Enter your domain. CiteGraph asks the questions your buyers ask, shows every answer the engines give, and hands you the fixes to get you in.
Not keywords. The real questions people type into AI assistants, mined from your own site and category.
Every answer, every cited page, sampled repeatedly across engines. The receipts are kept.
Every answer you are missing from, who was named instead, and the claim that would have put you in. CiteGraph writes the page, and Autopilot can publish it.
Exhibits 01 to 07 are real screens of our own project, citegraph.app, taken on 29 September 2026 from its weekly scan of 25 September, which ran three runs a question. The page in Exhibits 02 and 05 is shown as CiteGraph wrote it; the version live on our site since 15 September was edited by hand. Exhibit 07's Ask reply and MCP call were recorded on 26 September. The numbers are real, including the questions we lose.
Every buyer question the assistants answered without you, with each answer word for word, the rivals it named and what it picked them by. No score can tell you that. The answer can. Each lost answer ends in the one move to make next.
Every lost answer ends in one move: a page on your site, or a pitch to the page the answers cite. The Action plan puts the moves in order and says who does what. CiteGraph writes the page from the answers you lost and your own site, every sentence traced to a source, and leaves a slot for the detail only you can add.
The sites the assistants cite on the questions you lose, how often each one is cited, and whether it is a tool's own site or an independent one you can get onto. Each site's row shows the lost questions it is cited on; open one to see the answer that cited it. CiteGraph drafts the pitch; you send it.
One square per answer for the three rivals named in the most questions and for you, question by question, so you see who the assistants name and where your slot stays empty. Each row ends in its fix: write the page, or pitch the page the answers cite. Names the assistants use that you don't track are listed too; add one and the next scan measures it.
A page is followed from the answer you lost to live, to the first AI crawler that reads it, to the scan that names you. Weekly scans re-ask its question with the margin printed, and only a move beyond the margin is flagged. Ours went live on 15 September. No assistant names us for it yet, and the screen says so.
Not keywords. Ten questions written the way your buyers type them (compare, cheapest, alternative to, use case), each asked on all four assistants three times in every weekly scan, and five times in a scan you start on Growth or Scale. Open one to see, per assistant, whether it named you or cited your site, and who it named instead. Edit them any time; the next scan follows.
Ask reads the same answers, sources and plan you see, says when a number is inside the margin, and remembers what you tell it. It never posts anywhere public. The same tools run as an MCP server, so Claude, Cursor or VS Code can read your lost answers and apply the site fixes in your repo.
Enter your domain. CiteGraph asks the questions your buyers ask, shows every answer the engines give, and hands you the fixes that would put you in.
7-day trial · $1 to start
Or see pricing ↓The verdict first: which rivals the assistants name to your buyers, and how many answers name you, with the margin printed. Then the lost answer to start with, word for word, and the sentence that would put you in it.
Swipe the screen to see more →
citegraph.app · scan of 25 Sep 2026 · screens taken 29 Sep 2026
llms.txt does nothing. We checked. Competitors sell generators for it anyway.
Your mention rate is a vanity metric: the cited-source map is what you can act on.
One run is noise. Every number here is sampled, with its error bar printed.
Rank can’t be bought. Our own product is not on our own leaderboard. We published it anyway.
Answer engine optimization is the practice of getting your product named, and your pages cited, inside the answers AI assistants give (ChatGPT, Perplexity, Gemini and the rest) rather than inside a list of blue links. It differs from SEO in what it optimizes for: an answer engine reads a handful of pages and writes one recommendation, so the goal is to be the source that recommendation is built from.
Yes, in two separate ways, and they answer different questions. Your own server can report every request an assistant makes to your pages, so you see ChatGPT opening a page to build an answer, verified against the IP ranges OpenAI publishes. That tells you what was read. Separately, asking the engines your buyers' questions on a schedule and recording the sources in each answer tells you what was cited. Being read is not being named, but nothing gets named without being read. CiteGraph does both: a tracker you install on your site, and repeated scans across four engines.
It retrieves a handful of pages at answer time and writes a recommendation from what they say. So the pages that get mentioned are the ones that state a usable claim plainly: a price, a limit, who it is for. Most of the pages an assistant cites when recommending a product are not the brand's own, which is why listicles, forums and review sites often carry more weight than a homepage. Getting mentioned is less about your own site and more about being on the pages the engine already trusts for that question.
Find the buyer questions where an assistant already answers without you, read the pages it cited to build that answer, and get the missing claim about your product onto one of those pages, or onto a page of your own that answers the question better. CiteGraph measures which questions you lose, shows the verbatim answer and the cited sources, and writes the specific sentence that would have qualified you.
No. We tested it and found no measurable effect on citation rates. Assistants cite pages they retrieve and read at answer time; a manifest file at your root is not part of that path. Several tools sell llms.txt generators anyway. We say this publicly because it costs us nothing to be honest about it.
No, and the difference matters. A mention means the assistant named your product in its prose. A citation means your own page was used as a source. In our research (the week of 24 August 2026), for 34 of 70 top-ten products across 7 software categories, the page AI cited when recommending them belonged to somebody else, so a product can be recommended constantly while its own words never reach the buyer. Mention rate is the vanity metric; the cited-source map is the actionable one.
We write ten buyer questions for your category, ask each one on four AI engines through their official APIs (three times each in a weekly scan, five in a scan you start on Growth or Scale), and record which products are named and which URLs are cited. Rates are sampled, so every number is published with its margin of error, and we only flag a change when it moves beyond that margin. Nothing is scraped from a chat UI and nothing is simulated.
Signing up shows you a site readiness score and one live AI answer on your own domain for nothing. Your first full scan runs the moment you pick a plan: starting the 7-day trial charges $1 to confirm your card is real, and the plan itself is not charged until the trial ends. Plans start at $29 a month for weekly tracking of two domains; Growth is $69 and Scale $149, and every plan carries a 120-day money-back guarantee. There is no ongoing free tier because each scan spends real API money on live engine calls.
Then you get your money back. Every paid plan carries a 120-day guarantee: if no CiteGraph scan shows ChatGPT, Claude, Gemini or Perplexity naming your product within 120 days of your first payment, we refund every subscription payment from those days. Your side is to keep the plan and weekly scans on and publish three of the pages CiteGraph writes for the questions you lose. It is checked from your own scan history; the full terms are at citegraph.app/terms.
No. Positions on the public category leaderboards are measured from engine answers and cannot be bought, boosted or edited by anyone, including us. There are no sponsored rows and no featured slots. CiteGraph itself does not appear in its own AI search visibility leaderboard, and we published that anyway.
Yes. Autopilot finds the questions AI assistants answer without naming you, checks each one against a live engine, writes the post from the facts on your own site and sources quoted word for word, sets it in your site's own design, and commits it to your GitHub repository as a finished page. You choose whether each post waits for you or goes out on its own, and every change can be reverted from the app. It works today for sites whose code is on GitHub; on other platforms you can download each finished page.
Four answer engines (ChatGPT, Claude, Gemini and Perplexity) are queried for every scan through their official grounded-search APIs, so results reflect what a real buyer would see rather than a cached index. Each question is run multiple times per engine because answers vary between runs, and the per-engine breakdown is shown alongside the blended rate.
Named by an AI assistant within 120 days, or every payment back.
Keep your plan and weekly scans on, and publish three of the pages CiteGraph writes for you. If no scan in your first 120 paid days shows ChatGPT, Claude, Gemini or Perplexity naming you, we refund every subscription payment. Terms
Full plan limits and pricing FAQ → · Something custom (enterprise terms, more seats, an API): info@citegraph.app
We ask an AI assistant a real buyer question in your category and show you whether it names you and who it names instead. After that CiteGraph writes the pages that fix it, and Autopilot can publish them straight to your site.
Check your site →CiteGraph is an AI search visibility and citation tracking tool for founders, marketers and SEO teams. It runs the questions buyers ask through ChatGPT, Claude, Gemini and Perplexity, shows every answer and the pages the engines cite, and gives the fixes to win those citations. Plans start at $29 a month with a 7-day trial and a 120-day money-back guarantee. Alternatives it is compared with: Profound, Otterly.AI, Peec AI, Scrunch AI, Rankscale, CrowdReply and Bydefault.
