AI Visibility Index Methodology
A transparent framework for measuring how automotive businesses appear, rank and receive citations in AI-generated answers to buyer questions.
Edition 1 rankings are not yet final. London Auto Index will publish precise scores only after the relevant prompt runs, platform metadata, repeat tests and quality checks can be reproduced from the evidence archive.
1. Purpose and measurement question
The Index is designed to answer one bounded question: how visible is an automotive business within a defined set of AI-assisted buyer journeys during a stated research window?
It does not measure vehicle quality, dealership service quality, financial strength or whether London Auto Index recommends a business. AI outputs are probabilistic and can change between platforms, model versions, sessions and dates. Each published edition will therefore be a time-bounded benchmark rather than a permanent rating.
2. Research scope
The proposed Edition 1 scope covers ChatGPT, Gemini, Claude, Perplexity, Grok and Google AI Overviews. The final edition record will state which platforms were successfully tested, the visible model or product version where available, the research dates and any access limitations.
Buyer questions are assigned a stable prompt ID and executed verbatim. Prompts are tagged by intent, buyer stage, vehicle or service category and geography.
Businesses are grouped into comparable categories such as prestige dealers, franchise networks and leasing brokers. Inclusion does not imply a commercial relationship.
The target design uses three repeat runs for each eligible prompt-platform combination to measure output volatility.
Each completed run must retain the response, timestamp, platform, prompt ID, extracted entities, recommendation positions, citation URLs and review status.
3. Standard test conditions
- Use the same verbatim prompt for every business eligible for that prompt.
- Use clean research sessions without personal chat history or memory where the platform permits.
- Record date, time, platform and visible model information.
- Do not insert dealer names into open discovery prompts unless the prompt category specifically tests branded recognition.
- Archive refusals, errors and incomplete answers rather than silently removing them.
- Separate organic answer mentions from advertisements, sponsored placements or London Auto Index editorial recommendations.
4. Proposed score components
| Component | Proposed weight | Definition |
|---|---|---|
| Mention Rate | 40% | The proportion of eligible completed responses in which the business is recognisably mentioned. |
| Recommendation Rank | 30% | A normalised position score that rewards earlier placement within an AI-generated recommendation set. |
| Citation Share | 20% | The proportion of eligible completed responses containing a supporting link to the business’s verified domain. |
| Consistency | 10% | The stability of the business’s presence across repeat runs of the same prompt-platform combination. |
Composite score = 0.40M + 0.30R + 0.20C + 0.10S
These weights remain provisional until the Edition 1 validation cycle is complete. Any change will be documented before final rankings are published; weights will not be altered retrospectively to favour a participant.
5. Eligibility and missing data
A business is scored only on prompts for which it is genuinely eligible by category, geography and buyer need. Platform errors, blocked requests and technically incomplete outputs are recorded separately. An edition will state the minimum completed sample required for publication. Businesses below that threshold will be marked as insufficient data rather than assigned a precise rank.
6. Quality assurance
- Entity names and aliases are normalised before aggregation.
- Ambiguous mentions are manually reviewed.
- Citation URLs are resolved to their final domain where practical.
- A second review is required for disputed or high-impact classifications.
- Aggregate totals must reconcile with the archived run-level records.
- Published scorecards must include sample size, research window and methodology version.
7. Independence and commercial separation
No business can pay for inclusion, placement or a higher Index score. Purchasing an audit, consultancy service or future monitoring subscription will not change the underlying benchmark result. Commercial clients and non-clients will be assessed under the same published rules when they belong to the same research cohort.
Pages describing businesses should use research cohort, business monitored or business covered unless a formal partnership has been agreed.
8. Limitations
AI answers are non-deterministic and may be influenced by model updates, retrieval systems, location, session context and live-web availability. The Index observes answer behaviour during a defined window; it cannot prove how a model internally reached an answer. A higher score does not establish trustworthiness or commercial superiority.
9. Publication, corrections and disputes
Each edition will identify its research window, methodology version, platforms completed, prompt count, run count, sample threshold and known exclusions. Material corrections will be dated. A business may submit evidence of an identification or classification error, but cannot purchase a score change.
Until the Edition 1 evidence archive passes validation, the public Index will show research status rather than precise rankings.