scans.
What will the report measure?
Mention, recommendation and citation rates, share of voice, and cited source types across categories and engines, using a published method.
Can I use the data?
Yes, with attribution, once published. The method and prompt categories will be shared so results can be checked.