A citation gap is a question your buyers ask AI assistants where the answer cites other sources and not you, even though you have, or should have, the best page on the topic. To find them, run a fixed set of buyer questions through ChatGPT, Gemini, Perplexity and the other assistants your buyers use, log who gets cited, and compare that with what GA4 and Search Console say about your pages. Every gap then falls into one of a few types, and each type has its own fix.
GEO here means generative engine optimization: being named and cited in AI answers, not geographic targeting. For the wider picture of what gets a B2B company recommended, start with our guide to getting recommended by AI assistants. This post is the audit that sits underneath it.
What is a citation gap analysis?
It’s a content audit with a different question. A classic SEO audit asks “which pages rank, and which don’t?” A citation gap analysis asks “for the questions that lead to revenue, who do the assistants cite, and why not us?”
You need three inputs:
- A prompt log: what assistants answer and cite for your buyers’ questions.
- GA4: which pages already receive AI-referred visitors, and whether those visitors convert.
- Search Console: which pages rank for question-style queries and how their clicks are trending.
None of these is complete on its own. Assistants vary their answers, GA4 only sees the clicks, and Search Console doesn’t separate AI Overview impressions from ordinary results. Together they’re good enough to prioritize.
Step 1: How do you build the prompt set?
Write down 20 to 50 questions your buyers actually ask before they buy. Pull them from sales calls, demo forms, support tickets and Search Console queries, not from a keyword tool alone. Cover four kinds:
- Category: “best [category] software for [industry]”.
- Comparison: “[your product] vs [competitor]”, “alternatives to [competitor]”.
- Problem: “how do I [job your product does]”.
- Fit and price: “is [your product] good for [use case]”, “how much does [category] cost”.
Run every question in each assistant your buyers use, in a clean session, and record: whether you’re mentioned, how you’re described, which sources are cited, and which competitor appears instead. Keep the questions fixed and repeat the run monthly. Answers vary between sessions, so trends across the set matter more than any single screenshot.
Step 2: What does GA4 tell you about AI traffic?
GA4 can’t see an assistant reading your page, but it does see people who click through from one.
- Create an “AI assistants” channel. In Admin, add a custom channel group with a channel whose session source matches a regex such as
chatgpt|perplexity|gemini|copilot|claude, and place it above Referral. ChatGPT tags its links withutm_source=chatgpt.com, which makes it the easiest to catch. - Find the landing pages. In Explorations, build a free-form table with the AI channel as a filter, landing page as rows, and sessions, engaged sessions and key events as values.
- Read what it means. Pages that already receive AI referrals are being cited somewhere. Pages that convert well from AI traffic are the ones worth protecting and expanding. Pages with strong organic traffic and no AI referrals at all are candidates for a citation gap.
Two caveats. Some AI app traffic arrives without a referrer and lands in Direct, so treat the AI channel as a floor, not a total. And make sure bot traffic isn’t inflating your numbers; our guide to bot filtering in GA4 covers that. If your GA4 setup itself is in doubt, fix that first through our GA4 consulting service.
Step 3: What does Search Console add?
Filter the Performance report for question-style queries (containing how, what, best, vs, cost) and look at:
- Pages with high impressions and falling click-through rate for informational questions. AI Overviews often answer those directly, so the page may be used without being clicked, or replaced by another source.
- Comparison and “alternatives” queries where you get impressions but rank low. These are usually where a competitor’s page frames the answer.
- Queries with no page of yours at all. These become content gaps.
Your server or CDN logs complete the picture: they show whether crawlers such as OAI-SearchBot actually fetch the pages you expect to be cited.
Step 4: How do you sort the gaps?
Put every question from the prompt set in one row, with the evidence from steps 1 to 3, and assign it a type:
| Gap type | What you see | Fix |
|---|---|---|
| Cited and clicked | Mentioned in answers; AI referrals in GA4 | Protect: keep facts current, add depth, strengthen internal links |
| Ranked but not cited | Good organic ranking; assistants cite others | Restructure: answer first, question-shaped headings, specifics, tables, FAQ |
| Described wrongly | Mentioned, but with old or wrong facts | Fix the source: one consistent company description across your site and profiles |
| Competitor owns it | A competitor’s comparison page is cited | Write your own fair comparison or “alternatives” page |
| Not covered | No page of yours answers the question | Create a page for that one question |
| Not reachable | No crawler hits in logs; content rendered only by JavaScript | Fix robots.txt, firewall rules and rendering first |
Prioritize by revenue, not volume: a comparison question your buyers ask before signing matters more than a broad definition question with ten times the searches.
Step 5: What makes a page citable?
Assistants lift passages, not pages. The fixes for “ranked but not cited” are mostly editorial:
- Answer in the first two or three sentences. No warm-up.
- Use question-shaped H2s with the answer right below.
- Be specific: prices or “from” prices, limits, integrations, regions, and a real “updated” date.
- One page, one question. Split pages that try to answer everything.
- Tables for comparisons, a short FAQ for the rest, and structured data (Article, FAQPage, Organization) so the facts are machine-readable.
- Link related pages together, so the site reads as one body of expertise rather than isolated posts.
This is the same discipline that helps ordinary search and conversion. For Hemptouch, analytics fixes and a redesigned blog took the conversion rate from 0.5% to 2% in six months; clear, well-structured content pays off whoever reads it.
Step 6: How do you measure whether it worked?
Set a baseline before you change anything, then track monthly:
- The prompt log: how many of your questions mention or cite you, and whether descriptions are accurate.
- The AI channel in GA4: sessions, key events and, if you pass the source into your CRM, qualified leads.
- Branded search in Search Console, and direct visits to deep pages, which often follow an AI mention that wasn’t clicked.
- A “How did you hear about us?” field on demo forms. In B2B it often picks up “ChatGPT” before any tool does.
For measuring how assistants talk about you, not just whether they cite you, see tracking AI sentiment with GA4.
FAQ
Is there a GEO tool that connects to GA4 directly?
You don’t need one to start. A custom channel group in GA4 captures AI referrals natively, and a spreadsheet of fixed prompts covers citations. Third-party AI-visibility tools can automate the prompt runs, but check how many answers they sample before trusting their scores.
How often should we run a citation gap analysis?
Run the full audit once, then repeat the prompt log and GA4 checks monthly. Re-run the whole audit when you launch a product, change positioning or publish a batch of new pages.
Can a page be cited without bringing any traffic?
Yes, and often. An assistant may use your page to answer and the user never clicks. That’s why the prompt log and branded search matter alongside GA4 clicks.
Who should own this: SEO, content or analytics?
All three touch it. In our experience it works best when one person owns the prompt log and priorities, with analytics supporting the measurement and content making the fixes.
If you’d like us to run the audit on your site, get in touch.
