What Is an AI Visibility Audit? (And What's Actually in One)

An AI visibility audit queries ChatGPT, Claude, Gemini, and Perplexity with real buyer prompts about your brand category. Every AI response is captured verbatim. Citations are mapped. Brand facts are checked against what AI actually says. You get a report that shows exactly where you appear, where you don't, and what to fix.

I built Loamly's audit product after running hundreds of manual checks and realizing no one was delivering this at scale. Marketing teams want to know what AI says about them. The supply of one-time, deep-dive audit reports was essentially zero.

Want to see your baseline first? Free AI visibility check at loamly.ai/check, 15 prompts, no credit card.

---

What monitoring tools don't tell you

Most content about "AI visibility" describes monitoring dashboards, products like Otterly, Peec AI, or Ahrefs Brand Radar that track your score over time. Those tools are useful. But there's a gap.

They tell you that your score changed. They don't tell you what AI is actually saying, word for word, or what specific things to fix.

From analyzing 2,014 brand reports in the Loamly database (as of Feb 15, 2026), I know what the competitive distribution looks like: 85.5% of companies score 0-20. They're in the invisible tier. A monitoring tool showing them a score of 4 out of 100 doesn't tell them what's wrong. An audit does.

Here's what a proper AI visibility audit contains.

---

1. Buyer intent prompt research

Before running a single query, you need the right questions. The audit maps the buyer journey for your category, awareness prompts ("what is [category]"), consideration prompts ("best [category] tool for SaaS"), and decision prompts ("[Competitor] vs [Brand] pricing").

Decision-stage prompts matter most. That's where buyers choose. Most companies I audit appear in awareness-stage results, AI knows they exist, but vanish entirely at the decision stage. AI knows they exist. AI doesn't recommend them.

---

2. Verbatim AI response capture

This is the core. Every prompt runs on each platform, ChatGPT (GPT-4o with web search), Claude (web search enabled), Gemini (grounded), and Perplexity, and every response is captured word for word.

Not a score. The actual text.

This is the only way to find hallucinations. AI platforms misstate pricing, attribute competitor features to your product, repeat outdated claims from articles published years ago, and sometimes invent facts that have no source. You can't fix what you haven't read.

---

3. Sub-query capture

When ChatGPT or Perplexity search the web to answer a question, they generate internal search queries first. These are sub-queries: the actual keywords AI uses to find information about your category.

A single buyer prompt generates 3-7 sub-queries. A 50-prompt audit produces 300+ sub-queries total. This dataset doesn't appear in Google Search Console. It doesn't show up in Ahrefs. It's a direct window into what AI considers relevant for your category, in real time.

For SEO teams, this is the most immediately actionable output, a content targeting brief built from what AI is actually searching for. Ahrefs' analysis of AI traffic patterns documents how AI platforms are increasingly performing their own search queries rather than relying solely on training data.

---

4. Citation chain mapping

Every URL any AI platform cites in response to your prompts gets logged, tagged by domain, and analyzed. For a 50-prompt Professional audit, that means 300+ citation records.

The citation map shows:

• Which domains AI trusts in your category
• Which of your own pages get cited (usually not your homepage, AI cites specific blog posts and feature pages)
• Citation gaps: prompts where AI answers with no citation at all, which is where hallucinations are most likely
• Your citation share vs. named competitors

The Princeton GEO study (KDD 2024) showed that the content AI cites directly changes what it says. Sources cited in AI responses have 40%+ more influence on outputs than uncited sources. Knowing which URLs AI is citing for the prompts you're losing turns a vague problem into a specific task.

---

5. Brand accuracy audit

AI makes things up. Not intentionally: it synthesizes training data and live web content, and when that content is incomplete or contradictory, it fills gaps.

The accuracy audit checks 15-30 facts across platforms: product pricing, core features, company size, founding date, key integrations, customer use cases. Each fact gets a status: accurate, partially correct, incorrect, or missing.

For each inaccuracy, the audit traces the source. Usually it's an outdated article that AI trusts more than your own site, or a competitor comparison page that stated something about your product that AI repeated as fact.

The source identification is what makes it fixable. A pricing hallucination isn't fixed by updating your homepage: it's fixed by correcting or de-indexing the specific source AI is pulling from.

---

6. GEO technical assessment

GEO (Generative Engine Optimization) has a technical layer. The audit scores 50 pages of your website on:

• Schema markup: Organization, Product, FAQPage schemas
• llms.txt: a machine-readable summary for AI crawlers (the standard is documented at llmstxt.org, and Loamly's GEO guide covers implementation)
• Answer-first content: whether pages lead with the answer or bury it
• Citation-ready statistics: specific, sourceable data points
• Authority signals: Wikidata entity, Google Knowledge Graph presence
• Crawl accessibility: whether AI bots can access key pages

The GEO technical score affects citation probability. From the 2,014-company dataset, leaders average a GEO score of ~65 vs ~50 for emerging companies. It's not the primary differentiator, brand authority is, but it sets the floor.

---

7. Competitor battle cards

The audit identifies which competitors AI recommends instead of you for each prompt category, then explains why: what content they have, which sources cite them, what facts AI believes about them.

You get 5-10 competitor cards. Each shows which prompts they win that you lose, their citation sources, and the specific content gaps you could fill to compete.

---

8. 90-day action plan

The audit ends with a prioritized action plan: specific tasks ordered by estimated visibility impact. Not "publish more content", "publish an article answering [this specific question] because it appears in 8 prompts you're losing, and the competitor ranking for it has one Wirecutter mention that you could replicate."

Specific. Sequenced. No generic advice.

---

What an audit costs

Tier	Price	Delivery	Prompts	Platforms
Snapshot	$299	24h	20	3 (ChatGPT, Claude, Gemini)
Professional	$990	48h	50	4 (all platforms)
Enterprise	$2,490	48h	100	4 + strategy call + 90-day re-audit

One-time payment. No subscription.

The Professional tier is most popular because of the CSV exports. Three files: all 200+ verbatim AI responses, all 300+ sub-queries, all 300+ citations. For any team that wants to run their own analysis or feed the data into a content pipeline, that raw dataset is the value.

---

Limitations

These reports are a point-in-time snapshot. AI platforms update their training data and search results frequently, a citation that appears today may not appear in 60 days. The Enterprise tier includes a 90-day re-audit to account for this.

The audit captures what's happening now. It doesn't predict what changes.

---

Run the free check at loamly.ai/check to see your baseline in 3 minutes. If the results show you're in the 85.5% scoring below 20, the audit tells you exactly why and what to fix first.

No marketing spin. Just real data about your AI visibility.