How to run a citation gap analysis for AI search

AI answer engines synthesise a response and attribute it to a handful of sources. Being one of those sources is the only visibility available; ranking fifth on the Google results page underneath an AI Overview that cites two other sites is worth very little. A citation gap analysis measures that directly: for each prompt a buyer might ask, who gets cited, and is it you.

The category matters. In September 2026 SerionFlow tested six buying-intent prompts through ChatGPT with web search and was cited zero times; the sources cited included self-published "best programmatic SEO tools" pages on competitors' own domains. The gap was not authority in the abstract but the absence of citable pages for those exact prompts.

Write twenty to fifty prompts the way a buyer would type them into an assistant: "what is the best tool for programmatic SEO for a SaaS startup", "compare X and Y for content teams", "which platforms integrate with Cloudflare for GEO". Draw them from Search Console queries that look conversational, from sales call questions, and from People Also Ask boxes. Freeze the set; changing prompts between runs destroys comparability. Add new prompts as a separate cohort.

Run each prompt in ChatGPT with search enabled, in Perplexity, and in Google with AI Overviews visible, ideally from a clean session. For every answer, record the engine, the prompt, the date, each cited URL, its domain, and the page type it is, such as listicle, comparison, documentation, or review. A spreadsheet with one row per citation is enough. SerionFlow's free citation gap finder structures this capture for a set of prompts and two domains so the comparison is immediate.

Filter to prompts where a competitor is cited and you are not. Group them by the page type that got cited. In most software categories the pattern is the same: self-published best-of lists, X-vs-Y comparisons, and pages that answer a question directly in the first paragraph get cited; homepages and feature pages rarely do. The gap list is therefore a list of page briefs, each specifying the prompt to answer and the format that engines are already rewarding.

Each brief becomes a page with the prompt as its H1, a forty-to-sixty-word direct answer at the top, a table with real numbers, an honest section on where you are not the right choice, FAQ content, and full text in the static HTML so AI crawlers can read it without JavaScript. Confirm the AI crawlers are allowed in robots.txt and at the CDN. Then submit through Search Console and, for Bing-backed assistants, Bing Webmaster Tools with IndexNow.

Re-run the frozen prompt set monthly, log citations the same way, and chart your citation count and share per engine. Movement is slow and noisy; three months is the minimum for a trend. Pair this with Search Console impressions for the same prompts phrased as queries, since assistant-issued retrievals show up there too. SerionFlow's paid plans include sampled AI visibility monitoring and, on Pro and Agency, AI-related query signals from Bing, which complement a manual gap analysis without replacing it.