AI visibility audit: a practical 7-step framework

An AI visibility audit answers a question that classic rank tracking cannot: when a buyer asks an assistant what to choose, does your brand make the answer? The audit should test real purchase questions across several engines, preserve the cited URLs, compare your share of voice with competitors, and connect every recommendation to evidence.

This work builds on SEO rather than replacing it. Google's current guidance says its generative search features still rely on core Search ranking and quality systems, including retrieval from the Search index and query fan-out. At the same time, answer engines create a new measurement surface because the user receives one synthesized response instead of a list of ten blue links. Read Google's official AI search optimization guide for that distinction.

What should an AI visibility audit measure?

Measure outcomes first, then diagnose inputs. If the audit starts by scoring schema tags or word counts, it risks optimizing proxies without proving that any assistant actually names the brand.

The foundational GEO research described visibility as a measurable property of generative answers and found that optimization effects varied by domain. That is a strong reason to avoid universal scores with no prompt-level evidence. See the original GEO research paper.

Build the audit in seven steps

Define the buying journey

Start with the questions a real customer asks before purchasing. Cover discovery, comparison, fit, objections and final selection. A SaaS audit might include questions such as "best AI visibility platform for an SEO agency," "how to monitor brand mentions in ChatGPT," and "which GEO tool supports Spanish clients."

Do not create dozens of trivial rewrites. Google explicitly warns against producing separate pages for every possible query variation, and the same principle improves measurement. A smaller prompt set with clear intent produces a baseline that teams can understand and repeat.

Freeze the test conditions

Record the exact prompt, engine, model or product surface, market, language, date and whether live web search was available. Without this context, a later result is not comparable. A score changing from 42 to 51 means little if the second run used different questions or added another engine.

Capture the whole answer

For every engine and prompt, preserve four things:

• whether the brand was mentioned
• how it was described and positioned
• which competitors were named
• every cited or linked source

A simple yes or no loses too much information. A brand can be mentioned negatively, placed last, confused with another entity or cited through a third-party review rather than its own site. Those are different problems with different fixes.

Calculate outcome metrics

Use transparent formulas that a client can reproduce.

Never hide missing engine responses inside the average. A partial audit can look artificially strong if the engines where the brand performs poorly failed to answer.

Audit retrieval and crawlability

Now test whether engines can reach and understand the pages that should support those questions. Check HTTP status, canonical URL, indexability, server-rendered main content, internal links, sitemap inclusion and crawler access. OpenAI says publishers should allow OAI-SearchBot if they want content included in ChatGPT search summaries and snippets. Review the current OpenAI publisher guidance instead of relying on old bot lists.

Technical access is necessary, but it is not proof of visibility. A crawlable page can still be ignored if it repeats commodity advice, lacks evidence or does not answer the prompt clearly.

Diagnose content and authority gaps

Compare what the winning sources provide that your site does not. Look for first-hand data, clear definitions, pricing, methodology, expert authorship, current examples, product details and independent corroboration. Google's current guidance emphasizes unique, non-commodity content rather than mass-produced summaries.

If assistants repeatedly cite one third-party comparison, that URL belongs in the action plan. The fix may be digital PR or review coverage, not another paragraph on your homepage.

Prioritize and remeasure

Every recommendation should name the affected prompt, the evidence, expected impact, effort and owner. Then rerun the same sample after the change has been crawled.

What does a good audit deliver?

A useful deliverable contains the prompt matrix, raw cited URLs, outcome metrics by engine, competitor gaps and a prioritized backlog. It also states limitations. AI answers vary, personalization may affect results, and no third-party platform has access to an engine's private ranking system. Google explicitly advises caution around tools that claim to expose internal metrics.

SeoWave automates this evidence trail across multiple engines and turns it into a repeatable baseline. Review the public measurement methodology, compare the AI visibility workflow, or run a first scan from the homepage.