Be the source in the answer, not a link under it
Answer engines do not rank ten blue links. They synthesise one answer and cite a handful of sources, and either you are one of them or you are absent, with no page two to fall back to. Getting cited is a different discipline from ranking: it depends on whether a model can reach your page, parse it, and lift a clean, attributable claim out of it.
Most of the reasons a brand is not cited are diagnosable. A blocked retrieval crawler, a page that only renders in the browser, a claim buried under three paragraphs of setup: each has a specific fix, and each can be verified against the answers themselves. This page walks through that sequence in the order that finds problems cheapest first.
Why teams run citation work with Asva
You can see which sources actually win
Citation tracking shows the exact domains each engine quotes for the prompts in your category. The list is rarely what a team would guess: it leans on community threads, review platforms and comparison posts rather than vendor sites.
Access problems surface before content problems
A large share of "we are not cited" cases are mechanical: a CDN rule blocks OAI-SearchBot or PerplexityBot, the page body is assembled by JavaScript after load, or the structured data points at a page that no longer exists. These are cheap to fix and cheap to verify, and rewriting a page a crawler cannot read is expensive work that changes nothing.
Answer-shaped content is measurable
Whether a rewrite changed how often you are quoted is a question the platform answers per engine, on a fixed prompt set, before and after. That is a cleaner test than watching referral traffic, which AI surfaces under-report.
Retrieval and training are tracked separately
An answer grounded in a live fetch and one grounded in what the model learned months ago update on different clocks. Seeing which engines cite you from retrieval and which mention you from memory tells you whether a fix will show up in days or after the next model refresh, and stops the team declaring a change a failure too early.
How getting cited works
The sequence that moves a brand from absent to quoted in AI answers, in the order that finds problems cheapest first.
- 1
Confirm the retrieval crawlers can reach you
Check robots.txt, and any managed rules your CDN injects above it, for the agents that fetch at answer time: OAI-SearchBot and ChatGPT-User for ChatGPT, PerplexityBot and Perplexity-User for Perplexity, ClaudeBot for Claude. These are distinct from the training crawlers (GPTBot, CCBot, Google-Extended, Applebot-Extended). Then fetch your key pages as a plain HTTP client with no JavaScript and confirm the body text is present in the raw HTML. A crawler that receives an empty shell cannot cite you.
- 2
Publish machine-readable structure
Add complete JSON-LD for the page types you actually have: Organization on the homepage, Article on editorial pages, FAQPage where you answer questions, HowTo where you give steps, Product where you sell. Validate it rather than trusting the plugin. Add an llms.txt that points crawlers at the pages worth quoting. Structure is how a model separates a verifiable claim from a marketing sentence.
- 3
Baseline where you stand
Write the prompts your buyers actually type: the category question, the comparison questions, the use-case questions, the "is X worth it" questions. Run them across the engines that matter and record, per engine, whether you are mentioned, whether you are cited as a source, which competitors appear instead, and which third-party domains supply the answer. Keep the prompt set fixed from here on.
- 4
Rewrite for extractability
For each page that should be cited but is not, restructure so every section opens with a direct, self-contained answer and the supporting detail follows. Use the question as the heading. Keep one claim per passage and make each claim true without the surrounding context, because that is how it will be quoted. Prose that only makes sense after three paragraphs of setup loses to a forum comment that says the same thing in two lines.
- 5
Fix the gaps the data names, then re-measure
Where third-party sources win instead of you, the work is off your domain: pull the cited-source list, note which sources omit or misdescribe you, and hand it to whoever runs outreach. Then re-run the fixed prompt set on a schedule and read the results per engine, because they will not move together. Report mention rate and citation rate separately, and log which passage was lifted when a citation appears.
What teams use this for
Category and comparison prompts
The "best X for Y" and "X vs Y" questions where a recommendation directly precedes a purchase. These are the highest-value prompts to be cited on, and the ones most often answered from third-party comparison content rather than from the vendors being compared.
Defending an existing position
You are cited today. Sources shift, competitors publish, a model update changes which domains it prefers, and the citation quietly disappears. Watching the prompt set on a schedule catches that in the week it happens rather than in a quarterly pipeline review.
Launches into a category you do not yet appear in
A new product has no citation history. Knowing which sources the engines already trust for the category tells you where launch coverage has to land, and lets you check within weeks whether it actually reached the sources the models read.
Correcting what models say about you
Stale pricing, a discontinued plan, a feature you no longer ship. Models repeat whatever the sources say until the sources change. Tracing an inaccurate answer to its cited source turns a vague complaint into a specific page to update or publisher to contact.
Documentation and support content
In technical categories, how-to and troubleshooting questions are answered from documentation, and docs are frequently gated, JavaScript-rendered or blocked. Opening them up puts you in the answer at the moment a user is trying to get something working.
What differs per engine
Each engine reaches the web through a different crawler, weights sources differently, and updates on its own clock.
ChatGPT
ChatGPT cites from a live search when browsing is triggered, fetched by OAI-SearchBot and ChatGPT-User, and from training data otherwise. Rules for those two agents govern the retrieval path; GPTBot governs training only. Non-browsing answers will not change until the model does, whatever you publish.
Perplexity
Perplexity is retrieval-first: nearly every answer fetches live sources via PerplexityBot and shows them as numbered citations. That makes it the fastest engine to reflect a fix and the clearest to measure. It leans on recent, specific pages and community sources, so freshness and a direct answer near the top of the page matter most here.
Google AI Overviews / AI Mode
AI Overviews and AI Mode draw on Google's existing index, so ordinary Googlebot access and indexing apply rather than a separate AI crawler; Google-Extended controls training use, not whether you appear here. Pages already ranking for the query are the candidate pool, and extractable structure decides which get lifted.
Gemini
Gemini grounds answers in Google Search and shows sources when grounding is used, so the same index and extractability rules apply, with shopping and product questions increasingly pulling from merchant data. Citation display varies by surface, which makes it the engine where tracking through the platform rather than by hand matters most.
The numbers to report
What each figure does and does not tell you.
- Mention rate
- The share of runs, per engine, where your brand is named anywhere in the answer. It is the broadest signal and the first to move. Rising mention rate with flat citation rate means you are being talked about from sources that are not yours, which points at off-site work.
- Citation rate
- The share of runs where one of your URLs is attached as a source. Read it per engine, because retrieval-first engines move weeks before training-grounded ones, and against the specific URL cited so the content team knows which page and passage worked.
- Cited-source share
- Of all sources cited for your prompt set, the proportion that are your domain, a competitor's, or a third party. When third parties dominate, on-site work has a ceiling and the gap is an outreach job.
- Retrieval fetch rate
- How often the retrieval agents request your key pages. Zero fetches of a page you expect to be cited is a diagnosis on its own: the engine is not reading it. Fetches without citations mean it is reading and choosing something else, which points at the content.
Common mistakes and the fix
Mistake · Blocking every AI user agent in one rule
Fix · Decide separately about training crawlers (GPTBot, CCBot, Google-Extended, Applebot-Extended) and retrieval agents (OAI-SearchBot, ChatGPT-User, PerplexityBot, Perplexity-User, ClaudeBot). One Disallow covering both removes you from answers.
Mistake · Judging a change after a week
Fix · Set the expectation per engine before shipping. Perplexity and browsing ChatGPT can reflect a re-crawl in days; training-grounded answers may take months. Report monthly on the fixed prompt set and note which engines have re-fetched the page.
Mistake · Rewriting copy before checking the raw HTML
Fix · Fetch the page as a non-JavaScript client and confirm the body and the JSON-LD are in the response. If the content only exists after hydration, the rewrite will not be read. Fix rendering first.
Mistake · Writing for the whole page instead of the passage
Fix · Models lift passages, not pages. Open every section with the answer, keep one claim per paragraph, and make each claim true without the surrounding text. A section that needs the preceding three to make sense will not be quoted.
A worked example: a mid-market analytics vendor
A hypothetical mid-market product analytics vendor ranks on the first page for its category term but is never cited when a buyer asks ChatGPT or Perplexity which tools to consider. The baseline across a twenty-prompt set shows a mention rate near zero on both engines, and cited sources dominated by a software review platform, two comparison blogs and a handful of Reddit threads.
The first check is mechanical. The CDN has a managed rule blocking every user agent containing "bot", which catches OAI-SearchBot and PerplexityBot along with the scrapers it was meant for; the crawler log confirms neither has fetched a page in months. The rule is narrowed, and the documentation site, which rendered entirely in the browser, is moved to server-side rendering. Within two weeks both agents are fetching the pricing, comparison and docs pages.
Perplexity begins citing the vendor's comparison page on the "X vs Y" prompts within the month, once each section is restructured to open with a direct verdict and FAQPage schema is added. ChatGPT's browsing answers follow more slowly and still lean on the review platform for the category question, where the vendor's profile is incomplete; that item moves to the off-site list. Three months in, the next quarter's plan is written from the gaps the data names.
Frequently asked questions
How is getting cited different from ranking on Google?+
Ranking places a link in a list the user chooses from; citation makes you a source inside a single synthesised answer. A page can rank well and never be cited, usually because a standalone claim is hard to extract from it, or because the crawler feeding the answer engine is blocked while Googlebot is not.
Does blocking AI crawlers protect my content?+
It prevents your content being used, and it also removes you from the answers those engines generate. The distinction that matters is which crawlers you block: GPTBot, CCBot and Google-Extended govern training, while OAI-SearchBot, ChatGPT-User, PerplexityBot and Perplexity-User fetch at answer time to cite sources.
How long before a content change shows up in AI answers?+
It varies by engine and by how often your site is fetched. Retrieval-grounded engines can reflect a change within days of re-crawling; answers drawn from training data can lag until the next model update. Track per engine rather than expecting one number, and do not judge a change in its first week.
Do I need to be cited by every engine?+
Rarely. Concentrate on the engines your buyers use. For most B2B categories that means ChatGPT and Perplexity first, with Google AI Overviews mattering wherever buyers still start on Google; for shopping it increasingly includes Gemini.
Does structured data actually affect citations?+
It affects whether a model can confidently attribute a claim to you, which is the precondition for citing you. Schema does not make weak content citable, but it marks which text is an answer, which entity is speaking, and what the page is about. Necessary, not sufficient, and worth validating rather than trusting a plugin.
Why does a competitor with worse content get cited instead of us?+
Usually because the sources the engine trusts for that question mention them and not you, or because their page is reachable and yours is not. Check each cause in turn: access, rendering, structure, then off-site coverage.
Should we write new pages or fix existing ones?+
Fix existing ones first. A page that already ranks, has links and is already fetched by the retrieval crawlers is closer to citation than a new page with none of that. Restructure it answer-first, complete the schema, re-measure. Add new pages only for questions you genuinely do not cover.
Is llms.txt required?+
No engine has made it a requirement, and adoption by the engines themselves is uneven. It is cheap to publish, does no harm, and gives you a place to declare which pages are canonical and worth reading. Treat it as a low-cost signal alongside robots.txt and schema, not a substitute for either.
Explore the features behind this solution
Go deeper
Find out who gets cited in your category instead of you
Baseline your mention rate across every major engine, and see the sources winning the answers you want.