What Should an AI Visibility Audit Include?
A transparent standard for AI visibility audit scope, prompts, technical tests, content, entities, measurement and prioritised actions.
A credible AI visibility audit should produce evidence, diagnosis and a prioritised action plan. It should not be a collection of screenshots or a single proprietary score.
The audit must show where the brand is discoverable, mentioned, cited and recommended; why gaps may exist; and what owners should do next. Every conclusion should trace back to a prompt, page, source or technical test.
1. Audit scope and decisions
The opening section should define:
- brands and entities;
- markets and languages;
- products or services;
- target audiences;
- engines and surfaces;
- competitors;
- date range;
- business decisions the audit supports.
State exclusions. A focused UK professional-services audit should not imply global consumer coverage.
2. Governed prompt set
Provide the prompt library, not merely a sample. It should cover:
- discovery;
- category education;
- comparisons;
- recommendations;
- objections;
- local or sector intent;
- branded accuracy;
- post-purchase or support where relevant.
Each prompt needs an ID, intent, funnel stage, market and rationale. Record the exact execution dates and product modes.
3. Baseline visibility
Report separate outcomes:
- retrieval or source appearance;
- brand mention;
- domain citation;
- sentiment or factual accuracy;
- recommendation;
- competitor presence;
- referral traffic where available.
Do not merge these into one score without publishing the formula. Use repeated runs for priority prompts and describe variance.
4. Technical accessibility
Test:
- robots.txt for relevant crawlers;
- CDN and WAF behaviour;
- status codes and redirects;
- canonical URLs;
- raw and rendered content;
- internal links;
- sitemaps and IndexNow;
- structured data validity;
- preview controls;
- server-log evidence where available.
The AI crawler audit contains the detailed technical procedure.
5. Content and citability
Map target prompts to pages. Review whether those pages contain:
- direct answers;
- original evidence;
- clear definitions;
- comparisons or procedures;
- claim-level sources;
- current facts;
- visible authorship and review;
- useful media and accessible text;
- limitations.
Identify gaps, overlaps and pages that should be consolidated rather than multiplied.
6. Entity and authority
Check the consistency of:
- organisation name and description;
- products and services;
- people and credentials;
- locations;
- regulator profiles;
- Organization and Person markup;
- authoritative sameAs links;
- independent coverage;
- review and community themes.
Separate facts the company controls from external corroboration.
7. Measurement readiness
Inspect whether the organisation has:
- Search Console and Bing Webmaster Tools access;
- GA4 AI-assistant channeling;
- prompt tracking;
- answer evidence retention;
- agreed metric definitions;
- reporting owners;
- a review cadence;
- incident escalation.
An audit that cannot be repeated becomes a one-off opinion.
8. Competitor analysis
Use the same prompts, markets and rules for every competitor. Compare source types and content patterns, not just mention counts.
Distinguish a true competitor from a publisher, directory or regulator that appears because it supplies evidence.
9. Prioritised roadmap
Every recommendation should include:
- issue;
- evidence;
- affected prompts or pages;
- expected mechanism;
- owner;
- effort;
- risk;
- dependency;
- target date;
- verification method.
Prioritise foundational access and factual errors before speculative formatting tests.
10. Executive summary and evidence appendix
The executive summary should explain the few decisions leaders need to make. The appendix should preserve prompts, runs, URLs, screenshots or exports, technical tests and methodology.
A strong audit is both understandable and reproducible.
Red flags in an audit proposal
- guaranteed rankings or citations;
- a hidden prompt set;
- no repeated runs;
- no technical checks;
- a universal score with no formula;
- recommendations unrelated to findings;
- no competitor methodology;
- no evidence appendix;
- a sales deck presented as research.
Frequently asked questions
How long should an audit take?
Scope determines time. A small focused audit may take weeks; a multi-market enterprise audit takes longer. The provider should state sample sizes, reviews and dependencies.
Does an audit need paid software?
No. Software can improve scale and history, but official reports, manual testing and technical evidence can support a valid focused audit.
What happens after the audit?
Convert priorities into a 90-day implementation backlog with owners, baseline metrics and a fixed retest schedule.
