Reading AI citations

Interpret AI citation sources, analyze linked and unlinked brand mentions, identify realistic gaps, and move qualified URLs into Prospector.

Use Citations, under an individual report’s AI tab, to see which pages supported captured AI answers, which platforms cited each source, whether the page mentions the tracked brand, and which non-owned sources may be realistic outreach targets.

A citation means an AI platform referenced a source. It does not automatically mean the source mentions the brand, links to the client, endorses the brand, or caused the answer.

Before you start

The project needs a completed AI scan from at least one platform that tracks citations. If every enabled AI platform is configured for mentions only, the AI citations view is unavailable because no citation bucket is being measured.

Start with Reading AI answers when you need the answer text that produced the citation set.

Read the source table

Open the relevant project, choose AI, then Citations.
Select the platforms you want to include.
Review each source’s URL, title, platforms, best citation position, and weighted score.
Check whether the URL is the primary domain, another owned asset, or an external source.
Open citation history, changes, or volatility when you need evidence across multiple completed runs.

The source score is weighted by platform reach and citation position; it is not traffic, authority, or a guarantee that the page influenced an answer. A source cited by several platforms can rank above a one-platform source even when both appear once.

Analyze what the cited page says

Citation-page mention checks start when you choose Check mentions. Completing a scan or opening a report does not start page analysis. Unchecked pages show Not checked yet; that is unknown brand presence, not proof that the brand is absent. Existing findings keep their original check dates.

AI-answer brand detection and citation collection remain automatic during a scan. Checking the cited page is separate: it fetches the page and checks for the brand, recognized owned links, source type, opportunity status, and available sentiment evidence.

Open More actions and choose Check mentions to analyze current sources without findings.
After processing, use the filters to narrow the table.
Choose Recheck mentions in More actions to refresh existing findings. Review and confirm the displayed usage impact first.

Manual checks use pooled on-demand usage. Already analyzed canonical URLs are skipped when checking missing pages. Eligible evidence from this project’s traditional results can be reused without fetching the page again, and the reverse is also true. Reuse is limited to the last seven days and evidence recorded since the project’s brand settings last changed.

While a run is working, each source being analyzed shows Processing… in place of its mention state and the table updates itself as results land, so you can tell which rows this run covers without reloading the page. The status belongs to the run you started: it does not reappear after a reload, and it does not carry across keywords, locations, or the traditional results table.

Recheck mentions is different: it re-checks the current citation sources and updates results that have gone stale, including a false opportunity for a page that is now linked. Rankability shows the source count and relative usage impact for you to confirm first, and pauses when a pooled on-demand window reaches its limit.

Re-analysis covers exactly the sources the table is showing — the selected keyword and AI location, using each platform’s latest result. It does not reach URLs that appear only in older history, and confirming a later refresh starts a new on-demand attempt rather than repeating the one you already approved.

Read the resulting states as follows:

State	Meaning
Linked	The page mentions the brand and links to a recognized owned domain or page.
Linked elsewhere	The page names the brand as the text of a link, but the link points to another site (or to a section of the same page). It is a mention, not a backlink, and an easy outreach ask: point the existing link at the brand’s site.
Unlinked	The page mentions the brand but no recognized owned link was detected.
Opportunity	The analyzed page does not mention the brand and is classified as a realistic external target.
Gap	The analyzed page does not mention the brand, but it may not be a practical outreach target.
Not available	The page could not be reached or analyzed, such as a paywall, login wall, or temporary failure.
Not checked yet	No citation analysis exists for this source yet.
Owned	The URL matches the primary domain or the client’s owned-asset registry.

Read the source classification

Alongside its mention state, a cited source carries three short labels: what kind of page it is, how its location compares with the client’s, and the next action it suggests.

The page kind covers a directory list, a directory profile, a best-of list, an information page, a community thread, a competitor’s own site, a lead-generation doorway, a page the client owns, and an off-topic page. The location label reads as a match, the same state, the wrong city, another location, or unknown.

The action follows from the two: Get listed on a directory that omits the client, Earn inclusion in a best-of list, Join the conversation in a community thread, Earn a mention on an information page, or Create or claim profile where a competitor has one and the client does not. A source already covered reads Client listed or Client owned, and one to leave alone reads Competitor, Ignore doorway, Wrong location, or Off topic. Needs verification means the classification is not settled; open the page before acting on it.

Treat these labels as triage. They are derived from the page Rankability read, so confirm the page still says what the label implies before you spend outreach effort on it.

Use the filters

Start with Source evidence:

Actually cited is the default. It shows pages referenced in the selected AI answers.
Retrieved but not cited shows pages returned in exposed search-source lists that were not cited in those selected answers. Expand View retrieving queries on a page to see the associated searches. View related queries means the provider returned a query group or answer-level query set without identifying which individual query returned the page.

New captures preserve available source evidence for Grok, OpenAI, Claude, Gemini, and Perplexity. Claude links search results to their search calls. Gemini exposes grounding references and the answer passages they support; these references may not include every internally retrieved page. Perplexity exposes source lists and answer citation references, but its internal search queries may be unavailable.

Older Claude, Gemini, or Perplexity scans may show Saved sources (unverified) because their citation status was not preserved. These saved lists remain accessible. Retrieval evidence unavailable means unknown, not zero retrieved pages. If any selected answer has incomplete citation evidence, Tracker cannot verify a retrieved-only list for that selection. Select a platform with complete evidence or run a new scan. Partial source lists are identified. These are observable sources, not a guarantee that every internal search is exposed.

Retrieved-only pages do not contribute to citation scores, citation history, alerts, or citation-page mention checks. A page retrieved by one selected platform but cited by another belongs in Actually cited. Change the platform selection to inspect each platform separately.

Export CSV follows the selected evidence view. The retrieved-only export includes each page’s available queries, query association, platforms, and source metadata such as dates or snippets when returned by the provider.

Within Actually cited, the existing mention filters remain available:

All shows every cited source in the selected platform scope.
Has mention shows linked, linked elsewhere, and unlinked brand mentions.
Gaps shows analyzed pages that do not mention the brand.
Opportunities narrows gaps to non-owned sources classified as realistic outreach or content targets.

An opportunity is a prioritization aid, not a promise that the publisher will add a mention or that an AI platform will continue citing the page.

Keep owned assets accurate

Rankability automatically recognizes the primary domain and can recognize other assets registered for the client. Use the shield action to mark or unmark an off-domain page or domain as owned when the automatic classification is incomplete.

Owned classification affects how Rankability presents citation opportunities. Verify shared publishing platforms at the page level so one owned profile does not make an entire third-party domain look owned.

Move a source into Prospector

For a non-owned source, use the bookmark action to add the URL to an existing or new Prospector list. Rankability checks canonical list membership so a URL already saved in Prospector is shown as saved instead of being silently duplicated.

Use Prospector to qualify the source, assign outreach work, and keep the citation report focused on evidence rather than relationship management.

Understand history and alerts

Citation history and volatility compare persisted completed runs. A source can be added, removed, or change position as platforms alter their retrieval results. Compare the same keyword, platform, location, and time window before labeling a change as meaningful.

Open Report source history to select an exact completed report and see its URLs, platform, position, gained/lost state, analysis result, evidence, and original AI answer. Reports with unchecked pages show Not checked yet; their saved source URLs and answers remain available. Download CSV or JSON to work with the full timeline outside Rankability. A Lost row uses its last-seen report for the page evidence and identifies that source report in the export.

Report source history is available in the signed-in workspace report. The client portal does not currently include this history browser.

Recent citation alerts are signals to investigate. Dismissing an alert changes its notification state; it does not delete the underlying citation history.

Troubleshooting

If the table is empty, confirm an AI scan completed and the selected platforms returned citations.
If the tab is missing, check whether the project’s enabled AI platforms are configured for mentions only.
If a source says Not available, inspect the URL manually; Rankability may be blocked even when the page opens in your browser.
If an owned source is shown as external, update the client’s owned assets and rerun the free classification reconciliation.
If a citation’s own result looks stale, use Recheck mentions to recheck the current sources. To refresh the AI answer behind the citations, run the appropriate AI scan instead; citation analysis does not refresh the answer.