Skip to main content

​
What does the Pages table show?

Pages lists the URLs on your own domain that models attached as citations when they answered your tracked Prompts. One row is one URL. A URL appears only after a model actually cited it in a successful run inside the window you are looking at — this is not a crawl of your sitemap. A Citation is a single URL a model attached to an Answer. A Mention is your brand named in the answer text with no link. Pages counts citations only, so a page can be doing useful work in a market where your brand is named constantly and still show few rows here.

​
How do you read each column?

Each row carries the URL, how many citations it collected in the window, its share of your domain’s citations, and its average position in the citation list. Read them together: a page with few citations but a strong average position behaves differently from one cited often and always last.
Average position is a per-source-domain measure. It tells you where your URLs tend to sit in a list of links, not how you place against any competitor. There is no brand rank or position metric in Attensira.

​
Why is a number blank instead of zero?

Every rate in Attensira is a {value, n} pair, and every delta is a {value, real} pair. A blank value means not measured — no run produced data for that cell. A 0 with n above zero is a real, measured zero: runs happened and the URL was never cited. Deltas are tested for significance with a two-proportion z-test at 95%. When a change falls below that noise floor you get {real: false, value: null}, which reads in the table as no change we can prove. It is not a claim that nothing moved, and it is never a zero.
Treat a blank and a zero as different facts when you report internally. Rewriting “not measured” as 0% invents a failure that the data does not support.

​
Why does a page you care about not appear?

The table is bounded by what was queried. If a Prompt that would surface the page is not in your workspace, no run touched it. If the Model that cites it is not tracked for your workspace, it was never asked. A new workspace tracks two model channels by default — ChatGPT and Perplexity. Sampling depth also bounds what you see. Each Prompt × Model × country runs at most once per workspace-local calendar day, with n=1 on Starter and n=3 on Growth and Business. Rare citations need more samples and more days before they show up at all. Failed runs are stored but excluded from n, so a bad day narrows the window rather than poisoning it. Retention limits how far back you can look: 3 months on Starter, 12 on Growth, unlimited on Business.

​
What to do with the table

1

Sort by citations and read the top ten

These are the pages models currently trust to answer your Prompts. Keep them accurate and current before writing anything new.
2

Compare against your intended landing pages

If models cite a support article where you expect a product page, the product page is not answering the question the Prompt asks.
3

Check average position on the pages that matter

A URL consistently late in the citation list is being used as corroboration rather than as the answer.
4

Widen the window before concluding anything

At n=1 per day, single-day movement is mostly sampling noise. Read weeks, not days.

AI traffic

Crawler fetches of your pages, which are a different system from model answers.

Chats

Ask the agent to explain a page’s citation pattern in context.