How to Run an AI Citation Gap Analysis Against Your Competitors

An AI citation gap analysis is a structured comparison exercise: you run the same set of queries through ChatGPT, Perplexity, Claude, and Google AI Overviews, record which sources get cited for each one, and identify where a named competitor appears and your site does not. The output is not a vague sense that 'competitors do better at AI search'. It is a query-by-query list of exactly which answers you are missing from, which makes the next step, fixing it, tractable instead of guesswork.

Most teams that adopt AEO start with a solo audit: score their own site, fix the signals that are missing, and hope citations follow. That works, but it skips a faster diagnostic. If a competitor is already being cited for the exact queries you care about, they have effectively run the experiment for you. Their cited page tells you what a passing answer looks like in your category, right now, on the models you are trying to win.

Why a gap analysis beats a solo audit for prioritisation

A solo AEO audit tells you what is broken on your own site: missing schema, blocked crawlers, thin author attribution. It does not tell you which of those fixes actually matters for the queries your buyers ask. A gap analysis adds that missing layer of evidence. When a competitor without stronger domain authority than you is cited and you are not, the difference is almost always a specific, fixable signal, and comparing the two pages side by side usually reveals it within minutes rather than requiring you to guess across nine possible categories.

Step 1: Build your comparison query set

Start with 15 to 25 queries that map to your actual buying journey, not generic head terms. A narrow, high-intent list produces a more useful gap analysis than a broad one, because every query should be one your business genuinely wants to win.

• Definition and category queries: "What is [your category]?"
• Direct comparison queries: "[Your product] vs [named competitor]"
• Best-of and shortlist queries: "Best [your category] tools for [use case]"
• Problem-first queries: "How do I solve [problem your product solves]"
• Pricing and evaluation queries: "How much does [category] typically cost"

Step 2: Run the queries across each AI engine

Test the same query set across ChatGPT, Claude, Perplexity, and Google AI Overviews or AI Mode, in a fresh session each time so prior conversation history does not bias the answer. Perplexity is the easiest engine to start with because it shows numbered citations directly alongside every answer, making the source list unambiguous.

1. Submit each query to all four engines in a new, unauthenticated or freshly cleared session
2. Record whether your brand appears at all, and whether it is cited with a link or only mentioned by name
3. Record which named competitors are cited instead, and in what position (first, second, third source referenced)
4. Note the exact URL cited for each competitor mention, not just the domain
5. Repeat monthly, on the same date, so model updates and content changes are comparable over time

Step 3: Score the gap, not just presence or absence

What "cited" actually means

Treat a linked citation, a named-but-unlinked mention, and total absence as three different outcomes, not one. A competitor who is mentioned by name without a link is a weaker threat than one with a clickable citation appearing first in the answer, and your remediation priority should follow that distinction. Log all three states in your spreadsheet: query, engine, your status (cited / mentioned / absent), competitor status, and competitor URL.

Recording competitor advantage

For every query where a competitor is cited and you are not, open both the competitor's cited URL and the page on your own site that should have won that citation. If you do not have a page that targets the query at all, that itself is the finding, and it points to a content gap rather than a technical one.

Step 4: Diagnose why competitors are winning citations

Once you have a list of lost queries paired with the competitor page that won them, the diagnosis usually falls into one of a small number of recurring causes. Work through the cited competitor page against each of the following before assuming the gap is unfixable.

• Missing or incomplete schema markup, particularly FAQPage or Article schema, on your equivalent page
• AI crawlers blocked or restricted via robots.txt while the competitor allows GPTBot, ClaudeBot, and PerplexityBot full access
• Content that buries the direct answer under narrative framing, while the competitor leads with a 40 to 60 word answer
• No dedicated page for the query at all, a content gap rather than an optimisation gap
• Weaker entity clarity: no consistent author, Organisation schema, or third-party corroboration compared with the competitor
• Stale content: your page has not been meaningfully updated in over a year while the competitor page was recently refreshed

Step 5: Prioritise fixes by leverage, not by query count

Do not treat every lost query as equal. Group your findings by root cause rather than by individual query, because a single fix, such as adding FAQPage schema to a page template, often closes several gaps at once if the same missing signal recurs across many lost queries. Prioritise the root cause that appears most frequently across your gap list, then move to content gaps that require a new page, then to lower-frequency technical issues.

Turning gap analysis into a repeatable process

A single gap analysis is useful once. A monthly cadence is what actually moves your citation share, because competitor pages change, models update, and new queries enter your category as buyer language shifts. Pair the manual protocol above with an AEO readiness score on both your pages and the competitor pages winning your gaps, so you have a leading indicator between each round of manual testing rather than waiting a full month to see whether a fix worked. If your gap list keeps surfacing the same three or four competitors across every query, it is also worth reviewing your entity signals directly, since a consistently stronger entity footprint can explain a pattern of losses that no single page-level fix will close.