AEO & AI Search Published September 27, 2026 · By Nabin Thakur 8 min read

AEO Citation Gap Analysis: How to Audit Your Brand in AI Answer Engines

Answer engine optimization starts with a measurement most brands never take: which questions an AI answer names you for, which ones it names a competitor for, and why. That list of gaps is a citation gap analysis. Below is the method I run, and what it showed when I ran it on this site in September 2026, including the parts that did not go my way.

Short Answer

What is an AEO citation gap analysis? A list of the questions your buyers ask AI tools, with a record of who each answer names and which pages it cites. A gap is any question where a competitor shows up and you do not.

The useful part is sorting every gap by why it exists. A missing page, missing authority, a missing listing and a wrong description of your business look the same in the answer, and need four different fixes.

The 30-second version

Start with Search Console. Some of the questions people ask AI tools already show up there as queries.
Test signed out, with personalisation off. My logged-in and signed-out results disagreed.
Record four things per answer: named, cited, who instead, and what it said about you.
Sort gaps by mechanism, not by engine. That is what tells you the fix.

Why “AI visibility” is not one number

Most AI visibility tools give you a score. The score hides the only thing you need to know, because the answer engines do not pick sources the same way. In my own tests I saw three different mechanisms.

1. Live retrieval rewards the page itself. ChatGPT and Perplexity search the web and read pages when they answer. Depth and specifics on the page win. Signed out, Perplexity used this site for the opening definition in its answer to “how do I recover from the google helpful content update”, ahead of much larger sites.

2. Google’s AI Overviews lean on authority. They draw on Google’s own index. For “how to recover from google helpful content update”, checked signed out from India with personalisation off, this site ranked first in the normal results and was among the AI Overview’s sources. Another site still held the lead attribution. Ranking first did not buy the lead citation.

3. “Who should I hire” answers come from lists. When I asked Perplexity who the best agencies for Google penalty recovery are, every source it cited was a roundup or ranking site. Not one agency’s own website. Good content does not get you into that answer. Being on the lists does.

Step 1: Build the question list from real searches

Start with Search Console. Google now passes some AI-generated queries into the Performance report. They look odd: long prompts, sometimes with location instructions inside them, and short queries starting with “%” or “+”. The second kind are fan-out queries, the sub-searches an AI system runs on the user’s behalf.

Over the three months to September 2026 I found 42 of these on this site: 19 templated prompts, 20 fan-out queries and 3 long buyer prompts. Together they were about 2% of impressions and got no clicks. Their value is not traffic. It is that they show you the exact questions being asked. One was a buyer asking an assistant for “actual agency names” for recovery after a Google update.

This page exists because of them. Fan-out queries about “aeo citation gap analysis” kept showing one of my service pages at positions between about 3 and 10, which meant the AI was looking for a guide and finding a sales page.

Then fill in the four shapes buyers use. How do I fix X. Why did X happen. Who should I hire for X. What does X cost. Write 15 to 25 questions in your customers’ own words, not your keywords.

Step 2: Run every question clean

Google: add &pws=0 to the search URL. It switches off personalisation. It does not change your location.
ChatGPT: use Temporary Chat, which ignores memory and custom instructions.
Perplexity: sign out.

This matters more than it sounds. Early in September a logged-in Perplexity session cited this site for a helpful content update question. The same question signed out cited only Google’s documentation and Semrush. If I had trusted the first run, I would have reported a win that was my own browsing history.

Run the questions that matter more than once, on different days. AI answers change between runs. One run is a snapshot, not a measurement.

Step 3: Record four things per answer

Field	What to write down
Named?	Is your brand in the answer text?
Cited?	Is your page among the sources, and is it the lead source?
Who instead?	Every brand named, and every domain it cited.
What it said about you	Copy the sentence word for word. This is where the surprises are.

The last field caught my most useful finding. I asked Perplexity for an affordable recovery specialist for a UK small business. It named this site first of three, then added: “no reliable public recovery price found”. A UK competitor in the same answer showed a monthly price. The cause was mine: my US dollar prices only appeared after JavaScript ran, so the page the AI read showed rupees. I added a plain-text dollar price line the same day.

Step 4: Sort every gap by why it exists

Gap	What you see	What fixes it
Content	The answer to a how or why question comes from other pages. You have nothing that answers it directly.	A page that answers that exact question, first-hand and specific.
Authority	You rank, but Google’s AI Overview credits bigger sites first.	Mentions and links from sources Google already trusts. This one takes time.
Listing	“Best X” and “who should I hire” answers cite roundups, directories and marketplaces. You are not on them.	Get onto those lists: review directories, roundups, marketplace profiles.
Representation	You are named, but described wrongly: wrong price, wrong market, an old title.	Fix what machines can read: static text, schema, one consistent name. Then wait for the recrawl.

Two details from my own runs. A Google AI Overview answering a hiring question cited Upwork among its sources, so marketplace profiles belong in the listing column. And one answer described this site as convenient “if you are based in Guwahati”, while Search Console showed 59% of the site’s AI exposure coming from outside India. That is a representation gap, and no amount of new content fixes it.

What my own audit found, September 2026

Question and engine	Result	Gap
“how do I recover from the google helpful content update”, Perplexity	Cited for the opening definition	None
“how to recover from google helpful content update”, Google	First in the normal results; an AI Overview source, not the lead	Authority
India, recovery agency shortlist, Perplexity*	Named first of seven	None
UK small business, affordable specialist, Perplexity*	Named first of three, with “no reliable public recovery price found”	Representation (fixed)
“why did my website traffic drop after a google core update”, Perplexity	Not cited	Content
“who should I hire to recover my google rankings”, Google	Absent from the AI Overview and the results	Listing

*Run on my own logged-in account, so personalisation may have helped. Treat those two rows as indicators, not results.

The pattern is the useful part. The site wins “how do I fix this” questions and loses “who should I hire” questions. Those are answered from different documents, and only one of them is mine to write.

Add the Search Console AI report

Search Console now has a report for Google’s generative AI features, reached from a banner on the Performance report. For this site, 2,160 of about 13,800 impressions in the three months to 20 September were in AI features, 15.7% of the total. 59% of that AI exposure came from outside India.

It counts impressions only. It tells you where you appear inside AI features, not whether you were the answer’s recommendation, and not whether anyone clicked. Use it to choose which questions to test by hand, not as the result.

What a citation gap analysis cannot tell you

Cited is not recommended. Being one of the sources is not the same as being the answer’s choice.
It says nothing about clicks. In our seven-site Search Console study, AI exposure tracked lower click-through while ranking position did not.
Answers move. Rerun the same questions, the same way, every quarter. The comparison is the point.