Weekly AI visibility scans. See what really changed.
Your questions are asked again every week, on the day you choose. Turn on swing sampling and each change is marked as bigger than the usual swing, or too small to tell.
Visibility
7.5%3 of 40 non-branded
Citation rate
21.0%42 of 200 answers
Answers analysed
200
Named most instead
Northline
Visibility over time
Non-branded answers that named you, scan by scan. Each scan is divided by its own answers.
Visibility by assistant
Each assistant's share of answers that named you.
Only questions that do not use your name are counted.
Share of voice
Each brand's share of the names in non-branded answers.
Where you stand
Questions with your name and without it, kept apart.
Illustrative.
Why one scan is not enough
An assistant writes each answer fresh. Ask the same question twice and the answer can differ, including whether it names you. One scan is a single reading, and two readings a week apart can differ for no reason at all.
Researchers measured this for citations in a 2026 preprint. When they asked the same thing again with nothing changed, about one answer in seven flipped on whether a given source was cited. That is a preprint, not yet reviewed, and it measured citations rather than names, but it shows the size of the problem.
A change you cannot tell apart from chance is not a result. It is a coin landing differently.
How the weekly scan works
Once there are two scans, the Overview draws a line through them. It shows how often you were named and how often a page of yours was cited, scan by scan, beside the same figures for your competitors. The History screen lists every scan, newest first.
Weekly scans come with the plan, at $149 a month per website. The free first scan runs once and has no schedule.
You pick a day of the week and a time. The next scan is shown in your own time zone.
On that day, every question goes to every assistant again, exactly as before, and every answer is kept.
You can also start one extra scan by hand each week, for example straight after you publish a page.
When a scan finishes, or stops early, the owners and admins on your team get one email saying which.
How a change is called
Swing sampling measures how much your answers move on their own. During a scan, 20 questions that do not name you are asked three more times on each assistant, with nothing changed. Agentcites counts how often the answer flips between naming you and not.
It does not trust a small sample to show a small swing. Seeing no flips in 20 questions does not mean the swing is zero, so it uses the most the swing could be and still fit what it saw. With no flips in 20 questions, that is 16 percentage points.
| The screen says | When |
|---|---|
| Bigger than normal variation | The share of answers naming you moved by more than the swing. |
| Too small to tell | It moved, but by no more than the swing. It could be chance. |
| Nothing | No swing was measured, or fewer than 10 questions were sampled. |
Changes are compared as shares, never as counts. Going from 4 of 8 answers to 2 of 2 is a rise, not a fall of two, and comparing counts would get that backwards.
Counted as shares, 4 of 8 to 2 of 2 is a rise.
The share of answers naming you in two made-up scans. The count fell from 4 to 2, but the share went from half to all.
Why swing sampling is off until you turn it on
It costs more than the scan it checks. Twenty questions asked three more times on four assistants is 240 extra calls. For comparison, a scan of 40 questions on four assistants is 160 calls. Those numbers are an example. Because a set holds at most 50 questions, the check always makes more calls than the scan it checks. That is why it is a setting, and Settings shows the number of extra calls before you turn it on.
What it does not do
It measures the swing within one scan, seconds apart. The real swing between two scans a week apart is likely larger, so treat a cleared change as a good sign, not as proof.
It does not retry a failed scan on its own.
It does not change the questions between scans. The same questions are asked every week until you change them.
It does not say why a change happened. A page you published, a rival’s new page and a model update all look the same from here.
Answers from the assistants your buyers use
Each question goes to each assistant you choose, with web search on, the way a buyer would ask it.
- ChatGPT
- Perplexity
- Gemini
- DeepSeek
Questions
How often should I check my AI visibility?
Weekly is a sensible pace. Answers change from one asking to the next, so a single scan is a reading, and a trend needs at least two scans that agree.
Why did my AI visibility drop this week?
It may not have. An assistant can answer the same question differently with nothing changed. Swing sampling tells you whether a move is bigger than that, and says "Too small to tell" when it is not.
Can I run a scan right after I publish a page?
Yes. You can start one extra scan by hand each week, on top of the weekly one.
Sources
Read more about getting named in AI answers
See it for your own site
We ask 20 of your questions to four AI assistants, and you see every answer before you pay anything. You do not need a card.