Citations7 min readFielded

AI citation source analysis: sorting cited links by type and comparing assistants

In AI citation source analysis, every cited link in an AI answer is a clue about what the assistant trusts for your category. This field report shows how to collect those links, sort them into a handful of source types, compare the mix between assistants, and read what is missing, which in adult categories is often as telling as what is there.

Illustration of a bar chart with error whiskers on ivory, labeled Report 05

From the coded answers, pull each link the assistant shows, with the question, the assistant and the run it came from. Note which assistants showed no links at all, since their sources cannot be analyzed and should not be guessed.

Source typeExamplesWhy it matters
Own siteYour pagesDirect evidence you control
Retailers and marketplacesProduct listings elsewhereOften carry specifications and prices
Review and comparisonIndependent reviews, roundupsCan be scarce for adult products
CommunityForums, Q&A threadsFrequently fills the gap in adult topics
VideoVideo platformsDemonstrations and explainers
Reference and newsEncyclopedias, health and news sitesUsed for general and safety questions

Q3Compare the mix by assistant

  • Count each type as a share of all citations, separately for every assistant.
  • Look for assistants that lean on one type, since that shapes where effort pays off.
  • Check the same mix across months; sudden shifts often follow product updates.

Q4Read what is missing

In adult categories, mainstream review and news sources are often absent because many publishers avoid the subject. Answers then rely on fewer sources, frequently community threads and retailer listings. A thin source mix means a single accurate, well-organized page can matter more than it would elsewhere.

Q5Act only on honest openings

Useful responses include making your own pages clearer, correcting errors in public threads openly, and supplying accurate facts to legitimate reviewers. Paying for disguised placements is not one of them. Rival-level analysis is in our report on AI competitor analysis, and the service on competitive AI visibility.

Q6A source taxonomy

TypeExamples in adult categories (illustrative)
Own siteYour product, guide and policy pages
RetailerMarketplaces and specialist shops listing your products
Review and comparisonSpecialist reviewers and comparison articles
Forum and communityDiscussion threads and community sites
Health and educationSexual-health educators and clinics
Reference and newsEncyclopedias and news coverage, when present

Q7Owned, earned and absent

Sort each citation into owned (your own site), earned (independent sources that mention you) and absent (sources that cover your topic but not your brand). Owned citations show whether assistants trust your pages; earned ones show your reputation; absent ones show where buyers learn about the category without hearing about you. The absent list is usually where the most useful opportunities sit.

Q8Tracing errors to sources

When an assistant says something wrong about your brand, look at the sources it cited for that answer, then search those sources for the error. Often it comes from an outdated review, an old price on a retailer page or a confused forum post. Correcting or updating that source, where possible, and publishing the correct information clearly on your own site, gives assistants a better source to use next time.

Q9Outdated and broken sources

Assistants sometimes cite pages that have changed or disappeared. Check cited links each round. A cited page that no longer exists may still shape answers through older indexes; a redirect or an updated page on your own site can help. Dated, maintained pages are less likely to be replaced by stale ones.

Q10Correlation, not proof

Being cited often does not prove that a source causes your visibility, and changing a source does not ensure a change in answers. Treat source analysis as a map of likely influences, test changes one at a time, and judge them over several rounds. Monitoring between rounds is described in our guide to prompt set monitoring.

Q11Citation mix by assistant, illustrated

Assistant (illustrative)Largest source types cited
Assistant 1Review sites and retailers
Assistant 2Forums and community sites
Assistant 3Own sites and reference sources
Assistant 4Few links shown; answers rely on model knowledge

Made-up pattern. Real mixes differ by category and change over time, which is why we measure them in each round.

Q12Actions by source type

Source typeHonest action
Own siteClearer, dated pages answering the questions asked
RetailerAccurate product data and descriptions on listings
Review siteRequest a fair review; supply accurate information
ForumAnswer questions openly as the brand, where rules allow
Health and educationOffer accurate information; never pay for endorsement

Q13Sources you cannot reach

Some influential sources, such as large reference sites or major news outlets, will rarely cover an adult brand. That is fine. Focus on the sources you can reach honestly: your own site, retailers, specialist reviewers and communities. Over time, steady coverage in reachable sources does more than chasing a mention in one that will not cover the category.

Q14Recording sources over time

Keep a list of every cited source each round, with how often it appears and for which brands. Over several rounds, the list shows which sources gain or lose influence, and whether your actions changed anything. Rival source patterns are compared in AI competitor analysis.

Q15When your own site is cited

Citations of your own site are the clearest sign that assistants trust your pages. Note which pages are cited and for which questions. Those pages deserve priority for updates, accuracy checks and internal links, because they are already doing work for you. If a page is cited with outdated facts, fixing it quickly matters more than writing something new.

Q16In short

Collect every cited link, sort it by type, compare the mix across assistants and against rivals, trace errors to their source, and act only through honest openings. Repeat each round and watch which sources gain or lose influence.

Search engines that write AI answers often show sources from their own index, which means classic search visibility feeds those citations. Pages that rank well for a question are more likely to be cited in the AI answer for it. Source analysis and search work therefore support each other: improving a page that already ranks can raise both its search traffic and its chance of being cited.

Q18A source review each quarter

Each quarter, list the twenty sources cited most often in your category, mark which mention you, which mention rivals and which mention neither, and note any change since last quarter. Pick two honest actions for the coming quarter. Over a year, the review shows whether your presence in the sources that matter is growing.

Q19Next reads

How rivals use sources is covered in AI competitor analysis, and measurement in measuring AI visibility.

Q20Sources and seasons

Cited sources shift with the calendar. Gift guides and seasonal roundups gain influence before Valentine's Day and the end-of-year holidays, then fade. Check which seasonal pages assistants cite each year, and make sure your information is accurate on them well before the season starts.

FAQQuestions

Do assistants that show no links still use sources?

Often, but you cannot see them. Analysis is only possible where the assistant shows citations, so we note which assistants did.

Why are some source types rare in adult categories?

Many mainstream publishers and review sites avoid adult products, so answers fall back on fewer, often community-driven, sources.

Is a cited forum thread good or bad for us?

Neither by itself. What matters is whether it describes you accurately; inaccurate threads are worth answering openly and honestly.

LLM visibility audit

Find out how often AI assistants name your brand

Tell us your brand, your market and your main competitors. We run a sample of real customer questions through the major AI assistants several times, record mentions and citations, and send back a readout showing the ranges observed. The first one costs nothing and binds you to nothing.

Request an LLM visibility audit