Product
Vladimir Bazalenko, Founder, RendixUpdated Oct 10, 2026An AI citation is a link an AI answer gives as the source of what it says.
Rendix records every page ChatGPT, Perplexity, Gemini and Google AI Overviews cite for your buyers' questions, three times a week, and shows which pages are yours, your competitors' and everyone else's.
7-day free trial. Cancel anytime.
best tools to track website citations in perplexity and gemini
LLMrefs
ChatGPT did not show its searches on this run.
Definition
A citation shows where an answer came from. Three things Rendix's scans of its own category show about them:
A link to a page is not a mention of its brand.
46.2 % of the links to a tool's own site came in answers that did not name that tool.
9,643 links to the sites of tools the category's answers recommend, Aug 26 to Oct 7, 2026. Across all sources, Semrush counts 62 % of citations as linking a page without naming its brand.
Source: Rendix scans of its own category · Semrush, Jun 2026
A typical answer cites a handful of sources, and what an engine shows as a source differs.
7 to 9 sources in a typical answer, from Google AI Overviews to Perplexity and Gemini.
Distinct pages per answer across 2,251 answers, Aug 26 to Oct 7, 2026. Gemini's grounding reports domains only; Perplexity counts the results its answer points to.
Source: Rendix scans of its own category
Ask again and most of the sources change.
32.5 % of the pages an engine cited were cited again on the next scan of the same question.
2,067 pairs of consecutive scans, Aug 26 to Oct 7, 2026. Gemini is measured on domains.
Source: Rendix scans of its own category
Method
Every source of every answer, for every question you track, counted the same way on each scan.
Each scan keeps the links every answer gives: ChatGPT's source chips and the links in its text, the results Perplexity's answer points to, the sources under a Google AI Overview, and the domains Gemini drew on. Cited videos and Reddit threads get their own view.
Addresses are cleaned of tracking tags before counting, so one page counts once a week however many answers cite it.
Each domain is marked as yours, a competitor's, media, social, reviews, institutional, PR or other.
For every source: the questions that trigger it, its trend since first seen, the brands named in the answers that cite it, and whether its pages mention you or your rivals, checked in the page itself. A feed lists the sources won and lost.
| Engine | Sources | Sites of brands it named | Brands named, not linked |
|---|---|---|---|
| ChatGPT | 5 | 5 | 1 |
| Perplexity | 6 | 3 | 1 |
| Gemini | 10 | 2 | 9 |
| AI Overviews | 11 | 5 | 0 |
Evidence
No engine publishes its rules. These are the measured patterns, each from its own study.
In 16,851 ChatGPT queries, 88.6 % fanned out into exactly two web searches. The pages it cites come from what those searches return.
Source: AirOps, Apr 2026
A page ranked first in the engine's own search was cited 58 % of the time, a page ranked tenth 14 %.
Source: AirOps, Apr 2026
Pages with a heading close to the engine's search were cited 41 % of the time, against 29 % without one.
Source: AirOps, Apr 2026
Only 8 % of the pages ChatGPT cited ranked in Google's top 10 for the same prompt. For Google AI Overviews it was 37.9 % in January 2026.
Source: Ahrefs, Aug 2025 · Ahrefs, Mar 2026
Pages that added JSON-LD saw no major uplift in AI citations against matched pages that did not. On Google AI Overviews they dropped 4.6 %.
Source: Ahrefs, May 2026
These studies show what cited pages have in common. None of them shows that changing one thing gets a page cited.
Next views
A citation sits at the end of a chain: a search, a page the bots could read, a rival in the same answer, the page you publish next.
The searches each engine ran behind the answer, and the buyer questions they came from.
ResearchA page the crawlers cannot fetch or index is a page no engine can cite.
FoundationThe brands named in the answers that cite each source, and the questions your rivals win.
CompetitorsThe pages to write or fix next, each with a brief and a section outline.
ActionsEngines
Engines expose different parts of an answer. Rendix stores what each one shows and marks what it does not.
| Engine | Cites | Search results it read | Its own searches | Sources per answer |
|---|---|---|---|---|
| ChatGPT | Pages | When shown | When shown | 8 |
| Perplexity | Pages | Yes | Yes | 9 |
| Gemini | Domains | No | Yes | 9 |
| AI Overviews | Pages | No | No | 7 |
Plans
Straight from the pricing table: only what ships today.
Questions
/ Updated Oct 10, 2026Need more help? team@rendix.io
They search first, then cite from what the search returned. No engine publishes its selection rules.
In AirOps' study of ChatGPT, a page ranked first in the engine's own search was cited 58 % of the time, a page ranked tenth 14 %. Rendix shows the searches each engine ran and the pages it cited for every question you track.
Source: AirOps, Apr 2026
Get into the results of the searches ChatGPT runs, and answer the question near the top of the page.
ChatGPT search reads the web through OpenAI's own crawler, OAI-SearchBot, so a page it cannot fetch does not appear. Pages whose headings matched the engine's search were cited 41 % of the time, against 29 % without. Rendix's Actions turn the questions you lose into pages to write, and Foundation checks the bots can read them.
Source: OpenAI, crawler docs · AirOps, Apr 2026
Gemini's grounding reports only the domains an answer drew on.
Rendix reads Gemini through Google's Vertex AI with Google Search grounding and shows what it returns: the domains, and the searches Gemini ran. It does not guess page-level citations for Gemini.
Often. Of the pages an engine cited on one scan, 32.5 % were cited again on the next scan of the same question.
ChatGPT kept 22.3 % and Perplexity 42.7 % in Rendix's own category, Aug 26 to Oct 7, 2026. That is why Rendix counts each page once a week and keeps the history, rather than judging from one answer.
No study shows it does.
Ahrefs tracked 1,885 pages that added JSON-LD against matched pages that did not: no major uplift on ChatGPT or AI Mode, and a 4.6 % drop on Google AI Overviews. Google says generative AI search needs no special markup.
Source: Ahrefs, May 2026 · Google Search Central
No. Its search returned about fifteen pages per answer, and its answer pointed to about seven of them with [n] markers.
That is 160 answers through Perplexity's API, Sep 28 to Oct 8, 2026. Rendix counts only the pages the answer points to as Perplexity's citations and keeps the rest as the pages it read. Over Aug 26 to Oct 7, a typical answer in Rendix's own category cited 9 pages in Perplexity, 9 domains in Gemini, 8 pages in ChatGPT and 7 in Google AI Overviews.
Get started
/ Free 7-day trialOne scan shows your place in AI answers – and the first moves to climb. Free, no account needed.