AI Chat Feed
The raw answers the assistants gave to your tracked prompts. The most useful screen in RankX AI, and the first one to read when a number surprises you.
The AI Chat Feed is the answers themselves: what each assistant actually said when RankX AI asked one of your tracked prompts. Every other AI Visibility screen is a count derived from these answers, so this is the screen that explains all of them.
It is the first place to go when a number surprises you, and it is the best screen in the product on day one, when you have exactly one measurement and no trend to read.
What it shows
For each check, the feed carries the prompt asked, the assistant that answered, when it ran, the answer text, and the verdicts RankX AI derived from it: whether your brand was named, where it sat if the answer ranked things, which of your saved competitors were named, and which sources the answer cited.
Why the raw answers matter more than the rate
A mention rate tells you that you are absent. The answer tells you why, and the why is nearly always visible in one read:
Who got named instead. If the same three names appear in every answer, that is your real competitive set, and it is often not the one you listed during onboarding. Fix the competitor set and every comparison in the product gets more useful.
What the assistant thought the question was about. A prompt about "AI visibility tracking" that returns an answer about website analytics is measuring a question your buyers are not asking in the words you used. That is a prompt problem, not a brand problem.
Which sources it leaned on. The domains cited in the answer are where the assistant went to find out. Being present on those is usually more valuable than another page on your own site, and it is the part of this work that does not depend on more runs.
Whether you were described accurately when you were named. Being named badly is a different problem from being absent, and only this screen shows it.
Reading a feed with gaps in it
Three things you will see that are not failures:
An answer with no ranked list. Many AI answers are prose. There is no position to report, so the position is blank rather than last.
A check with no verdict. The answer came back in a shape the analyser could not read, so the check contributes to neither half of any rate and is counted separately. It is not a "not mentioned".
A failed assistant. RankX AI names which assistant failed, refunds the check, and does not surface the upstream error text. A gap for one assistant on one run is a gap, not a zero.
Using the feed to fix something
A workable loop, and it is the one the visibility Agent Skill automates:
- Filter to prompts where you are absent, on the assistant that matters most to your buyers.
- Read five answers, not one. The pattern across five is the finding; one answer is an anecdote.
- Write down the domains cited. Three or four will repeat. Those are your off-page targets, and they are the same list AI Answer Citations will show you in aggregate.
- Write down the brands named. Compare against your competitor set and correct it if it is wrong.
- Check whether the answer even matched the question. If it did not, the prompt needs rewriting more than your site does.
- Turn what you find into tasks, so the work survives the reading session. See Tasks.
What the feed cannot tell you
Whether an answer is typical. Assistants are non-deterministic: two runs of the same prompt frequently return different brand lists. A single answer is one sample, and reading three consecutive days of one prompt as a trend is the most common analytical error available here.
Why the assistant chose those sources. RankX AI records what was cited, not the retrieval decision behind it. Nobody outside the vendor can see that.
What the assistant knows from training rather than from the web. A brand can be named with no citation at all, and that is a real and different signal, but it is not separable from retrieval by any tool.
Retention
How far back the feed goes is a per-plan retention limit, and it is one of the more material differences between plans on a product whose value is measured over time. The figures are in plans and limits.
Where to go next
- AI Answer Citations, the same sources in aggregate.
- AI Prompts, to act on what you read.
- Tracked platforms, for what each assistant is good at.
AI Prompts
The panel of questions RankX AI asks the assistants. Writing good ones, the three statuses, the per-plan cap, and why the prompt list is your credit bill.
AI Answer Citations
The domains and pages AI assistants cited when answering your tracked prompts. What the list is good for, and why it is not AI Overview Citations.