Marketing & GrowthFree
AI Search Visibility Report of Cited Domains
Your buyers' questions asked repeatedly in AI assistants, every cited source recorded, and each cited site sorted by whether you can get onto it.
This review is a measurement with a denominator. Your category's questions get asked repeatedly in each AI assistant River can reach, alongside answers you paste in yourself. Every run is recorded as a row: the question, the assistant, whether you were named, and every source the answer cited. Because answers vary between identical runs, the report is a share of runs rather than a screenshot. Each cited domain then gets classified by whether there is a route onto it, which is what turns a visibility figure into a work list.
Most of what ranks for this search is a list of tactics, and the ones that are not are dashboards counting how often a brand name appears in answer text. Neither is usable. A brand mention you cannot trace has no next step. In this page's illustrative example, a freight audit firm we call Calderwood was named in 74 of 480 runs while its own site was cited in 21. Mentions come from the model. Citations come from pages, and pages are things you can go and get.
Built for SEOs being asked whether AI answers are taking their traffic, product marketers who want to know which comparison page the machine is reading, and founders who saw a competitor named and want to know why. Feed the query set from the Search Console keyword map, and use the messaging from customer language pack when the answers describe your category in words you never chose. It runs in the marketing workspace.
Check whether you were eligible before you measure anything
Two robots.txt lines decide most of this, and they are not the same line. OpenAI's crawler documentation states that its search and training agents are independent, and that a site opted out of the search crawler will not be shown in ChatGPT search answers. Plenty of sites blocked the training agent in 2023 and blocked the search one by accident. The illustrative Calderwood account had exactly that: zero citations in 120 runs on one assistant, and 21 in 360 on the other three, from one line nobody could account for.
The second gate is a snippet control. Google documents that nosnippet stops a page appearing as a snippet in every surface, and also prevents the content being used as a direct input for AI Overviews and AI Mode. Calderwood's pricing page carried one, added in 2022 so competitors could not read the price from the results page. It is the page an answer needs in order to say what freight audit software costs, and it had been withheld for four years.
What no amount of markup fixes is the rest. Google states plainly that you do not need to create new machine readable files, AI text files or markup to appear in these features. The same page notes that traffic from these features is counted inside the Web search type in the Performance report, so there is no AI-only report to read. That absence is the reason this has to be sampled rather than queried, and why every figure needs a run count beside it.
How it works
Name the questions
What a buyer asks before they know your name, in the words they would actually use.
River checks eligibility
Crawler access and snippet controls first, because a blocked site cannot be cited at all.
Read the source table
Every cited domain with its share of runs and the route onto it, or none.
Rerun the same set
Same questions, same run count, same assistants, next quarter, so the comparison still means something.
What you get
- Every run recorded as a row, so each figure carries the number of runs behind it
- The eligibility check first: which assistants can reach your site, and which cannot
- Whether you were named in the answer, kept separate from whether you were cited
- Every cited domain ranked by how many runs cited it, across the whole query set
- Each cited domain classified by the route onto it, or marked as having none
- The questions where you never appear once, which is a content finding not a link one
Common questions
How many runs per question do you actually need?
Enough that the share means something, which in practice starts around five and gets more useful at ten. The report states the run count next to every figure and marks any cell too thin to support a claim. A one-in-five result at five runs is noise, and saying so is more honest than reporting twenty percent.
Which AI assistants does River check?
The ones it can reach on the open web without your login, asked through their public pages, plus any answers you paste in from assistants you use yourself. The report lists every assistant it covered, any it could not reach and why, and keeps your pasted answers in their own rows so nobody mistakes them for River's runs.
Is there a file or markup that gets me cited?
No, and Google says so directly: you do not need new machine readable files, AI text files or markup to appear in these features. What exists is the reverse, controls that stop you being used. Those are worth auditing precisely because they are binary, free to remove, and nothing in your analytics reports them.
Why report cited sources rather than whether we are mentioned?
Because a mention has no next step and a citation has a URL. In the worked example the firm was named in 74 runs and cited in 21, and the 74 came from the model rather than from a page. Every cited domain is a page somebody publishes, which means somebody can be asked, pitched, reviewed or paid.
Should we block AI crawlers or allow them?
That is a business decision and the review does not make it for you. What it does is separate the two so the decision is informed: the search agent and the training agent are independent rules, and blocking the training one costs you nothing in answers. Sites regularly block both while intending only one.
What do I do about a competitor's own page being cited?
Nothing, and naming it is the point. In the worked example, roughly a sixth of citation events sit on competitor pages and encyclopaedia entries where there is no legitimate route in. Reading what that page claims is a competitor teardown instead. Marking the dead ends explicitly concentrates effort on the 83 percent that has a route.
How does this compare with tracking rankings?
It is closer to a survey than to a rank check. Positions are stable enough to track daily; answers are not, so this runs as a periodic sample against a fixed question set. Keep the questions and the run count identical between rounds, or the comparison measures your method rather than your visibility.
Sources
Built by
Chandler Supple
Co-founder & CEO, River
Chandler has spent the last six years working in AI, building tools for millions of users and Fortune 100 companies. In 2025, he co-founded River to build a new kind of operating system for AI agents.
AI Search Visibility Report of Cited Domains
Describe what you need, add any files you have, and your workspace opens with the work already underway.