AEO Audit Process: How I Audit a Brand's AI Visibility

An AEO audit is a structured check of whether AI answer systems and search features can find, understand and accurately describe your brand, followed by a ranked list of fixes. My process has eight steps: define the buyer questions, take a baseline across engines, check crawl access, review answer-readiness of key pages, check entity clarity, compare third-party sources, find competitor gaps, and turn all of it into a prioritised plan. This is my own method, described as of 4 October 2026. I own this site, so treat it as one practitioner's approach, not a standard.

Key Takeaways

• Start with the questions buyers ask, not with a tool. The prompt set drives everything else.
• Measure visibility as a percentage across repeated runs. AI answers vary, so single screenshots mislead.
• Check basics first: indexation, snippet eligibility and bot access. Google says no special optimizations are required for AI Overviews or AI Mode.
• Audit what third parties say about you. AI systems draw on more than your own pages.
• Output is a ranked fix list with effort and expected impact, not a 90-page PDF.
• I do not guarantee any result from the audit. It tells you what to fix and how to track it.

What is being audited, and what is not

This audit looks at visibility in AI answers: ChatGPT, Google AI Overviews and AI Mode, Gemini, Perplexity, Claude and similar. It sits on top of an ordinary SEO check, which I treat as the first layer rather than something separate. For a view of what to look for in any provider, Savage Audit's pre-hire checklist lists technical foundations, content structure, entity clarity, schema, intent alignment and AI visibility baselines. My steps cover the same ground in a different order.

The 8-step process at a glance

Step 1: Build the buyer question set

I start with a conversation, not a tool. I collect real questions from sales calls, support tickets, search queries, community threads and comments, then group them by stage: problem, comparison, and decision.

For each question I write a few natural variants. SparkToro's study found human-written prompts for the same intent were highly varied, with a semantic similarity of just 0.081 in its tests, so one phrasing never represents a topic. DoodleWeb's guide also treats real buyer-question research as a mark of a genuine AEO provider.

Deliverable: a prompt list of roughly 30 to 60 questions, signed off by the client.

Step 2: Take a baseline across engines

I run each prompt several times on each engine, logging whether the brand is mentioned, whether it is cited with a link, what is said about it, and which sources appear.

Why repeated runs? SparkToro reported less than a 1 in 100 chance of the same brand list appearing twice across 2,961 prompts from 600 volunteers. Its authors considered visibility percentage a reasonable metric and exact ranking position unreliable. So I report ranges and dates, not "you are number two".

I cover the mechanics of a tracking set in my guide to AI visibility tools; the short version is that a tool can help but a spreadsheet is enough for a first baseline.

Deliverable: a baseline table by engine, with run counts and the date.

Step 3: Check crawl access and eligibility

Before touching content I check the basics.

• Indexation and snippet eligibility. Google's AI features documentation says pages must be indexed and eligible to be shown in Search with a snippet to appear in AI Overviews and AI Mode, with no extra requirements or special optimizations.
• robots.txt and meta controls. Check nothing important is blocked by accident, and that nosnippet or max-snippet settings are intentional.
• OpenAI bots. OpenAI's crawler documentation describes OAI-SearchBot as the bot that surfaces sites in ChatGPT search, and says sites that disallow it will not appear in its search answers, though they may still show as navigational links. GPTBot is separate and relates to training. Allowing one and disallowing the other is a legitimate choice, but it should be a decision, not an accident.
• Server logs. Where logs are available, I look for AI-related bot visits. My notes on this are in AI bot log file analysis.

Deliverable: a short access report listing blocked or broken items and the choices that need a business decision.

Step 4: Review page readiness

For the 10 to 20 pages that should answer the highest-value questions, I check:

1. Is there a clear, direct answer near the top?
2. Is the page's purpose and audience obvious?
3. Are claims specific, sourced and dated?
4. Is the structure scannable, with descriptive headings, tables where they help, and text rather than text hidden in images?
5. Is it written by an identifiable author or team, with a clear reason for existing? Google's helpful content guidance frames this as who, how and why.

Google's documentation also says to make sure structured data matches visible text. I do not add schema to chase citations; I add it where it accurately describes the page. Google's structured data intro recommends JSON-LD in most cases.

Deliverable: page-level briefs listing what to change, in order.

Step 5: Check entity clarity

I write a one-page fact sheet: the brand's name, what it does, who it serves, location, leadership, key products and claims that are verifiably true. Then I compare that against the website, social profiles, directories and what AI answers say.

Mismatches are common: old product names, wrong categories, outdated pricing. Structured data can help express identity; Schema.org's Organization type includes a sameAs property for linking to reference pages that identify the item. That helps clarity but does not guarantee a mention.

Deliverable: an entity fact sheet and a list of inconsistencies to correct.

Step 6: Audit third-party sources

When AI answers cite sources, I log them. Often the cited pages are review sites, comparison articles, community threads, news coverage or directories rather than the brand's own site.

I ask: which sources show up repeatedly for our questions, do they mention us, and is what they say accurate? SEO.com's red flags article lists authority-building through reviews, mentions and backlinks as part of legitimate AI SEO, and I agree it belongs in the audit.

I do not recommend manufactured reviews or planted mentions. Fixes are things like correcting a directory listing, pitching for an honest inclusion, or publishing original data worth citing.

Deliverable: a source gap list ranked by how often each source appears.

Step 7: Map competitors in the answers

From the baseline I list which brands appear for each question, how often, and in what framing. This shows where a competitor is repeatedly recommended and why: a strong comparison page, widespread reviews, or clearer positioning.

I use this to find gaps, not to copy. If a competitor wins a question because they publish original data and you do not, the fix is to earn your own evidence, not imitate their page.

Deliverable: a competitor map by question.

Step 8: Turn findings into a ranked plan

I score every finding on three dimensions: impact on a high-value question, effort, and confidence in the evidence. The result is a one-page roadmap.

I attach a reporting plan: re-run the prompt set monthly or quarterly with the same wording, report visibility as a percentage per engine, and connect it to leads or pipeline where possible.

Deliverable: a roadmap and a measurement plan.

What the audit does not do

• It does not guarantee citations. Outputs vary and the engines belong to other companies.
• It does not replace SEO. Fundamentals come first.
• It does not capture everything. Prompts are samples of real behaviour, not the whole of it.

For how to check whether a provider is promising too much, see my post on GEO agency red flags.

A starter version you can run yourself

1. Write 20 buyer questions in your customers' words.
2. Run each three times on two or three engines.
3. Log mention, citation, claim and sources in a spreadsheet.
4. Compute your mention percentage per engine.
5. List the sources that appear most often.
6. Check that your key pages are indexed and not blocked.
7. Fix the cheapest issues first and re-run in 30 days.

FAQ

What is an AEO audit?

It is a structured review of whether AI systems can find, understand and accurately cite your brand, ending in a prioritised fix list. It usually covers technical access, content readiness, entity clarity, off-site sources and a measured baseline.

How long does an AEO audit take?

It depends on site size and the number of prompts and engines. A focused audit of a small site with a 30-prompt set is realistic in a couple of weeks; larger sites take longer. Treat any fixed promise with caution.

Do I need special tools?

Not for a first baseline. A spreadsheet and repeated manual runs work. Tools help at scale, but they measure; they do not fix anything.

Do I need to add special schema or an llms.txt file?

Google's documentation says no special markup or AI-specific files are required for AI Overviews or AI Mode. Structured data is fine when it accurately describes a page, but it is not a shortcut.

Should I block GPTBot?

That is a business decision. OpenAI documents GPTBot as relating to training and OAI-SearchBot as relating to ChatGPT search visibility, and says each is independent. Many brands allow search bots and decide separately on training.

How many prompts do I need?

I start with 30 to 60 across problem, comparison and decision stages. More is better for reliability, but coverage of real buyer questions matters more than raw count.

How often should I repeat the audit?

A light re-run of the prompt set monthly or quarterly is common. Run a full audit after major site changes, rebrands or product launches.

Can an audit guarantee better AI visibility?

No. It identifies what you can control and gives you a way to measure change. Visibility still depends on competitors, the engines and variation between responses.

Do you offer this audit?

Yes, I run audits like this for edtech and startup brands. A short paid audit is how I prefer to start.

Want an audit for your brand?

If you would like this process applied to your brand, see my work at younusfardeen.in and send a note through the contact form. I bring 4+ years of marketing experience in edtech and startups, and I will tell you plainly what an audit can and cannot do before you commit.