Short answer: an LLM visibility audit measures three different things: whether the brand appears in relevant answers, whether it appears accurately and which sources enable that visibility. The audit then connects those results to entity, content, evidence and external-authority gaps to build a roadmap.
If a readable page is still missing from the answers you observe, use the website not appearing in ChatGPT diagnosis guide to distinguish access checks, content gaps and the next evidence needed before investing.
Budget the visibility sample and technical investigation as named deliverables. The 2027 guide to technical GEO audit scope and fees helps distinguish the two and compare what a proposal will actually deliver.
Asking ChatGPT once “do you know my company?” is not an audit. The answer can vary by wording, context, time, market and engine. Useful analysis requires a stable question set, intent categories and a methodology that can be repeated over time.
1. Define the business objective before the queries
An audit for a hotel brand should not use the same questions as an industrial B2B company. Define which decisions matter: discovery, shortlist inclusion, provider selection, reputation, product, destination or support.
The objective determines the question set and prevents teams measuring visibility that could never create value.
2. Build five groups of questions
A useful baseline combines category, problem, comparison, selection and brand questions. Examples include “which companies provide X?”, “how do I solve Y?”, “X vs Z”, “best providers for…” and explicit questions about the brand.
This separates brand recall from generic-category discovery.
3. Add natural variations
Users do not phrase every question the same way. Include short, long, technical, conversational and sector-specific versions. You do not need hundreds of mechanical synonyms; you need to represent real research paths.
4. Separate languages and markets
A brand may be visible in English and absent in Spanish. Sources can differ too. Assess priority languages separately and add location where it changes the decision.
5. Record brand presence
The simplest metric is whether the brand appears. Define what counts: exact company name, product, domain or another unambiguous mention. Keep the rule consistent between audits.
6. Record competitor presence
An answer where you are absent becomes more diagnostic when you know who is present. Build a competitor matrix by topic cluster and identify recurring brands.
AI competitors do not always match the competitors the commercial team had in mind.
7. Analyse cited sources
When an engine exposes sources, record domain and URL. Classify owned source, media, directory, community, official documentation, review, research or competitor.
Patterns show which evidence types dominate the category and which sources you might legitimately build or earn.
8. Separate mentions from citations
Your brand may appear without the official site being cited. That means the entity exists in external sources while owned pages may not be winning citations. The reverse can also happen: an owned page is cited for sector information without the company being recommended.
Those outcomes require different actions.
9. Measure factual accuracy
Review country, services, audience, product, public pricing if mentioned, size where relevant, dates and any sensitive statement. Classify errors by severity.
Accuracy becomes especially important in regulated sectors or high-value buying decisions.
10. Evaluate semantic positioning
Presence alone is not enough. Which category does the engine associate with the brand? Agency, software, consultancy or industrial supplier? Does it mention the specialism you want to own?
An audit can reveal that an entity exists but is categorised in an unhelpful way.
11. Evaluate competitive context
When several brands appear, inspect why. Is one associated with price, another with specialisation and another with geographic coverage? These relationships reveal which attributes the information environment attaches to competitors.
12. Audit the canonical entity page
The home and About/Company pages should consistently explain name, category, services, markets, location and differentiators. If engines get basic facts wrong, start here.
13. Audit service pages
Every priority service needs a clear source: definition, scope, process, deliverables, FAQs, cases and boundaries. Check whether the page answers the questions in the monitored set.
14. Audit first-hand evidence
Inventory cases, data, studies, methodology, tools, authors and publishable experience. A brand with no evidence has less differentiated material competing for citations.
15. Audit external authority
Find independent sources that describe the company. Assess accuracy and freshness. Identify meaningful absences: associations, sector media, profiles, marketplaces, professional directories or publications where the brand should naturally exist because of real activity.
16. Audit entity consistency
Compare name, description, URL, locations, categories and services across the website, profiles and relevant sources. Fix contradictions before producing more content volume.
17. Audit retrievability
Review robots, sitemap, indexation, canonicals, HTTP states, rendering and internal links. An excellent page that engines cannot retrieve consistently cannot function as a source.
18. Audit important content trapped in weak formats
If the best answers exist only in PDFs, video, graphics or closed tools, create accessible text equivalents. Keep the original formats; add a web layer that explains and contextualises them.
19. Score by cluster rather than one universal percentage
A single “LLM visibility score: 42%” can hide too much. It is more useful to see discovery 20%, comparison 50%, brand 90% and problem 15%, for example. That connects the roadmap with intent.
The score should be an internal comparison tool, not a universal truth about an algorithm.
20. Turn findings into concrete actions
Every gap should create an action: improve a service page, publish a case, clarify an entity, create a comparison, correct an external profile, earn a legitimate mention, add a definition or fix a technical retrieval problem.
21. Prioritise by value, gap and evidence capacity
Do not publish an article simply because visibility is low for one question. Prioritise valuable queries where the company has something real to contribute and where the gap can be corrected.
22. Repeat the audit with a stable methodology
One measurement is a snapshot. Repeating the same set reveals trend. Version your question set and document major changes so a new methodology is not mistaken for improved visibility.
23. Combine LLM data with SEO and analytics
Search Console can show growth in brand and topic queries. Analytics can show engine referrals and later conversions. CRM can capture natural mentions of ChatGPT or Perplexity when leads are asked how they found the company. Combine signals rather than forcing one-source attribution.
24. Add an AI Overviews layer
For Google, include queries that trigger AI Overviews and record sources and brand presence. Our Google AI Overviews guide explains the main diagnostic signals.
25. Add a Perplexity layer
Perplexity makes sources easy to observe in many answers. Use that transparency to understand which pages win citations and which content types are missing. The Perplexity optimization framework develops this process.
26. Document errors and corrections
When you find wrong information, record which source may be contributing and which action is taken. Over time, this log shows recurring error types and influential sources.
27. Include human review
Automation can collect answers, but someone who understands the business must judge accuracy, relevance and context. A detected brand name does not automatically equal a positive or correct recommendation.
Deliverables of a useful LLM visibility audit
- Versioned question set by intent, language and market.
- Brand and competitor presence matrix.
- Cited-source map.
- Factual-error register.
- Entity and category-positioning assessment.
- Source-page and evidence gap analysis.
- Technical retrievability audit.
- Prioritised action roadmap.
- Baseline for repeat measurement.
At The GEO Company, we treat the audit as a decision tool. The goal is not a decorative scorecard; it is knowing what to change so generative systems have better sources about the brand.
Frequently asked questions about LLM visibility audits
Which engines should an audit include?
It depends on the market, but ChatGPT, Gemini, Perplexity and Google’s generative search experiences are common priorities. Add other engines when audience behaviour justifies them.
How many questions does an audit need?
An initial baseline can be useful with 30–100 well-selected questions. Companies with several product lines, countries or languages may need larger sets.
Can visibility scores be compared between tools?
Not reliably unless methodologies match. Tools define presence, engines, prompts and frequency differently. Use scores primarily to compare your own progress under a stable method.
How often should the audit be repeated?
Monthly or quarterly depending on market speed and the amount of work being deployed. Major brand or architecture changes can justify a new baseline.
How to present the baseline to leadership
Do not lead with one global score. Summarise three findings: where the brand is already understood correctly, which commercially important questions still exclude it, and which errors or source gaps need attention. Then present five prioritised actions with an owner and target date. The audit becomes valuable when it ends in decisions about content, reputation, technical retrieval or distribution.
Keep the raw answer archive available for specialists, but give leadership a stable trend view. Presence, accuracy and citation mix over time are easier to act on than dozens of isolated screenshots.
Keep a change history for the audit
A useful audit should also keep a change history for high-impact findings. Record when a source was corrected, a new entity page was published, an external profile changed or an inaccurate AI description disappeared. This turns repeated audits into a learning system rather than disconnected snapshots. Over several cycles, the team can see which kinds of evidence improve accuracy fastest and which visibility gaps require external corroboration rather than more owned content.