What is an outsourced AI visibility audit, and how does it work?
An outsourced AI visibility audit is a service that puts numbers on whether AI answers cite your brand, in what context they mention it, and how its share compares with competitors, and then presents improvement priorities. The process runs in this order: defining what to measure, designing prompts, repeated measurement on each engine, and a metrics report.
AI search engines don't list web pages by rank; they synthesize answers directly. So even when a category question in ChatGPT or Perplexity brings up only competitor brands, conventional search ranking tools can't detect that gap. Maeil Business Newspaper (Maekyung) reported that clicks are falling as the search box turns into a 'command box.' When a brand fails to make it into the body of AI answers, it loses that much of its traffic path.
What is an AI visibility audit?
An AI visibility audit measures how often ChatGPT, Perplexity, Gemini, and Claude cite and recommend a brand in their generated answers, so it measures something different from an SEO audit, which tracks link rankings. AI answers have no link rankings; a brand is either mentioned or left out.
| Item | SEO audit | AI visibility audit |
|---|---|---|
| What it measures | Keyword rankings, traffic volume | Citations in the body of answers, citation context, mention ratio relative to competitors |
| Form of results | Link rankings on search results pages | Answers synthesized in sentences |
| Improvement direction | A page's link authority and keyword density | Getting engines to recognize brand content as a trusted, structured information source |
The audit turns the degree to which engines recognize brand content as an information source into numbers.
What steps does the audit follow?
The audit runs in four stages, and each stage's results become the input for the next.
Stage 1: Define the brand's current status and what to measure First, set the brand's category, 5–7 main competitors, and the target purchase journey stage. You can only build the list of questions to measure once you've decided which questions the brand should appear in the answers to.
Stage 2: Design the prompt set Write 15–25 questions that resemble how real users ask AI search engines. Leave out brand-name searches and build the set from category and feature questions such as "Recommend a hydrating serum for dry skin" and "Compare B2B SaaS contract management tools." Only questions like these let the measurements reflect actual visibility.
Stage 3: Measure four engines in parallel Enter the same prompts into each of the four platforms (ChatGPT, Perplexity, Gemini, and Claude) at least five times and collect the responses. Because each engine cites differently, results are tallied separately by engine.
- Perplexity: Shows source links in its answers.
- ChatGPT: Often works brand names into the body of the answer as part of a sentence.
- Gemini: Draws on content from the Google Search index as the basis for its answers.
- Claude: Adds sources only to answers that use web search.
Stage 4: Calculate the four-tier metrics and write the report Tally the collected responses into four metrics: Presence, Prominence, Sentiment, and Source Ecosystem. A single measurement carries a lot of noise, so the first measurement serves as the baseline, and weekly repeated measurement adds the trend.
Brands under industry regulation, such as Korea's prior review of medical advertising, need a different approach from the prompt design stage onward. Questions centered on diagnosis names don't produce valid measurements, and reviews can't be used as content material either. What hospitals and clinics can do is covered separately in GEO services for hospitals and clinics: Working within medical advertising rules.
What the audit checks
The audit checks whether brand content can serve as a training and real-time search source for AI engines, across three areas: technical, content, and competition.
Technical check
-
Whether an
llms.txtfile exists and whether AI crawler access is allowed - How far JSON-LD structured data has been applied (schema.org Product, FAQPage, Organization, and so on)
- Whether server-side rendering is applied (the problem of AI crawlers being unable to read content in SPA environments)
- Consistency of meta information (OG tags, canonical)
Content format check
- Definition sentences: whether there is a sentence defining the entity in the form "[Brand] is ~"
- FAQ: whether real user questions and direct answers appear in the body
- Comparison and table content: whether there is comparison information AI can excerpt as is
- The proportion of citable authoritative source links included
Share of Voice relative to competitors
- Comparison of mention frequency between your brand and 5–7 competitors on the same prompt set
- Distribution of share by category and by purchase journey stage
- Analysis of the content types through which competitors get cited (reverse-tracing which formats and sources AI answers adopt)
What the audit report metrics mean
The audit report is built on four-tier metrics, and strategic priorities emerge only when you combine them.
| Metric | What it measures | Interpretation |
|---|---|---|
| Presence (presence rate) | The share of responses, across all measured prompts, in which the brand is mentioned at least once | Close to 0% means AI engines don't recognize the brand, so fixing the technical and content foundations comes first |
| Prominence (salience) | When mentioned, whether the brand appears early in the answer or near the top of recommendation lists | High Presence with low Prominence means the brand is mentioned but buried on the periphery |
| Sentiment (context quality) | Whether the brand is cited in a positive, neutral, or negative context | "Expensive but effective" and "disappointing for the price" are both mentions, but their strategic implications differ |
| Source Ecosystem (source ecosystem) | The types of sources AI relies on when citing the brand | Where to invest in content depends on whether the source is your own website, media articles, or review platforms |
What to check when outsourcing the audit
Many services advertise AI visibility audits, but the rigor of their measurement methods varies widely. Check the items below before you commission one.
| Item to check | What it should meet | Why |
|---|---|---|
| Platform coverage | Lists the engines it will measure and the number of repetitions per engine in its proposal | A partial audit that measures only some engines reflects only part of the AI search landscape and can't serve as a basis for strategy design |
| Prompt set defined and shared in advance | The list of questions for the audit is shared and agreed on before commissioning | Changing the prompts after measurement makes comparison with earlier measurements impossible |
| Repeated measurement and trend reports | Averages after collecting at least five repetitions, and provides trend data at least once a week | AI engines generate a new result with each response, even for the same question |
| Fixed competitor comparison set | Fixes 5–7 competitors and measures repeatedly with the same prompt set | If the comparison targets change every time, changes in share can't be interpreted |
| Separating ads from organic citations | Proposes organic optimization and advertising separately | ChatGPT ads appear in a separate sponsored area outside the body of the answer |
The claim that running ads gets a brand cited in the body of AI answers is not accurate, so screen out agencies that explain it that way.
PION provides parallel measurement across four engines (ChatGPT, Gemini, Google AI Overviews, and AI Mode) and regular trend reports as its default package. Other AI engines, such as Claude and Perplexity, can be added. For teams new to AI visibility audits, it is efficient to start by handing it all over at once, with brand status definition and prompt set design included in the initial consulting.
To check how far your internal preparation has come before you commission an audit, use the AI answer visibility basics check.
Frequently asked questions
How can I get an AI visibility audit?
You commission a specialist GEO/AEO agency. The agency first sets the categories and competitors to measure and writes 15–25 real user questions. It enters these questions repeatedly into the four engines (ChatGPT, Perplexity, Gemini, and Claude) to calculate the brand citation rate and Share of Voice. When choosing an agency, check for coverage of four or more platforms, agreement on the prompt set in advance, and trend reporting.
How is an AI visibility audit different from a conventional SEO audit?
An SEO audit measures search rankings and traffic volume by keyword, while an AI visibility audit measures whether a brand is cited in the body of AI answers, the context of those citations, and its mention ratio relative to competitors. AI search engines don't list links by rank but synthesize answers in sentences, so conventional SEO tools can't detect this gap.
Which AI engines should be measured in the audit?
Measure several engines, such as ChatGPT, Perplexity, Gemini, and Claude. Each engine cites differently. Perplexity shows source links, ChatGPT works brand names into sentences in the body text, Gemini reflects Google indexing status, and Claude adds sources only to answers that use web search. Measuring only some engines reflects only part of the AI search landscape.
What metrics can I check from the audit results?
You receive four-tier metrics: Presence (the rate at which the brand is mentioned), Prominence (position within the answer), Sentiment (positive or negative classification of the citation context), and Source Ecosystem (the types of sources AI relies on). The first measurement serves as the baseline, and weekly repeated measurement data tracks how the numbers change after content is revised.
What items must I check when outsourcing an AI visibility audit to an agency?
Check platform coverage (the engines to be measured and the number of repetitions per engine), a prompt set defined and shared in advance, averages calculated after at least five repeated collections, trend reports at least once a week, and a fixed comparison set of 5–7 competitors. The claim that ChatGPT ads can earn citations in the body of answers is not true, so choose an agency that proposes organic optimization and advertising separately.
- Category
- Guide