AI citation gap analysis: a framework for finding where competitors are cited and you are not

AI citation gap analysis is the practice of mapping which third-party sources AI models cite when recommending your category, comparing your citation footprint to competitors, and prioritizing which missing sources to earn next. It treats citations, not rankings, as the unit of work. The output is a ranked hit list of a few dozen domains where your competitors appear and you do not, scoped to the 33% of cited sources that are actually influenceable.

Why citation gaps are the right unit of analysis

Ranking positions no longer determine whether AI recommends you. Citations do. Ahrefs' analysis of ChatGPT's top 1,000 citations found that 67% come from sources marketers cannot directly influence, Wikipedia, government domains, and Reddit, leaving roughly 33% of citable surface area as actionable territory (Ahrefs). Seer Interactive's study of 541K AI responses across 20 brands showed that when a brand is mentioned in a cited source, the brand's content citation rate reaches 53.1%; when not mentioned, it drops to 10.6% (Seer Interactive). The presence or absence of your brand inside a cited source is a bigger lever than anything you can do to your own pages. Citation gap analysis is how you find those sources systematically instead of guessing.

What counts as a citation gap

A citation gap is a specific, named source that AI cites for your category where a competitor appears and you do not. Not a missing backlink. Not a keyword you are not ranking for. A source in an AI answer, a G2 category page, a Reddit thread, an industry listicle, a review comparison article, a trade publication feature, that mentions one or more competitors while omitting you. Gaps come in three flavors: absent (the source exists and never mentions you), thin (you are mentioned once in passing while competitors get paragraph-length coverage), and stale (an older mention that AI no longer surfaces because the source has been updated without you). The taxonomy matters because each gap type requires a different close tactic, from earned media pitches to refresh requests to outright new coverage.

Parse's position in this workflow

Parse tracks AI visibility across ChatGPT, Google AI Overviews, Perplexity, and Claude so the citation graph behind each answer is visible, not inferred. That matters for gap analysis because a citation gap is only actionable when you can see exactly which URL AI surfaced, which competitor was inside it, and which prompt triggered the retrieval. Without that visibility, teams fall back to proxy metrics, backlinks, mentions, SOV, that correlate weakly with what AI actually cites. The Citations tab is where gap analysis stops being guesswork and starts being a ranked list a content lead and a PR lead can split between them.

Inputs you need before running the analysis

A useful gap analysis needs five inputs locked before you open the spreadsheet. First, a prompt set that maps to revenue, 50 to 150 prompts grouped by buyer-stage and category, averaged across multiple runs to filter model noise. Second, a named competitor set of three to eight brands; more than eight dilutes the signal. Third, per-platform citation data for each prompt, not an aggregate, BrightEdge found only 38% agreement between ChatGPT and Google AI Overviews on brand recommendations, so platforms must be read separately (BrightEdge). Fourth, a classification of every cited domain as influenceable or off-limits. Fifth, a 90-day calendar that has capacity allocated for gap-close work before the analysis runs, not after.

The four-step methodology

Steps one and two are data work. Step three is the analysis. Step four is where most teams fail, because they either ship a 400-row spreadsheet no one reads, or they ship a list with no owners and no dates. The method is only useful if the output of step four fits on one page and names a human for each row.

How to classify a source as influenceable or off-limits

The 67/33 split is the single most important filter in this entire workflow. An off-limits source, Wikipedia, a government site, a Reddit thread dated 2022, a Quora answer with 11K upvotes, might cite a competitor, but you cannot practically move it in a 90-day window. Including those rows in the active hit list wastes capacity. Classify every domain in your citation set against this rubric: earned-media (you can pitch a journalist), marketplace-or-directory (you can claim, optimize, or request updates), review-platform (you can drive verified customer reviews), comparison-or-listicle (you can outreach the author for inclusion), community (indirect influence through authentic participation), or structural (Wikipedia, Wikidata, government, long-horizon entity work only). Earned-media, marketplace, review, and comparison surfaces absorb most of a realistic 90-day plan. Muck Rack's analysis found 82% of AI citations come from earned media and 94% from non-paid sources, which is consistent with where actionable gaps live (Muck Rack).

Weighting by platform, vertical, and prompt value

Not every gap is worth the same. TryProfound's source analysis shows the ranked citation leader flips by platform: Wikipedia is ChatGPT's top source at ~7.8% of citations, while Reddit leads AI Overviews and Perplexity (TryProfound). A gap on G2 matters more for B2B SaaS than for ecommerce; a gap on a trade publication matters more for regulated industries than for consumer tech. AmICited's study of review platforms puts G2 alone at 22.4% influence across ChatGPT, Perplexity, and AI Overviews, and the top five review platforms collectively at 88% of all review-based AI citations (AmICited). Weight each gap by three multipliers: platform share (what percent of AI traffic in your category comes from the platform citing the gap), vertical fit (how often your category is the subject), and prompt revenue value (is this a top-funnel category prompt, a middle-funnel comparison, or a bottom-funnel buy-intent prompt). Close the high-weight gaps first.

The prioritization scorecard

A workable gap score combines reachability and impact on a single rubric. Rate every gap on three dimensions, 1–3 each, and sort by total:

A score of 7–9 means close this quarter. A score of 4–6 means next quarter or pipeline for earned-media work. A score of 3 or below means parking. This is coarse on purpose, the point is a ranked list a content lead and a PR lead can act on Monday, not a weighted multi-factor model that takes a week to re-run.

If you want to see which sources shape AI answers about your brand, run a free brand check — it takes a minute.

Closing gaps: the four most common tactics

Once the list is ranked, the tactics fall into four buckets. Profile-and-listing work covers G2, Capterra, Gartner Peer Insights, Clutch, GetApp, and the vertical marketplace your category uses, claim, fill, optimize, and drive review velocity. Comparison and listicle inclusion means reaching out to the author of every "top 10" or "best X" article a competitor is in, offering a specific reason and quotable data for your inclusion. Digital PR and earned media means pitching journalists and trade-publication editors with data, quotes, or original research that puts your brand inside the coverage AI is already citing. Community presence means building authentic, sustained participation on the subreddits and Q&A threads AI pulls from; this is slower but compounds, and Idea Grove's research puts earned-media share of AI brand mentions at 91%, this is where the volume lives (Idea Grove). Each tactic maps to a specific gap type; do not run PR for a listings gap.

How AI retrieval shapes which gaps matter

A gap only matters if AI actually surfaces the source for prompts in your category. ALM Corp's work on ChatGPT retrieval shows only 15% of retrieved pages make the final answer and 85% are evaluated and discarded; 32.9% of cited pages came from fan-out searches rather than the original prompt, and 89.6% of prompts trigger two or more follow-up queries (ALM Corp). Two implications for gap analysis. First, the source must survive reranking, thin mentions rarely do, which is why "thin" gaps often matter more than "absent" ones once you score them. Second, the fan-out layer means sources AI cites for adjacent queries also shape which brands it is aware of at all, so your gap scope should include one-hop adjacent prompts, not just your primary category set. Princeton's GEO study (KDD 2024) found that adding citations, quotations, and statistics to a page can lift visibility by up to 40%, confirming that content-inside-cited-sources is the structural lever (Princeton, KDD 2024).

Running gap analysis as a quarterly operating loop

Gap analysis is not a one-time audit. Run it quarterly, in a predictable rhythm, tied to your planning cycle. Week one of the quarter: pull the full citation matrix and refresh the competitor set. Week two: classify and score. Week three: brief the content lead and the PR lead with a one-page ranked hit list, names next to rows. Weeks four through eleven: execute, with a weekly check against your regular AI visibility review. Week twelve: re-pull the citation matrix, attribute wins and losses, close the loop. The quarterly cadence matches how AI models actually refresh their source preferences, Perplexity updates in days, ChatGPT with browsing in hours-to-days, and ChatGPT base training in three to six months, so a quarter is the shortest window where you can expect all platforms to reflect your work. Link gap close to the AI visibility scorecard so the movement shows up in the same metric your executives see.

Ghost citations and why "absence" is the wrong default assumption

Before declaring any gap, check whether AI is already using your content without naming your brand. Seer Interactive calls this a ghost citation, the model pulls facts, framing, or data from your page but never mentions the brand behind it (Seer Interactive). Their study found content citation rates fall from 53.1% to 10.6% when the brand is not mentioned in the source itself. If your gap analysis lists a source as absent but the page is actually cited for the prompt, the fix is not new placement, it is making your brand visible inside the content being cited. See our ghost citations breakdown for the diagnostic steps. Skipping this check leads teams to chase placements they already have structurally, just without brand attribution.

The top citation sources and what they imply for most gap lists

Sources: Visual Capitalist, TryProfound. The takeaway: most gap lists correctly weight Reddit, review platforms, trade publications, and comparison content heavily, because that is where the influenceable 33% of citation surface actually lives. For a deeper source-by-source breakdown see which domains AI models cite most, Parse's own data on the source domains behind AI answers, and how YouTube stacks up against Reddit as a cited source.

Common failure modes to design out

Three failure modes account for most gap-analysis projects that die by quarter two. First, the mega-spreadsheet: a 400-row list with no prioritization that no human ever works through. Fix: cap the quarterly hit list at 15–25 rows. Second, unowned rows: a ranked list with no name next to the row. Fix: every row gets a single owner, content, PR, community, or product, before the hit list is approved. Third, re-running the analysis before the last close cycle completes: a new matrix every month generates churn, not progress. Fix: one full run per quarter, with weekly checkpoints against the existing list rather than new analyses. Gap analysis is most valuable when the same ranked list survives contact with execution for a full 90 days.

If you want a live view of which sources AI is citing for your category, and which of those sources include competitors but not you, see your citation gaps.

FAQ

How is AI citation gap analysis different from a traditional SEO backlink gap?

A backlink gap tells you which sites link to competitors but not to you. A citation gap tells you which sites AI actually pulls into answers when users ask category questions. The two are correlated but not identical. Seer Interactive's work shows content can be cited without brand mention, and vice versa. Citation gaps matter because they drive the answer AI generates, not the ranking a human scrolls past. Backlink gaps are still useful as a leading indicator; citation gaps are the trailing metric of what AI is already doing in your category.

How many prompts and competitors do I need to run this?

Most teams start with 30–80 prompts and three to eight competitors. Below that, the signal is too sparse to prioritize. Above 150 prompts or more than eight competitors, the analysis becomes unreadable in a quarterly rhythm and teams stop acting on it. Prompts should cover top-of-funnel category questions, mid-funnel comparisons, and bottom-funnel buying prompts in roughly equal measure. Competitors should be the brands AI is already citing alongside yours, which is itself a data question your gap analysis will answer.

How do I know if a gap is closeable in 90 days?

Use the reachability-effort-impact rubric. A 7–9 score means yes. Profile and review-platform gaps almost always score 7+ because the claim path is direct. Earned-media gaps in trade publications typically score 5–7, depending on how warm the relationship is. Wikipedia, government, and long-tenured Reddit threads score 3–4 and belong in the long-horizon entity track, not the quarterly close list. If a gap scores below 4, it is a structural problem that requires a six-to-twelve-month program, not a quarter.

Should I include off-limits sources in the analysis at all?

Yes, but only to calibrate the addressable universe. Knowing that 67% of a competitor's citation set is Wikipedia, Reddit, and government domains tells you two things: that the competitor has strong entity authority and that your quarterly close list should focus elsewhere. Off-limits rows belong in the matrix as context, not as action items. The practical rule is to keep them visible but move them to a separate tab so they do not absorb planning time.

How often do AI models refresh their source preferences?

It varies by platform. Perplexity uses real-time retrieval, so changes in cited sources can appear in days. ChatGPT with browsing updates in hours to days for cited content. ChatGPT's base training cycles take three to six months. Google AI Overviews typically refresh within weeks. A quarterly gap-close cadence is the shortest window where all four platforms can reasonably be expected to reflect your work, which is why quarterly is the operating rhythm rather than monthly.

How do I prove the gap work is paying back?

Tie gap close to a top-line metric that executives already see: category Share of Model, aggregate AI citation count, or brand-mention frequency in cited sources. Measure the pre-quarter baseline and the post-quarter read on the same prompt set, scored the same way. The noise band (5–10 points week-over-week) should be set before you start so wins and losses are distinguishable from model variance. Report both the metric move and the concrete source moves, "added to 4 of 7 target listicles, profile optimized on G2, 2 trade pitches landed", so leadership sees the mechanism, not just the number.