Exactly what a GEO Audit tests
A Generative Engine Optimisation audit measures your presence inside AI assistant answers the way a customer would actually encounter you — then backs every figure with the verbatim text that produced it.
Your URL is enough
You give us a domain and a few business details. From there we build the whole measurement instrument — no lengthy onboarding, no hand-fed keyword lists.
- We classify your business, category, and audience from your site
- We identify your real market competitors
- We generate the buyer questions a genuine prospect would ask
- You can review and override the panel before we run it
Buyer questions, not vanity prompts
We never simply ask “is [your brand] good?”. We ask the neutral, category-level questions your buyers ask — where you have to earn the mention.
Before we measure AI visibility, we check the foundations
Technical Visibility Foundation is a deterministic preflight performed before the AI audit begins. It checks whether technical conditions could affect how search and AI retrieval systems discover or interpret your website.
We examine signals including crawl accessibility, indexability, crawler access, sitemap configuration, canonicalisation, structured business data, language and market signals, internal discoverability and basic content accessibility.
The purpose is not to produce another SEO score. It is to identify technical context that should be understood before poor AI visibility is interpreted as a content, authority or competitive problem.
- Crawl accessibility
- Indexability
- Crawler access
- Sitemap configuration
- Canonicalisation
- Structured business data
- Language and market signals
- Internal discoverability
- Basic content accessibility
- Machine-readable identity consistency
AI Buyer Shortlist Snapshot
When AI is directly required to choose five businesses for the buying need, does your brand make the shortlist — and who is selected instead?
Two frozen, brand-blind buying questions are run three times across each of the three audited engines. A response only counts when it contains exactly five identifiable businesses in explicit ranked order, with no duplicates.
If your brand is absent from a valid list, a controlled follow-up asks for evidence-based context. Those answers are reported as model-stated explanations, never as proof of what caused the omission.
Typical position uses the median when selected. Competitor appearance and exact selection reasons remain evidence-linked.
Six dimensions of buyer-journey visibility
These broad-panel measures remain separate from the Shortlist Snapshot. Each is computed from valid, analysed samples only and reported with its reliability so a thin result is never dressed up as a solid one.
Brand mention rate
How often each engine names your brand at all, per question and overall — the baseline of AI visibility.
Recommendation rate
How often the assistant actively recommends you, not just mentions you in passing. Always a strict subset of mentions.
List position
When the answer produces an explicit ranked shortlist, where you land — reported as a median across repeats, with best-to-worst spread.
Competitive share of voice
Which competitors appear, how frequently, and in which buyer questions they beat you to the answer.
Sentiment & consistency
Whether you're described positively, neutrally, or negatively — and whether that description is consistent across engines.
Cited sources
The third-party domains cited in AI answers about your category — the citation opportunities where competitors are appearing.
From technical foundations to improvement priorities
- 1 Technical Visibility Foundation
- 2 Buyer-question research
- 3 AI Buyer Shortlist Snapshot
- 4 Multi-engine AI measurement
- 5 Competitor and source analysis
- 6 Findings and improvement priorities
Built for rigour, not screenshots
| Principle | What it means for your report |
|---|---|
| Repeated sampling | Every question is asked multiple times per engine, so we can report stability — not a single lucky answer. |
| Valid-sample denominators | Frequencies are computed only over answers that actually returned and analysed cleanly; failures are disclosed, never hidden. |
| Evidence-locked findings | Every positive mention or recommendation links to the verbatim sentence that supports it. Unsupported claims are rejected. |
| Per-field provenance | Each figure is labelled by how it was determined — observed in the text, deterministically derived, or model-inferred. |
| Correlational only | We describe what the AI said, never claim we know why. No fabricated causation. |
| Engine separation | Results are never pooled misleadingly across engines — a strong ChatGPT rank and a weak Gemini rank stay visible as distinct facts. |
A forwardable, client-ready report
Executive verdict
A plain-language summary of where you stand, safe to put in front of a decision-maker.
Question-by-question breakdown
Won, contested, mentioned-only or lost — for every buyer question, with the competitors present.
Evidence appendix
The exact quotes behind the numbers, each tagged with its provenance and the engine that produced it.
Prioritised actions
Correlational, evidence-anchored recommendations — coverage gaps, cited sources, engine differences.
Data table (CSV)
Every sample, one row each, with full audit identity — for your own analysis.
Limitations, stated plainly
Sample sizes, failures, and reliability flags — so the report is honest about its own confidence.
See where your brand stands in AI answers
Request an audit and we'll show you how ChatGPT, Perplexity, and Gemini describe you today — with the evidence behind every figure.