CitadexDocs

Citation Sources: find the sources AI trusts and get cited there

How Citation analysis (with each page’s citation status) and Market health & repair work, how to read the gap score and the citation gap map, and the step-by-step routes from a gap to a citation.

Product Manuals21 min read

What this page is for

AI answers draw on a fixed set of websites. This page shows which domains and pages AI cites, whether you and your competitors are in them, how long citations last, and which sources to prioritize. It also checks, per market, whether your brand basics are in place.

When to use it

  • Coverage isn’t improving and you need to see which sources AI cites.
  • Deciding where to publish and where to collect reviews this month.
  • Checking that a market’s brand pages and third-party listings are complete.

The page

A bar runs across the top and stays there as you scroll; in Chat mode it shows “project / Citation Sources” on the left. Below it is one page, Citation analysis, with one filter row that starts with Market and Prompts; a set filter reads “Market: Japan ✕”, and ✕ clears it. (Citation retention was merged into it as each page’s Citation status; its old links open the analysis filtered to Decaying. Source Map was retired; its old links open Market health & repair. The opportunities tab was retired; old links open Prompt Tracking → Tracked filtered to Uncovered); the Market health & repair button on the right of the bar (showing c/n covered) opens the market health window.

Citation analysis

  • The filter row: after Market and Prompts come Engine (All engines, Google (all 3) (AI Overviews, AI Mode and Gemini together; shown when at least two of them are scanned) or one engine) and Site type (Owned / Competitor / Review site / Social / Authority / Media / Corporate, or Sites that block an AI; it also narrows the most-cited pages and the gap cards), then three buttons, All, Rival-only (AI answers citing the site named a rival and never you) and Neutral sources (encyclopedias, government and academic sites), which narrow the same three blocks; on the right, the domain search, the period (Last 30 / 90 / 180 days) and 文A one-click translation. Google's surfaces draw on Google's own index and cite different sources than ChatGPT or Perplexity, so their gaps are worth working on separately.
  • The KPIs sit in one strip, each with a small trend line: Own-site cited rate, Your citations, Cited domains and Uncovered top sources.
  • Cited source domains: sorted by citations, with Type, Citations, Retrieved (the share of your prompts whose answer listed this domain), You (how often you appear when it is cited), Gap and AI can read. A kind chip marks Community, Creator, Government, Academic and Licensed (a publisher with an announced AI content deal). Click a row to see Pages AI cited and Prompts that cited it; each page also shows Co-mentioned rivals (rivals named in answers citing it) and an N-week trend of citations, and each prompt has View the AI answer.
  • AI can read comes from each site’s robots.txt: whether the ChatGPT, Gemini, Claude and Perplexity crawlers may read it (the cell reads All allowed, Blocks … with the engines it blocks, or could not check; hover it for each engine). Content you place on a site that blocks an engine usually doesn’t show up in that engine’s answers. Filter with Sites that block an AI.
  • Most-cited pages and Citations by source type (Owned / Competitor / Review site / Social / UGC / Authority / Media / Corporate).
  • Citation gaps · priority sources: sources AI cites often where competitors appear but you don’t; the higher the gap score, the higher the priority. Competitors’ own sites are never suggested, and vendor pages rank after independent sources. Each source has actions by type: community sites get Draft post, Find live threads and Draft reply; video sites get Video script; others get View solution (a content brief); media sites also get Outreach.
Citation Sources · Cited source domains
Source domainCitationsYouGap
g2.com4212%88
reddit.com315%79
acme.com17You—
👆 Start with the highest gap score

Citation status

The most-cited pages and the pages in each site’s drill-down carry a citation status (last 12 weeks, this market, all engines): Stable, New (recently started being cited), Decaying · half-life N wk (still cited, but the citation rate is falling; the drill-down also shows when it is expected to lose citations), Lost (cited before, but not in the most recent weeks that had scans) or Returned. Weeks without scans never count as a drop; with too little data no status is shown. The Citation status filter shows one status at a time (pages, the domain table and the gap cards together). These are modeled; hover for how. Your own pages that are decaying or lost go to Weekly topics → Content refresh with a diagnosis and next step; when a rival’s page AI cites most is decaying, the New articles evidence says so — consider publishing on that topic. Each AI engine’s citation half-life, weekly churn and re-entry are an industry benchmark across customers (shown with at least 5 customers, aggregates only): ask the AI assistant “how long does ChatGPT keep citing a page?”, or use the MCP tool get_industry_benchmark (citation_lifecycle). The old Citation retention tab was merged here; its links open Citation analysis filtered to Decaying.

Market health & repair

Click Market health & repair on the right of the Citation Sources bar to open it as a window (Before your project starts in the AI assistant and links in the chat open it too). The markets are chips at the top; it opens on the project's default market (a project without markets has only All markets, the global / English list). Under them: Industry: … Change (the list follows your industry, detected automatically; Change picks another), Based on: last 90 days · P prompts · N AI answers · M engines. With few prompts the window is in starter mode (one prompt scanned many times is still one question), with Add prompts next to the note; once you have enough prompts the list follows your own AI answers. A site counts as one AI reads here only when answers to several different prompts cite it. The top of the window shows Start-up must-dos c/n and Re-detect. The list is the built-in sites for this market and industry plus every site AI actually cites in this market's answers, so a site missing from the built-in list still shows (tagged Not on our list · AI reads it); a built-in site AI hasn't cited yet is tagged Common in your industry · not cited by AI yet; Often cited for your industry · N% of answers means N% of other Citadex customers' AI answers in the same industry and country cite it (an aggregate of at least 5 customers; never who, and never another industry). The built-in list grows with the data: sites AI cites for several customers of one industry but missing from the list are checked by Citadex and added for that industry and market. With the tracking script or a store connected, a row also shows 90 days: N visits · M orders · amount (visits to your site referred by that site, and orders whose first visit came from it; graded Likely); a site already sending you visitors or orders moves to the top. N% of peers saw results after listing means that of other customers in the same industry and country who marked it registered, N% were named with it more often or got orders from it some weeks later (an aggregate of at least 5 customers; before/after, not proof of cause). When you click I'm listed Citadex saves how the site looked then and compares later, counting the visits and orders it sent since. The window is your project's start-up checklist, in two groups, both open when you arrive: Get listed / claim (maps, reviews, booking, marketplaces, directories) and Encyclopedias & company databases (Wikipedia, Wikidata, company databases; you can't write your own Wikipedia article; propose on the talk page), with the folded Left out at the bottom (image servers such as images.openai.com, the engines' own domains, trackers, short links, and sites you marked not relevant, with Restore).

Which sites are Must-do is decided by the AI assistant’s analysis of the market: Citadex lists every site of the market with what it read (how often AI cites it, for which questions, the competitors named there, the visits and orders it sent), and the AI assistant decides which ones you must be on, giving a reason for the ones it leaves out. A site that is common in your industry but that AI rarely cites in this market is not a must-do. The Based on line at the top shows the date of the analysis and the number of must-dos; View analysis opens it in full, and while an analysis runs the line shows it. With too few prompts (under 10) there is nothing of your own for the AI assistant to analyze, so the must-dos follow your industry’s must-haves, what peers’ AI cites and sites sending orders, with a nudge to add prompts; with enough prompts but no analysis yet, the must-dos are provisionally based on AI citations and Analyze now starts one (it uses credits); projects with assistant follow-ups on are analyzed automatically, then once a week. The rest of a group sits under one folded row, N more (optional — after the must-dos), and doesn't count toward progress. Finishing a must-do never adds a new one; must-dos change only when the AI assistant analyzes the market again. Each group header shows Must-do c/n (or Optional n when it has no must-dos); in Start-up must-dos c/n at the top, You're on it and Registered · waiting for re-check both count as done.

The last line of Encyclopedias & company databases is Official profiles your site’s schema links (sameAs) x/y, with a chip for each profile (LinkedIn, X, YouTube, Crunchbase, GitHub, G2 and so on); Get the code opens paste-ready Organization / FAQ JSON-LD inside the window, with the profiles found filled in.

Do these 3 first, Official accounts (communities & video), Get covered / included, Competitors' and other companies' sites and Own-site base pages are no longer in this window. Forums, Q&A, blogs and video are ongoing content work, and an account alone doesn't make AI cite you, so they aren't a start-up item: in Citation analysis open a site; threads have Find live threads, Draft reply and Draft post, YouTube has Video script, and article platforms such as note.com or Medium have View solution. Outreach to media and roundup articles: in Citation analysis open a site and click Outreach, or use Outreach in Weekly topics → Content refresh. Competitors' and other companies' sites: View solution on that page in Citation analysis, or a New article in Weekly topics. Your own definition, category, comparison and FAQ pages: New article in Weekly topics and Generate strategy article in Prompt Tracking, which work from your low-coverage prompts. Structured data: the sameAs line above and the agent readiness check (AI Traffic → Agents).

Each row shows its type, AI cited it N× (AI reads it, which is not proof you're there), competitors named there and its state: You're on it, Registered · waiting for re-check or Not confirmed yet. A site is marked done only after it is verified. When AI cites a page on a site that is your own (its address or title names your brand, or it sits under the profile URL you gave at I'm listed), that site turns You're on it by itself, sites outside the built-in list too; built-in sites are also checked on the web and against Wikidata, and on some platforms (such as ITreview) Citadex opens your product page there directly, so a page you just made counts before AI cites it. View shows which pages. The button opens the site's own vendor sign-up or claim page (Open site / Claim / Register / Apply to list ↗, marked Sign-up link checked); where we haven't checked that link it opens the homepage and says Homepage only · sign-up link not checked. Many sites create a page for a product on their own: before registering, Is there a page of mine? searches that site on the web, so you can claim the page instead of making a second one. Come back and click I'm listed: a small box asks for Your profile URL on {site} (optional), with Skip and Save. With your profile URL there (for example your note.com account page), Citadex opens that page right away: when it loads and names your brand, the site turns You're on it at once; a site that turns the check away (G2, for example) shows Registered · waiting for re-check and is tried again daily. Any page under that address that AI cites also counts as you being there, even without your brand in the title. An address on another site, or the site's homepage, is refused (red hint). Why this one lists the pages and questions AI cited the site for, whether you were named, and See the answer. A site outside the built-in list can be Relabelled: competitor, a platform you can join, or not relevant; a built-in platform can be relabelled as not relevant. I’m listed counts per site: marking G2 in one market marks G2 in the others; a map or booking listing is per location and counts only in its market. Platforms for other industries are not listed, unless AI actually cites one often in this market: then it sits under the optional rows, tagged Not on our list · AI reads it, with Open site, never a must-do. View evidence, Why this one and Get the code open inside the window; ← Back returns to the list. The check can confirm that you are on a platform, not that you are absent, so an item it can't find shows Not confirmed yet rather than a gap. Sites you mark I'm listed are re-checked automatically (are you confirmed there yet) and again later (does AI cite the site, how often you're named with it), and the AI assistant tells you the result in the chat; every Monday it also points out one must-do site AI keeps citing where competitors show up and you aren't confirmed, without repeating the same site often. You can also ask the assistant "Which sites should we register on first?".

On a review site you are listed on, the row also shows N reviews · rating · newest date, as your own page there publishes them (Citadex reads it weekly; nothing is shown or estimated when the page does not publish them), and Ask customers for reviews. It lists the site’s rules for asking first: send it to every customer, not only the happy ones; don’t ask for a score or suggest wording; offer no gift, discount or prize draw (a site’s own review program aside); staff and anyone you pay don’t review. Then Write the invite writes an email for you to send from your own email or customer tool; Citadex sends nothing. Sites that ask businesses not to request reviews (Yelp, Tabelog) get no invite.

Common tasks

Core workflow: from a gap source to a citation

The full path is: find a gap source → open the solution and generate a brief → send it to the writer → publish → re-test. Different source types have different buttons; each route below follows the click order.

Route 1: media / review / corporate sources → View solution → brief → article

  1. 1

    Pick sources in the citation gap map

    Take two or three with the highest gap score. The Citations / Retrieved / You numbers on the card show that AI cites it often while you are barely in it. You can also expand a row in Cited source domains and work on one specific page.

  2. 2

    Click View solution to open the Writing Brief

    The top shows which engines cite this source and that you are not listed in it. "This ONE article covers all n prompts" lists the related prompts (up to 8; expand to see more). The list is read-only here; you add or remove prompts in step 5. At first the brief is a generic template ("Generic template · click 'Generate brief' for a tailored one").

  3. 3

    Click Generate brief

    AI builds a tailored brief from those prompts, your business info and the target market: Suggested title, Article structure (H2 / points), Competitors to mention, Evidence to include, Where to publish and Publish sequence, plus the Re-test after publishing note. Not happy? Regenerate. Copy brief copies the whole brief for your content team or another writing tool. A generated brief is kept, so reopening the same source shows it immediately.

  4. 4

    Click Send to writer

    Available only after the brief is generated. You land on Answer Assets with everything filled in: the brief goes into Outline as context, the related prompts go into the prompts box one per line, competitor names come along, Content type is chosen from the source type (comparison → Comparison, best-of list → Listicle, FAQ → FAQ-heavy article, guide → How-to guide, otherwise Standard article), and AI agent · deep research is switched on.

  5. 5

    Add or remove prompts on Answer Assets

    The prompts box takes one prompt per line: delete a line or type a new one, or use Pick from tracked prompts (N), which adds or removes a line as you tick, showing how many articles each prompt already has and its coverage. With many prompts, filter by group or search at the top of the picker, and Add all N shown adds a whole group at once. Comparison and list articles work best with 3 to 15 prompts. You can also change Content type and Article language here.

  6. 6

    Click ✨ Generate Content

    Generation takes about 3 to 5 minutes. The panel shows the generation progress, and you can close it to keep it running in the background. When it is done you are taken to Content, where the article carries the Counter tag; back on Citation Sources, the source shows how many times it has been generated.

  7. 7

    Review, publish, re-test

    Open the article in Content, check it and click Publish Now. Seven days after publishing, the same prompts are re-scanned automatically. Then check those prompts’ coverage in Prompt Tracking, or see whether the article shows up as Article worked under Achievements.

Route 2: community and forum sources (Reddit, Quora and similar)

  1. 1

    Click Find live threads

    A thread finder opens in a new tab, already loaded with the first prompt, your main competitors and the source link, and starts searching for live discussions on the topic.

  2. 2

    Click Draft reply or Draft post

    Draft reply writes a natural reply using the source title and link. Draft post writes a new post with all the source’s prompts (up to 12), competitor names and your brand name, in the language of the source page.

  3. 3

    Copy it and post on the platform

    Community content doesn’t go through Content. After you post it, watch the related prompts in Prompt Tracking.

Video sources (YouTube, Vimeo) only have Video script, which opens the script tool with the first prompt.

Route 3: media sources → Outreach

Sources of type Media, and company roundup or comparison articles (titles like "best tools" or "top 10"), show both View solution and Outreach. Paid placements must be labelled as ads.

  1. 1

    Click Outreach

    An outreach email is drafted automatically while the system looks for the article’s author and contact details ("Finding the article author & contact…"). Once an author name is found, the email is redrafted to address them by name.

  2. 2

    Choose Angle, Tone and Language

    Angle: Suggest inclusion (politely suggest adding your brand to the article), Offer data (trade your exclusive data for a citation) or Expert take (offer a quotable expert comment). Tone: Friendly, Concise or Formal. Language follows the project by default. Changing any of them redrafts the email.

  3. 3

    Choose a contact

    Real contacts (read from the article) lists the Article author and Editorial desk addresses. If there is only a profile and no email, click Open author profile. If nothing was found, use Find author on LinkedIn, Search author’s email on Google or Open the article.

  4. 4

    Edit Subject and Body, then send

    Both are editable and each has Copy. Copy email copies the whole email; Open in mail client opens your mail app and moves Stage from To contact to Contacted. Later, update Stage by hand: Replied, Placed or Declined.

  5. 5

    Attach material

    Suggested attachments lists a feature comparison against competitors, citation-share data and product screenshots you can quote.

Route 4: Prompt Tracking → Uncovered → Generate strategy article

  1. 1

    Open Prompt Tracking, set Coverage to Uncovered and select prompts

    These are the prompts where AI rarely names you; tick the ones you want written up.

  2. 2

    Click Generate strategy article in the bulk bar

    The dialog below opens directly with the ticked prompts.

  3. 3

    Adjust the prompts in the strategy article dialog

    "This article will cover {n} prompts" is a checklist: untick a prompt to leave it out. Type in Add a related prompt to cover… and click Add to add a new one; it is also added to your tracked prompts, and the system checks whether it fits the topic (a prompt that doesn’t is marked ⚠). Not sure which to pick? Switch to AI auto-pick chooses by topic cluster.

  4. 4

    Choose the angle and how to generate

    Angles: Comparison / Roundup, How-to, In-depth guide and Explainer. With the AI agent turned on, it writes one article from 3 to 15 prompts; turned off, it can write one article per angle. Comparison / Roundup always uses the AI agent with web research.

  5. 5

    Click Generate article

    If the prompts span too many topics, a warning offers Let me adjust, Auto-remove outliers or Generate anyway. The finished article goes to Content; click Go to Content management / publish →.

Fill in a market’s basics

  1. 1

    Market health & repair → choose a market

    The top shows Start-up must-dos c/n and each group Must-do c/n. The must-dos are enough; the N more rows are optional.

  2. 2

    Must-do sites: Claim / Register

    The site's claim or sign-up page opens in a new tab. When you're done, come back and click I'm listed, add your profile URL there if you like, and the status becomes Registered · waiting for re-check.

  3. 3

    Site schema: Get the code

    On the sameAs line at the bottom of Encyclopedias & company databases, click Get the code and add the Organization / FAQ JSON-LD to your site.

  4. 4

    Click Re-detect to confirm

    The system checks the web; confirmed sites change to You're on it.

Notes

  • The prompt list in the brief dialog is read-only; add or remove prompts on Answer Assets after Send to writer. In Route 4 you tick and add prompts right in the dialog.
  • Send to writer stays grey until you click Generate brief.
  • Only Route 1 articles carry the Counter tag and count toward the source’s generated total.
  • Outreach only appears for media articles and roundup / comparison articles. Review sites have no outreach button; claim them under Market health & repair with Claim or Register.
  • Competitors’ own websites never appear as gaps or outreach targets.

Common questions

What do Retrieved and You mean? Retrieved is the share of your prompts whose answer listed a URL from the domain. You is how often your brand appears when that domain is cited. Your own domain shows You; a competitor’s shows n/a.

Where did Source Map go? It was retired, because its scores mixed in estimates. Old links open Market health & repair; the official-profile (sameAs) check is the last line of Encyclopedias & company databases there, with Get the code.

Does Not verified mean I haven’t done it? Not necessarily. The check uses a web search; if it can’t find you it shows Not verified. If you know you’re listed, click I'm already on it → and it is re-checked next time.

A source blocks an engine in AI can read. Content you place there is unlikely to help with that engine; favour sources that engine can read.

Troubleshooting

  • No citation data in this window yet.: widen the time window, or run checks in Prompt Tracking first.
  • A platform opened by Go claim won’t load: it is a third-party site; try another network or try again later.

Was this article helpful?